October 30, 2017

IJCNN Sessions

IJCNN-1 Non-iterative Approaches in Learning
IJCNN-2 Special Session on Machine Learning and Deep Learning Methods applied to Vision and Robotics (MLDLMVR)
IJCNN-3 Complex-Valued and Quaternionic Neural Networks
IJCNN-4 Special Session on Deep Neural Audio Processing
IJCNN-5 Data Driven Approach for Bio-medical and Healthcare
IJCNN-6 Feature Extraction and Learning on Image and Text Data
IJCNN-7 Advances in Reservoir Computing
IJCNN-8 Empowering Deep Learning Models
IJCNN-09 Spiking Neural Networks (SNN)
IJCNN-10 Deep Learning for Structured and Multimedia Information
IJCNN-11 Cognition and Development
IJCNN-12 Biologically Inspired Computational Vision
IJCNN-13 Advanced Machine Learning Methods for Large-scale Complex Data Environment
IJCNN-14 Parallelism in Machine Learning: Theory and Applications
IJCNN-15 Intelligent Power Systems
IJCNN-16: Hybrid Neural Intelligent Models and Applications
IJCNN-17 Concept drift, domain adaptation & learning in dynamic environments
IJCNN-18 Learning from Big Graph Data: Theory and Applications
IJCNN-19 Advanced Cognitive Architectures for Machine Learning
IJCNN-20 Neurocomputation and Cognition
IJCNN-21 Deep Reinforcement Learning
IJCNN-22 Ordinal and Monotonic Classification
IJCNN-23 Deep Learning and Reinforcement Learning for Games
IJCNN-24 Neural Techniques for Artificial and Natural Locomotions
IJCNN-25 Multi-agent Reinforcement Learning and Adaptive Dynamic Programming Designs
IJCNN-26 Data Mining and Knowledge Discovery in Cyber-Physical Systems
IJCNN-27 Extreme Learning Machines
IJCNN-28 Adversarial machine learning in information security
IJCNN-29 Machine Learning for Massive and Complex Urban Data Analytics
IJCNN-30 Machine learning for big data: scalable algorithms and applications
IJCNN-31 Neural Models for Behavior Recognition
IJCNN-32 Biologically-inspired Neural Networks for Robotics and Mechanics
IJCNN-33 Neural Intelligence After Tomorrow
IJCNN-34 Ensemble models for Pattern Recognition and Data Mining
IJCNN-35 Evolutionary Computation for Neural Networks
IJCNN-36 Deep, Transfer and Reinforcement Learning for Robotics and Intelligent Agents
IJCNN-37 Advances in Document Analysis and Recognition
IJCNN-38 Neural Approaches for Natural Language
IJCNN-39 Industrial Applications
IJCNN-40: Advanced Machine Learning Techniques for Computational Biology
IJCNN-41 Machine Learning for Encoding‐Decoding the Brain neural activity

IJCNN-1 Non-iterative Approaches in Learning

Organized by P. N. Suganthan (epnsugan@ntu.edu.sg ) and Filippo Maria Bianchi

Optimization, which plays a central role in learning, has received considerable attention from academics, researchers, and domain workers. Many optimization problems in machine learning can be tackled with non-iterative approaches, which can be presented in closed-form manner. Those methods are in general computationally faster than iterative solutions and less sensitive to parameter settings. Even though non-iterative methods have attracted much attention in recent years, there exists a performance gap when compared with older methods and other competing paradigms. This special session aims to bridge this gap.

Scope and Topics

The first target of this special session is to present the recent advances of non-iterative solutions in learning. Secondly, the focus is on promoting the concepts of non-iterative optimization with respect to counterparts, such as gradient-based methods and derivative-free iterative optimization techniques. Besides the dissemination of the latest research results on non-iterative algorithms, it is also expected that this special session will cover some practical applications, present some new ideas and identify directions for future studies.

Original contributions, comparative studies with both iterative and non-iterative methods are welcome. Typical paradigms include (but not limited to) random vector functional link (RVFL), Echo State Networks (ESN), kernel ridge regression (KRR), random forests (RF), etc…

The topics of the special session include, but are not limited to:

  • Methods with and without randomization
  • Regression, classification and time series analysis
  • Kernel methods such as kernel ridge regression, kernel adaptive filters, etc.
  • Feedforward, recurrent, multilayer, deep and other structures.
  • Ensemble learning
  • Moore-Penrose pseudo inverse, SVD and other solution procedures.
  • Gaussian Process regression
  • Non-iterative methods for large-scale problems with and without kernels
  • Theoretical analysis of non-iterative methods
  • Comparative studies with competing for iterative methods


IJCNN-2 Special Session on Machine Learning and Deep Learning Methods applied to Vision and Robotics (MLDLMVR)

Organized by José García-Rodríguez (jgarcia@dtic.ua.es), Alexandra Psarrou, Isabelle Guyon, Andrew Lewis

Over the last decades there has been an increasing interest in using machine learning and in the last few years, deep learning methods, combined with other vision techniques to create autonomous systems that solve vision problems in different fields. This special session is designed to serve researchers and developers to publish original, innovative and state-of-the-art algorithms and architectures for real-time applications in the areas of computer vision, image processing, biometrics, virtual and augmented reality, neural networks, intelligent interfaces and biomimetic object-vision recognition.

Scope and Topics

This special session provides a platform for academics, developers, and industry-related researchers belonging to the vast communities of *Neural Networks*, *Computational Intelligence*, *Machine Learning*, *Deep Learning*, *Biometrics*, *Vision systems*, and *Robotics *, to discuss, share experience and explore traditional and new areas of the computer vision, machine and deep learning combined to solve a range of problems. The objective of the workshop is to integrate the growing international community of researchers working on the application of Machine Learning and Deep Learning Methods in Vision and Robotics to a fruitful discussion on the evolution and the benefits of this technology to the society.

The methods and tools applied to vision and robotics include, but are not limited to, the following:

  • Computational Intelligence methods
  • Machine Learning methods
  • Self-adaptation and self-organization
  • Robust computer vision algorithms (operation under variable conditions, object tracking, behavior analysis and learning, scene segmentation)
  • Extraction of Biometric Features (fingerprint, iris, face, voice, palm, gait)
  • Convolutional Neural Networks CNN
  • Recurrent Neural Networks RNN
  • Deep Reinforcement Learning DRL
  • Hardware implementation and algorithms acceleration (GPUs, FPGAs)

The fields of application can be identified, but are not limited to, the following:

  • Video and Image Processing
  • Video tracking
  • 3D Scene reconstruction
  • 3D Tracking in Virtual Reality Environments
  • 3D Volume visualization
  • Intelligent Interfaces (User-friendly Human-Machine Interface)
  • Multi-camera and RGB-D camera systems
  • Multi-modal Human Pose Recovery and Behavior Analysis
  • Gesture and posture analysis and recognition
  • Biometric Identification and Recognition
  • Extraction of Biometric Features (fingerprint, iris, face, voice, palm, gait)
  • Surveillance systems
  • Autonomous and Social Robots
  • Robotic vision
  • Industry 4.0
  • IoT and Cyber-physical Systems


IJCNN-3 Complex-Valued and Quaternionic Neural Networks

Organized by Marcos Eduardo Valle (valle@ime.unicamp.br), Igor Aizenberg, Akira Hirose, and Danilo Mandic

Complex-valued neural networks (CVNNs) and quaternionic neural networks (QNNs) constitute a growing research area that has attracted continued interest for the last decade. One of the most important characteristics of CVNNs is the proper treatment of phase and the information contained in phase, e.g., the treatment of wave- and rotation-related phenomena such as electromagnetism, light waves, quantum waves, and oscillatory phenomena. QNNs, which have potential applications in three- and four-dimensional data modeling, have been effectively used for processing and analysis of multivariate images such as color and polarimetric SAR images.

More generally, hypercomplex-valued neural networks – which include CVNNs and QNNs, treat multidimensional data as a single entity. There are several new directions in hypercomplex-valued neural networks: from the formal generalization of the commonly used algorithms to the hypercomplex case that is mathematically richer than regular neurons, to the use of original activation functions that can increase significantly the neuron and network functionality. There are also many interesting applications of HVNNs in pattern recognition and classification, nonlinear filtering, intelligent image processing, brain-computer interfaces, time-series prediction, bioinformatics, robotics, etc.

The CVNNs special session, which indeed covers hypercomplex-valued models, has become a traditional event of the IJCNN conference. Eight special sessions organized since 2006 (WCCI-IJCNN 2006, Vancouver; WCCI-IJCNN 2008, Hong Kong; IJCNN 2009, Atlanta; WCCI-IJCNN 2010, Barcelona; IJCNN-2011, San Jose; WCCI-IJCNN 2012, Brisbane; IJCNN-2013, Dallas; IJCNN-2014, Beijing, WCCI-IJCNN 2016, Vancouver) attracted numerous submissions and had large audiences. They featured many interesting presentations and very productive discussions.

Scope and Topics

IJCNN 2018, which is an integrated part of IEEE WCCI 2018, will be a very attractive forum for complex-valued and quaternionic neural networks. It will be possible to organize a systematic and comprehensive exchange of ideas in the area, to present the recent research results, and to discuss future trends. We hope the proposed session will attract not only potential speakers but also many kinds of research interested in joining the hypercomplex-valued neural network community. We expect that this session would be very beneficial for computational intelligence researchers and other specialties that are in need of the sophisticated neural networks tools.

Papers that are, or might be, related to all aspects of the hypercomplex-valued neural networks, including CVNNs and QNNs, are invited. We welcome contributions on theoretical advances as well as contributions of applied nature. We also welcome interdisciplinary contributions from other areas that are on the borders of the proposed scope. Topics include, but are not limited to:

  • Theoretical Aspects of CVNNs and QNNs
  • Complex-Valued and Quaternion Activation Functions
  • Learning Algorithms for CVNNs
  • Complex-Valued and Quaternionic Associative Memories
  • Pattern Recognition, Classification and Time Series Prediction using CVCNNs and QNNs
  • CVNNs and QNNs in Nonlinear Filtering
  • Dynamics of Complex-Valued and Quaternionic Neurons
  • Learning Algorithms for CVCNNs and QNNs
  • Chaos in Complex Domain
  • Feedforward CVCNNs
  • Spatiotemporal CVNNs Processing
  • Frequency Domain CVNNs Processing
  • Phase-Sensitive Signal Processing
  • Applications of CVNNs and QNNs in Image Processing, Speech Processing, and Bioinformatics
  • Quantum Computation and Quantum Neural Networks
  • CVNN in Brain-Computer Interfaces
  • CVNNs and QNNs in Robotics
  • Clifford and Hypercomplex-valued Neural Networks

IJCNN-4 Special Session on Deep Neural Audio Processing

Organized by Emanuele Principi, Aurelio Uncini, Björn Schuller, Stefano Squartini (s.squartini@univpm.it)

Computational Audio Processing techniques have been largely addressed by scientists and technicians in many application areas, like entertainment, human-machine interfaces, security, forensics, and health. Depending on the problem under study, these techniques have been successfully applied to speech signal processing (speech/speaker recognition, speech enhancement, emotion and speaker state recognition, privacy-preserving speech processing), music information retrieval and automated music generation, and in generic sound processing for acoustic monitoring, acoustic scene understanding and sound separation, detection, and identification. In the case of animal vocalization analysis, some efforts have been recently spent in the automatic classification and recognition of animal species by means of their emitted sounds.

In the different fields, state of the art performance has recently been obtained by using data-driven learning systems, often represented by variants of deep neural network architectures. Several challenges remain open, due to the increasing complexity of the tasks, the presence of non-stationary operating conditions, the difference between laboratory and real acoustic scenarios, and the necessity to respect hard real-time constraints, also when the amount of data to process is big and battery-powered devices are involved. In some other application contexts, the challenge is facing a scarce amount of data to be used for training, and suitable architectures and algorithms need to be designed on purpose. Moreover, also the employment of cross-domain approaches to exploit the information contained in diverse kinds of environmental audio signals are often needed, as recently investigated by some pioneering works.

Scope and Topics

The aim of this special session is to provide a forum for the presentation of the most recent advancements on Deep Neural Networks algorithms applied to Digital Audio problems, with particular attention to speech analysis and enhancement, music information retrieval and generation, as well as acoustic scenes and events detection and classification, exploring also new and emerging methods, such as end-to-end and one-shot/zero-shot learning. Topics include, but are not limited to:

  • Computational Audio Analysis
  • Deep Learning Algorithms in Digital Audio
  • Neural Architectures for Audio Processing
  • Transfer, Weakly Supervised, and Reinforcement Learning for Audio
  • Music Information Retrieval
  • Music Performance Analysis
  • Neural Methods for Music/Speech generation and synthesis
  • Computational Methods for Physical Instrument Modeling
  • Music Content Analysis
  • Voice conversion
  • Speech and Speaker Analysis and Classification
  • Sound Detection and Identification
  • Acoustic Novelty Detection
  • Computational methods for Wireless Acoustic Sensor Networks
  • Acoustic Scene Analysis
  • Cross-domain Audio Analysis
  • Signal enhancement with neural networks
  • End-to-End learning for Digital Audio Applications
  • Privacy preserving computational speech processing
  • One-shot/Zero-shot learning for digital audio applications


IJCNN-5 Data Driven Approach for Bio-medical and Healthcare

Organized by Paul J Kennedy (Paul.Kennedy@uts.edu.au), Mukesh Prasad and Alexei Manso Correa Machado

Healthcare and biomedical sciences have become data-intensive fields, with a strong need for sophisticated data mining methods to extract the knowledge from the available information. For example, data analysis methods are applied on biomedical datasets, namely DNA microarray data or Next Gen sequencing data to predict treatment outcomes of paediatric Acute Lymphoblastic Leukaemia patients. Moreover, clustering methods are routinely used to investigate the interpretation of the correlated genes associated with cellular and biological function.

Scope and Topics

Biomedical data contains several challenges in data analysis, including high dimensionality, class imbalance and low numbers of samples. Although the current research in this field has shown promising results, several research issues need to be explored as follows. There is a need to explore feature selection methods to select stable sets of genes to improve predictive performance along with interpretation. There is also a need to explore big data in biomedical and healthcare research. An increasing flood of data characterises human health care and biomedical research. Healthcare data are available in different formats, including numeric, textual reports, signals and images, and the data are available from different sources. An interesting aspect is to integrate different data sources in the data analysis process which requires exploiting the existing domain knowledge from available sources. The data sources can be ontologies, annotation repositories, and domain experts’ reports.

This special session aims to bring together the current research progress (from both academia and industry) on data analysis for biomedical and healthcare applications. It will attract healthcare practitioners who have access to interesting sources of data but lack the expertise in using the data mining effectively. Special attention will be devoted to handle feature selection, class imbalance, and data fusion in biomedical and healthcare applications.

The main topics of this special session include, but are not limited to, the following:

  • Information fusion and knowledge transfer in biomedical and healthcare applications
  • Data Analysis of the biomedical data including genomics.
  • Text mining for medical reports.
  • Statistical analysis and characterization of biomedical data.
  • Machine Learning Methods Applied to Medicine
  • Large Datasets and Big Data Analytics on biomedical and healthcare applications.
  • Information Retrieval of Medical Images
  • Machine learning technique for single cell sequencing analysis
  • Medical imaging and genomics


IJCNN-6 Feature Extraction and Learning on Image and Text Data

Organized by Domingo Mery (domingo.mery@uc.cl), Jefersson Alex dos Santos, Nabin Sharma and Mukesh Prasad

The current resurgence of neural networks in the form of deep learning have shown remarkable results in fundamental tasks such as segmentation, tracking, detection, recognition, classification, clustering and feature learning. The deep features extracted from deep neural network architectures are robust and have good representation for most of the fundamental computer vision tasks. Although, deep learning has shown tremendous amount of success in the fundamental tasks in the areas of image and text analysis, the intuitive understanding of the architectures are yet to be explored in details. There is a need of further exploration of architectures, whichis suitable for a specific tasks. Training of neural network architecture and then transfer the learned features to another unknown task, which requires transfer learning and fine tuning. Therefore, transfer learning also has a significant research scope, both from theoretical and application perspective. Traditional machine learning approaches rely on hand crafted features such edges, texture, SIFT etc. The fusion of such features are used to tackle many complex computer vision problem and they poorly generalized to the unknown scene.

Scope and Topics

Recent advances in both feature extraction and learning on image and text data have allowed us to develop promising solutions that are being increasingly used in our society. For this reason, this special session aims to bring together researchers, scientists, engineers and students to discuss the state of the art and the new trends in feature extraction and learning on image and text data. The idea of the session is to present recent theories and applications in deep learning, transfer learning, reinforcement learning and some other feature extraction/learning techniques for various image and text oriented tasks, such as object recognition, image retrieval/classification, annotation, multimedia processing, image super-resolution, text mining and text analysis. Special attention will be devoted to handle advanced issues of network architecture design, real-time performance criteria for various applications and diverse application areas. The main topics of this special session include, but are not limited to, the following:

  • Cross domain transfer learning
  • Real-time Object segmentation, detection and recognition in complex environment
  • Human gesture/activity recognition
  • Visual analysis of crowds, surveillance systems and applications
  • Document image analysis and systems
  • Handwriting recognition
  • Writer identification
  • Text classification/analysis
  • Sparse representation and low-rank representation for feature extraction
  • Contextual scene understanding and summarization
  • Ensemble of traditional and deep learning techniques
  • Methods aplicable to Forensic Science
  • Drone based applications.
  • Novel Reinforcement Learning algorithms with deep representation layer


IJCNN-7 Advances in Reservoir Computing

Organized by Claudio Gallicchio (gallicch@di.unipi.it), Alessio Micheli, Simone Scardapane and Peter Tiňo

Website: https://sites.google.com/view/reservoir-computing-ijcnn-2018

During the last decade, Reservoir Computing (RC) has attested itself as a state of the art paradigm for efficient learning in the temporal domain. The extreme efficiency of the approach follows from limiting the training algorithm to only a readout component, while the temporal encoding process is carried out by a dynamical pool of recurrent neurons that, under certain conditions, is able to develop a rich representation of the temporal information even if untrained.
The common paradigm has been instantiated into several models, among which the Echo State Network and the Liquid State Machine represent the most widely known ones.
Since its origins, we have witnessed a progressively increasing popularization of the RC approach, especially by virtue of its ease of implementation and its extreme training efficiency. At the same time, several theoretical research lines strongly contributed in understanding the bias of untrained recurrent models, grounding the RC methodology on a solid mathematical basis and better delineating the potentialities and downsides of the approach.

Scope and Topics

After 10 years since the introduction of the term “Reservoir Computing” into the neural networks literature, it is now time to sum up all the recent research initiatives in the field and start drawing its future. This session calls for both theoretical and applicative contributions emerging in the field of RC, aiming at stimulating an open discussion in the neural network community.

The list of relevant topics for this session includes, but is not limited to, the following:

  • Novel Reservoir Computing models
  • Theoretical analysis of Reservoir Computing
  • Physical implementations of Reservoir Computing
  • Evolutions of the RC paradigm (e.g., conceptors)
  • Biological motivations and applications in Neuroscience of Reservoir Computing
  • Adaptation of reservoir dynamics and of system dynamics
  • Deep Reservoir Computing models
  • Reservoir Computing for learning in structured domains (trees, graphs, networks, …)
  • Reservoir Computing for Big Data
  • Novel application fields for the RC paradigm


IJCNN-8 Empowering Deep Learning Models

Organized by Nicolò Navarin (nnavarin@math.unipd.it), Luca Oneto, Luca Pasa and Alessandro Sperduti

In recent years, Deep Learning has become the go-to solution for a broad range of applications, often outperforming state-of-the-art. Deep learning allows computational models that are composed of multiple processing layers to learn representations of data with multiple levels of abstraction. These methods have dramatically improved the state-of-the-art in speech recognition, computer vision, drug discovery, genomics and many others.

Scope and Topics

The goal of this special session is to provide a forum for focused discussions on new extensions of deep learning models and techniques and to gain a deeper understanding of the difficulties and limitations associated with state-of-the-art approaches and algorithms. Practitioners should provide practical insights to the theoreticians, which in turn, should supply theoretical insights and guarantees, further strengthening and sharpening practical intuitions and wisdom.

Examples of these possible extensions are:

  • Multimodal and Multitask Deep Learning
  • Deep Transfer Learning
  • Deep Recurrent and Recursive Neural Networks
  • Deep Learning on Structured Data
  • Interpretability of Deep Learning
  • Private and Federated Deep Learning
  • Generative and Adversarial Deep Learning
  • Randomized Deep Learning (Deep ELM, Deep ESN, Deep Reservoir Computing)

The focus of this special session is to attract both solid contributions or preliminary results which show the potentiality and the limitations of new ideas, refinements, or contaminations between the different fields of deep learning and other fields of research in solving real world problems. Both theoretical and practical results (e.g. Social Data Analysis, Speech, Natural Language Processing, Cybersecurity) are welcome to our special session. This special session is supported by the IEEE Task Force on Deep Learning http://deeplearning.math.unipd.it .

IJCNN-09 Spiking Neural Networks (SNN)

Organized by Professor Nikola Kasabov (nkasabov@aut.ac.nz), Maryam G Doborjeh (mgholami@aut.ac.nz), Dr Elisa Capecci, Zohre Gholami

Spiking Neural Networks (SNN) are a rapidly emerging means of neural information processing, drawing inspiration from the brain processes. They have the potential to advance technologies and techniques in fields as diverse as medicine, finance, computing, and indeed any field that involves complex temporal or spatiotemporal data. SNN, as the third generation of neural networks, can operate on noisy data, in changing environments at low power and with high effectiveness. Due to their basis in biological neural networks, SNN research is strongly positioned to benefit from advances made in the fields of molecular, evolutionary and cognitive neuroscience. There is presently considerable interest in this topic. We believe that this area is quickly establishing itself as an effective alternative to traditional machine learning technologies, and the interest in this area of research is growing rapidly.

Scope and Topics

The aim of this special session is to bring together research works of contemporary areas of SNN, including theoretical, computational, application-oriented, experimental studies, and emerging technologies such as neuromorphic hardware. This special session invites researchers to present state-of-the-art approaches, recent advances and the potential of SNN.

The topics relevant to this special session include, but are not limited to, the following:

  • Theory of SNN
  • Learning algorithms for SNN, including Deep Learning
  • Computation with and within SNN
  • Theory or practice in biologically-plausible neural simulation or biomimetic models
  • Big data and stream data processing in SNN
  • Multiple sensor networks data processing in SNN
  • Neuromorphic hardware systems and applications
  • Optimization of SNN
  • SNN models of cognitive development
  • Information encoding for SNN
  • SNN applications in neuroinformatics, bioinformatics, medicine and ecology.
  • SNN in BCI
  • SNN in neuro-robotic
  • Any other topics relating to SNN, their theory, or applications.


IJCNN-10 Deep Learning for Structured and Multimedia Information

Organized by Davide Bacciu (bacciu@di.unipi.it), Silvio Jamil F. Guimarães and Zenilton K. G. Patrocínio Jr


A key factor triggering the deep learning revolution has been its ground-breaking performance on image and video processing applications. These have been built mostly on a (multi-dimensional) raw data representation of the visual information.  Multimedia content, on the other hand, calls for more articulated data representations catering for the multimodal nature of this information. These are often based on a structured representation that can capture the complexity of the contextual, semantic and geometrical relationships among the visual, phonetic and textual entities and concepts.

Scope and Topics

The goal of this special session is to provide a forum for researchers working on the next generation of deep learning models for machine vision and multimedia information, which are capable of extracting and processing information in a structured representation and/or with a multimodal nature. We welcome contributions proposing innovative deep models dealing with:

  • learning hierarchical or networked representations of multimedia information;
  • processing of structured multimedia information under the form of sequences, labelled trees, as well as more general forms of graphs;
  • understanding and synthesizing of multimodal data;
  • fusion of multimodal information.

This special session is meant to attract researchers from deep learning, machine vision and multimedia information communities. We aim to bring together researchers with consolidated tradition on structured data processing (such as in machine learning and NLP) with those with machine vision and multimedia processing insight, but mostly working with flat-data representations.

Topics of interest for this special session include, but are not limited to, the following:

  • deep learning models for structured data;
  • representation learning in machine vision and multimedia processing;
  • hierarchical/structured visual processing;
  • deep models for visual data streams;
  • generative and variational deep learning for multimedia data;
  • multimedia data synthesis;
  • attentional and bio-inspired models for the processing of visual and audio information;
  • applied deep learning to machine vision and multimedia processing, such as: biomedical images and biobanks, pose and gesture estimation from graphs, etc.;
  • innovative software and libraries for deep learning and multimedia content understanding.


IJCNN-11 Cognition and Development

Organized by Nikolas J. Hemion (nhemion@softbankrobotics.com), Anna-Lisa Vollmer, Angelo Cangelosi, Pierre-Yves Oudeyer

The special session aims at the presentation of the latest results in the investigation of machine learning and cognitive robotics models that are taking inspiration from our understanding of sensorimotor, cognitive, and social development in humans and other animals.

Scope and Topics

The aim is twofold: On the one hand, the special session provides a platform for presenting insights about the functions and processes underlying developmental learning. On the other hand, it encourages discussion of the role of developmental aspects in learning, in particular in the light of the ongoing success of deep learning methods.

Relevant research questions include:

  • How can deep learning models be improved by including developmental processes, such as intrinsic motivation, sensorimotor exploration, or social learning and interaction?
  • Is current deep learning methodology suited for autonomous and open-ended learning in cognitive robots? Or will it be necessary to consider developmental processes, and if so, which ones?
  • What are the mechanisms that allow a child or robot to autonomously develop cognitive capabilities?
  • How does the social and physical environment, with which the child interacts, shape and scaffold the child’s developing cognitive skills and knowledge? And how can machine learning systems (including cognitive robots) exploit these beneficial factors?
  • What is the relative contribution of Nature and Nurture in the development of human and machine intelligence?
  • What do qualitative stages during development, and body and brain maturational changes, tell us about the mechanisms and principles supporting development?

The special session also encourages submissions of relevant research on cognition and development from other disciplines, such as child psychology, developmental linguistics, neuroscience, philosophy, and interdisciplinary approaches to cognition and development.

List of main topics:

  • developmental robotics
  • epigenetic robotics
  • developmental deep learning
  • neuro-robotics
  • bio-inspired and cognitive robotics
  • cognitive modelling
  • intrinsic motivation
  • sensorimotor development
  • cognitive development
  • social development
  • language acquisition


IJCNN-12 Biologically Inspired Computational Vision

Organized by: Khan M. Iftekharuddin (kiftekha@odu.edu)


Constructive understanding of computational principles of visual information processing, perception and cognition is one of the most fundamental challenges of contemporary science. Deeper insight into biological vision helps to advance intelligent systems research to achieve robust performance similar to biological systems. Biological inspiration indicates that sensory processing, perception, and action are intimately linked at various levels in animal vision. Implementing such integrated principles in artificial systems may help us achieve better, faster and more efficient intelligent systems.

Scope and Topics

This special session provides an integrated platform to present original ideas, theory, design, and applications of computational vision. Topics of interest include, but are not limited to the following:

  • Theoretical approaches and modeling in computational vision
  • Neuronal mechanisms of visual processing
  • Low level vision and its relationship to biological machinery
  • Artificial learning systems for image and information processing and evidential reasoning for recognition
  • Intelligent search in communications networks
  • Modeling issues in ATM networks, agent-oriented computing architectures
  • Perception of shape, shadows, poses, color and illumination in object recognition
  • Tracking for inferring shapes and 3D motions
  • Active visual perception, attention and robot vision
  • Functional Magnetic Resonance Imaging (fMRI) studies of visual segmentation and perception
  • Application of computational vision in areas of
    • Automated target identification and acquisition systems in defense and industry
    • Biomedical imaging
    • 3D photography
    • Face recognition
    • Learning to segment camouflaged objects
    • Motor actions and robotics
    • Image databases and indexing
  • Hardware implementation of computational vision
  • Contemporary deep learning algorithms, structures and methods for computer vision applications
    • Transfer learning
    • Meta learning
    • Recurrent learning
  • Any other topics related to biological approaches in computer vision


IJCNN-13 Advanced Machine Learning Methods for Large-scale Complex Data Environment

Organized by Jia Wu (jia.wu@mq.edu.au), Bo Du, Michael Sheng, Chengqi Zhang


Traditional machine learning methods have been commonly used for many applications, such as text classification, image recognition, and video tracking. For learning purposes, these data are often required to be represented as vectors. However, many other types of data objects in real-world applications, such as chemical compounds in bio pharmacy, brain regions in brain networks and users in social networks, contain rich feature vectors and structure information. Such a simple feature-vector representation inherently loses the structure information of the objects. In reality, objects may have complicated characteristics, depending on how the objects are assessed and characterized. Meanwhile the data may come from heterogeneous domains, such as traditional tabular-based data, sequential patterns, social networks, time series information, and semi-structured data. Novel machine learning methods are desired to discover meaningful knowledge in advanced applications from objects with complicated characteristics.

Scope and Topics

This special session expects to solicit contributions on the advanced machine learning methods and applications from complicated data environment. The topics of interest include, but are not limited to:

  • Supervised/Unsupervised/Semi-supervised Learning
  • Semi-structured Learning
  • Graph-based Learning
  • Graph Classification/Clustering/ Streaming
  • Multi-Graph Learning
  • Deep Graph Learning
  • Online Graph Learning
  • Time Series Learning
  • Complex Social Networks
  • Multi-view/instance/ label Learning
  • Heterogeneous Transfer Learning
  • Web/Text/Image Mining
  • Multimedia Learning
  • Big Data Analytics for Social Media
  • Big Data and the Internet of Things


IJCNN-14 Parallelism in Machine Learning: Theory and Applications

Organized by Veronica Bolon-Canedo (vbolon@udc.es), Jorge Gonzalez-Dominguez, Amparo Alonso-Betanzos, Beatriz Remeseiro

Machine learning (ML) is a prolific research area focused on the study and definition of algorithms able to learn from and make predictions on data. The current technologies and the use of Internet have revolutionized the way in which people acquire, store, or share data, resulting in huge amounts of information. In order to deal with massive volumes of data, ML techniques have to learn complex models with millions of parameters, increasing the computational cost involved.

Parallel programming and distributed learning are gaining attention in the last years as a mean to alleviate the effects of this extremely increase of computational cost. The efficient exploitation of High Performance Computing (HPC) resources such as multicore CPUs, hardware accelerators (GPUs, Intel Xeon Phi coprocessors, FPGAs, etc.), clusters or cloud-based systems can significantly accelerate many ML algorithms. This increase of speed allows ML users to either reduce the time needed for their applications or to search a larger space in the same period of time.

Scope and Topics

We invite papers on both practical and theoretical issues about incorporating parallel and distributed approaches into machine learning problems. In particular, topics of interest include, but are not limited to:

  • Development of parallel machine learning algorithms on multicore and manycore architectures: multithreading, GPUs, Intel Xeon Phi coprocessor, FPGAs, etc.
  • Development of distributed machine learning algorithms.
  • Exploitation of cloud, grid and distributed-memory systems to accelerate machine learning algorithms: Spark, Hadoop, MPI, etc.
  • Scalability analysis of parallel and distributed methods for machine learning.
  • Performance comparison of parallel and distributed machine learning algorithms.
  • Deep learning models trained across multicore CPUs, GPUs or clusters of computers.
  • Applications: bioinformatics, medicine, multimedia, marketing, cyber security, etc.


IJCNN-15 Intelligent Power Systems

Organized by Donnot Benjamin (benjamin.donnot@gmail.com), Marot Antoine, Isabelle Guyon and Louis Wehenkel

Website: https://sites.google.com/view/intelligent-power-systems-wcci/accueil

Brief Description:

The electricity power grid is one of the largest and most complex artificial systems currently in operation, with high criticality level given its wide social and economic impact. Such complexity is expected to increase in years to come to face possible climate changes while integrating intermittent renewable energy resources and distributed decisions taken at the consumer level. To limit building new power lines and expanding the power grid with costly equipment investments, new approaches relying on digital technologies are needed to leverage existing infrastructures by optimizing the management and operations of power grids: this is the “smart grid” era. In this context, new methods are needed to better understand and anticipate the behavior of the grid, with spatial and temporal multi-scale resolutions, both for predictive modeling and decision making. To complement historical physical models of the grid, coupled with standard optimization that are currently in use in study-tools of grid operators, learning-based methods promise to take into account expert knowledge to better assist them in analyzing demand forecasts and managing crisis situations.

Scope and Topics

Neural networks and machine learning have made great advances in the direction of continuously adapting to changes, acquiring new knowledge of systems, and improving context awareness. We anticipate that recent developments in neural networks will translate into applications in power systems, particularly to model ever larger datasets with spatio-temporal structure and capitalize on shared internet information. In addition, techniques that favor explainability and/or interpretability needed to assist operators in decision making are still in their infancy. Therefore, this special session aims at compiling the latest efforts and research advances in applying neural networks, machine learning, and other computational intelligence algorithms to augment existing study tools and simulators. Work that aims at accelerating research on a broader scale, enabling easier and more transparent iterations in the community, like open-sourcing dataset, releasing benchmark platforms or organizing data challenges will be very welcome.

Compared to previous years, the emphasis of this new call will be on:

  • Building a computational intelligence power system community for decision making (assisting the operators, not only displaying new information with the integration of new tools)
    • Towards explainability/interpretability, some context awareness
    • Coupling decisions at different time scales
  • Leveraging the use of existing physical simulators with learning
    • Meeting/crossing observational data, simulators and recent advances in AI
  • Creating open initiatives between TSOs and Research in Universities:
    • Shared data on real system with confidentiality preservation
    • New Benchmarks
    • Challenge organization

The scope of the special session comprises all aspects of machine learning applied to smart grid management and operations, including but not limited to the following topics:

  • Stochastic Load and/or Renewable Generation forecasting
    • At different spatial and time scales
  • Power grid clustering and segmentation with visualization:
    • Operators naturally focus manually on zone of interests to study and solve a problem, how can we do it more automatically?
  • Grid operation support for real-time:
    • Decision proposals to the operator under uncertainties.
  • Learning with simulators:
    • Simulators are of great help to acquire data for free and learn from it, how can we leverage their usage augmenting them with machine learning.
  • Calibrating simulators:
    • Learning the physical parameters of a given power grid with measurements
  • Causal learning:
    • What are the factors that determine some observations on the grid? How can we influence those variables then?
  • Multi-scale Reinforcement Learning:
    • From monthly to weekly to daily, how can we couple decisions taken at different time scale to better optimize grid management, especially between planned maintenance operations and near to real-time control room dispatching operations.
  • Predictive Asset Management:
    • Grid Development and renewal requires smart planning to keep the system continuously running in normal conditions. Asset renewal will be quite high in the coming years for aging grids. There is a need today to estimate materials wearing given their historical operational and environmental conditions to prioritize renewals, as well as model the impact of applying different renewal policies.
  • Collective Intelligence with open-source initiatives:
    • This applies to smart grid but to research as well: how can we ease collaboration and iterations with shared materials like data, benchmarks and challenges?


IJCNN-16: Hybrid Neural Intelligent Models and Applications

Organized by Patricia Melin (pmelin@tectijuana.mx), Alma Alanis (alma.alanis@cucei.udg.mx)

SC methodologies at the moment include (at least) Neural Networks, Fuzzy Logic, Genetic Algorithms and Chaos Theory. Each of these methodologies has advantages and disadvantages and many problems have been solved, by using one of these methodologies. However, many real-world complex industrial problems require the integration of several of these methodologies to really achieve the efficiency and accuracy needed in practice. This session will include papers dealing with methods for integrating the different SC methodologies and neural networks in solving real-world problems. The Special Session will consider applications on the following areas: Robotic Dynamic Systems, Non-linear Plants, Manufacturing Systems, Pattern Recognition, Medicine and Time Series Prediction. Hybrid models offer advantages when a prudent combination of methods is performed and in this case can be a powerful tool in solving complex problems. This Special Session is being organized as one of the main activities of the Task Force on Hybrid Intelligent Systems of the NNTC.

Scope and Topics

This special session aims to promote research on hybrid neural intelligent models all over the world and provides innovative approaches for solving real problems.

The topics include but are not limited to:

  • Hybrid Neural Systems,
  • Neuro-Fuzzy Systems,
  • Genetic Neural Systems,
  • Neuro Fuzzy Genetic Systems
  • Hybrid Intelligent models for application on Pattern Recognition, Time Series Prediction, Modeling, Control, Medicine, Robotics, among others.


IJCNN-17 Concept drift, domain adaptation & learning in dynamic environments

Organized by Giacomo Boracchi (giacomo.boracchi@polimi.it), Robi Polikar, Manuel Roveri, Gregory Ditzler

website: http://home.deib.polimi.it/boracchi/events/ijcnn2018_SS/index.html

An increasing number of systems and algorithms rely on predictive models that are learned directly from data, and the computational intelligence community has widely investigated the abilities of these modes in various real world environments. Despite the steady performance improvements shown over the past few years, there are still many challenges that need to be addressed, one of the most prominent is that most learning algorithms and data-driven models still rely on two fundamental assumptions:

  • The whole training dataset is initially provided to the algorithm for learning
  • The data are sampled from a fixed – albeit unknown – probability distribution

As a result, many state-of-the-art algorithms that can learn from a stream have been designed for data that are independent and identically distributed. Unfortunately, real-world learning scenarios violate this assumption far too often, and the models that are learned need an adaptation mechanism to provide reliable predictions.

Adaptation algorithms and strategies, together with techniques to detect changes and learn process dynamics, have been one of the most frequently addressed research challenges in the last few years. Many of these methods, however, are primarily heuristic in nature, with many parameters requiring fine-tuning. The problem of learning in a dynamic and nonstationary environment is still far from being solved.

Scope and Topics

This special session gathers the latest solutions to provide learning systems with the ability to operate in dynamic and changing environments, including methods for online learning, transfer learning, domain adaptation and change detection. Papers addressing either a theoretical or application-oriented perspective are welcome, as well as contributions presenting relevant applications that require learning in dynamic environments. The special session also welcomes papers addressing other challenges related to learning in dynamic environments such as online learning, class imbalance, anomaly detection, cognitive-inspired architectures and adversarial machine learning.

Researchers working in any of the related areas of learning in dynamic/nonstationary environments, concept drift or domain adaptation are encouraged to submit their contributions to this special session.

The special session topics include, but are not limited to:

  • Methods and algorithms for learning in dynamic/non-stationary environments
  • Transfer Learning and Domain Adaptation
  • Incremental learning, lifelong learning, cumulative learning
  • Online learning and stream mining algorithms
  • Change and covariate-shift adaptation
  • Semi-supervised learning methods for nonstationary environments
  • Ensemble methods for learning in nonstationary environments
  • Learning under concept drift and class imbalance
  • Learning recurrent concepts
  • Change-detection and anomaly-detection algorithms
  • Information-mining algorithms in nonstationary data streams
  • Cognitive-inspired approaches for adaptation and learning
  • Applications that call for learning in dynamic/non-stationary environments, or change/anomaly detection, such as:
    • adaptive classifiers for concept drift
    • adaptive/Intelligent systems
    • fraud detection
    • fault detection and diagnosis
    • network-intrusion detection and security
    • intelligent sensor networks
    • time series analysis
  • Development of datasets/benchmarks/standards for evaluating algorithms learning in non-stationary/dynamic environments
  • Adversarial machine learning


IJCNN-18 Learning from Big Graph Data: Theory and Applications

Organized by Shirui Pan (Shirui.Pan@uts.edu.au), Guodong Long, Chengqi Zhang

Recent years have witnessed a dramatic increase of graph applications due to advancements in information and communication technologies. In a variety of applications, such as social networks, communication networks, internet of things (IOTs), and human disease networks, graph data contains rich information and exhibits diverse characteristics. Specifically, graph data may come with the node or edge attributes showing the property of an entity or a connection, arise with signed or unsigned edges indicating the positive or negative relationships, form homogenous or heterogeneous information networks modeling different scenarios and settings. Furthermore, in these applications, the graph data is evolving and expanding more and more dynamically. The diverse, dynamic, and large-scale nature of graph data requires different data mining techniques and advanced machine learning methods. Today’s researchers have realized that novel graph learning theory, big graph specific platforms, and advanced graph processing techniques are needed. Therefore, a set of research topics such as distributed graph computing, graph stream learning, and graph embedding techniques have emerged, and applications such as graph-based anomaly detection, social recommendation, social influence analytics are becoming important issues for the research community.

Scope and Topics

This special session expects to solicit contributions on the advanced data mining and machine learning methods and applications for big graph data analytics. The topics of interest include, but are not limited to:

  • Feature Selection for Graph Data
  • Distributed Computing on Big Graphs
  • Dynamic and Streaming Graph Learning
  • Graph Classification, Clustering, Link Prediction Tasks
  • Graph Embedding
  • Learning from Unattributed/Attributed Networks
  • Learning from Unsigned/Signed Networks
  • Learning from Homogenous/Heterogeneous Information Networks
  • Social Recommendation
  • Social Influence Analytics
  • Anomaly Detection in Graph Data
  • Knowledge Graph and Its Applications
  • Sentiment Analysis
  • Cyberbullying Detection in Social Networks


IJCNN-19 Advanced Cognitive Architectures for Machine Learning

Organized by Jose C. Principe (principe@cnel.ufl.edu), Badong Chen (chenbd@mail.xjtu.edu.cn)

Current work in machine learning treats perception of the real world as pattern recognition. While this has been shown possible in pre-defined domains, with the availability of large data sets and labels, it is unclear that the approach scales up to autonomous vision, where the complexity of the world may supersede the gains associated with the unreasonable effectiveness of data. Biological organisms evolve in an unknown and uncertain world by creating models of the environment, storing past solutions that worked and using this knowledge effectively in the future. It may be possible to achieve similar performance for autonomous vision applications if we rethink the architectures and models currently being utilized in deep learning to include more parsimonious architectures and encapsulate in mathematics the voluminous literature available in cognitive science.

Scope and Topics

The goal of this special session is to provide a forum for focused discussions on extensions of conventional neural network architectures for space and time processing (conv and recurrent nets), how to go beyond labels, how to organize the representations achieved in current deep learning models in ways that the system can use and generalize without resorting to retraining of all the parameters, etc.

The focus of this special session is to attract both solid contributions and preliminary results which show the potentiality and the limitations of new ideas, refinements, or crosslinkage among the different fields of machine learning, AI and cognitive sciences to solve real world problems.

Examples of these possible extensions are:

  • Hierarchical Models for autonomous vision
  • Bidirectional processing architectures
  • Generative dynamical models of data
  • Parsimonious factorization of space time joint distributions
  • Incorporation of external memory in conventional machine learning architectures
  • Alternative learning paradigms beyond backpropagation
  • Self-learning and autonomous learning approaches
  • Brain inspired learning
  • Lifelong Learning
  • Statistical syntactic approaches for machine learning


IJCNN-20 Neurocomputation and Cognition

Organized by: Larry Manevitz (manevitz@cs.haifa.ac.il), Bernardete Ribeiro and Alex Frid

The field of neurocomputation is concerned with the possibility of computation in computers by following the paradigm and analysis of computation that occurs in neurons and the brain. In recent years this has resulted in breakthroughs in pattern recognition, machine learning theory, clustering, associative memory and fault tolerant computation.

Consequently, the precision resulting from the computational and mathematical viewpoint has led to insights helping to clarify aspects of one of the ultimate human research endeavors: understanding the manner in which human thought emerges from the organization of the human brain.

Scope and Topics

In this special session we will focus on three main directions.  Many researchers have interest in at least two of them.

  1. Recognizing and Classifying Cognitive and Brain activities using Neurocomputation and related technologies. That is, this is a very complex area wherein the neurocomputation serves as a tool; to help uncover subtle relationships.
    1. Example: Work by Nawa, Ando et al (CiNet, Osaka, Japan) on the ability to judge valence of free recall human biographical memories.
    2. Example: Work to do diagnosis and early prognosis of Neuro-degenerative diseases (such as Parkinson’s Disease) from various sorts of features and data (See e.g. Frid, Kantor et al)
    3. Example: Epileptic Seizure Predictions (See e.g. Baraia, Ribeiro et al)
    4. In this area, there is much “old fashioned” work on feature selection and choice of learning method. Yet systematic methodology is still in modus ascendii and one aim of the meeting is to clarify this. Some of this work does not fit easily into, e.g. deep learning techniques because of the relative paucity of data. Thus, also experimental theoretical work relating to how to manage with “small data” would be appropriate as long as it has links to the cognitive aspect.
    5. Example: (Bitan, Shalelshivili et al work on “Classification from Generation” in the context of deep grammatical task.)
  2. Developing Neurocomputation Models as a means to develop or test cognitive theories via simulation, especially cognitive modeling and computational models of creativity.
    1. Example: Computational Models for reading; computational models for autism and corresponding results. Included in this would be such ideas related to neural architectural ideas for such modeling. Example:  Work by Peleg et al  on reading
    2. Example: Computational models of cognitive phenomena (e.g. emotion, creativity, etc.)
    3. Example: Temporal Storage, Reservoir Computing and so on. Example: Work by Hazan et al on how requirements of robustness gives topological constraints on brain models.
    4. Example:  Temporal Sparse Distributed Memory (Manevitz)
  3. Discovery of Objective “Computational” Biomarkers.
    1. Example: Using neurocomputation and machine learning tools for neurodegenerative diseases   See, e.g. work by Frid et al
    2. Example: Discovery of Secondary Declarative Memory using modeling and feature selection. See e.g. work by Gilboa, Koilis et al

The special session invites submissions in any of the following (and related) areas:

  • Neurocomputation techniques as related to human cognitive issues
  • Understanding brain information processing underlying real-world tasks.
  • Validation of cognitive models using machine learning methods
  • Computational Biomarkers for Diseases
  • Computational Biomarkers for Cognitive Activity
  • Neurocomputational and Architectural Models of Creativity
  • Use of Neurocomputation and Machine Learning Tools to identify Physiological Features
  • Biologically Inspired Neural Computing


IJCNN-21 Deep Reinforcement Learning

Organized by Qichao Zhang (zhangqichao2013@163.com), Dongbin Zhao, Chaomin Luo

Pursuing higher “intelligent” systems has become the tendency of current artificial intelligence. It is expected that the new technique is capable to perceive some complicated problems and make decisions properly. For recent years, deep learning and reinforcement learning have made remarkable contribution in the field of “perception” and “decision” respectively. Deep learning has brought new techniques to perceiving high-dimensional data and processing complex information, while reinforcement learning has provided advanced solutions for nonlinear system control problems. Therefore, the combination of the above two methods, which is called deep reinforcement learning (DRL), has been essential to advanced artificial intelligence, and has also become the research hotspot currently. Recently, deep reinforcement learning has been applied to many fields such as the game Go, intelligent driving, Atari video games and so on.  This special session aims to discuss the recent development of DRL and its applications especially in intelligent driving.

Scope and Topics

The aim of this special session will be to provide an account of the state-of-the-art in this fast moving and cross-disciplinary field of deep reinforcement learning. It is expected to bring together the researchers in relevant areas to discuss latest progress, propose new research problems for future research. All the original papers related to deep reinforcement learning and its applications especially in intelligence driving are welcome.

The topics of the special session include, but are not limited to:

  • Deep reinforcement learning algorithms
  • Multi-agent deep reinforcement learning
  • Reinforcement learning and adaptive dynamic programming
  • Transfer learning
  • Deep reinforcement learning for video games
  • Deep reinforcement learning for intelligent driving
  • Navigation/Perception/Prediction/Planning/Control schemes for Intelligent driving
  • Adaptive dynamic programming for cooperative adaptive cruise control
  • End-to-End learning for intelligent driving
  • Dataset, Hardware implementation and algorithms acceleration for DRL


IJCNN-22 Ordinal and Monotonic Classification

Organized by Pedro Antonio Gutiérrez (pagutierrez@uco.es), Salvador García

Special session website: http://www.uco.es/grupos/ayrna/ordmon-wcci2018

Ordinal classification covers those classification tasks where the different labels show an ordering relation, which is related to the nature of the target variable. For example, financial trading could be assisted by ordinal classification techniques predicting not only a binary decision of buying an asset, but also the amount of investment. The decision could be categorized by {“no investment”, “low investment”, “medium investment”, “huge investment”}. Machine learning methods should consider the natural order among the classes and penalize differently the errors. Specific solutions have been recently proposed in the machine learning and pattern recognition literature, resulting in a very active field. Moreover, if a set of monotonicity constraints between independent and dependent variables has to be satisfied, then the problem is known as monotonic classification. In these problems, a higher value of an attribute in an example, fixing other values, should not decrease its class assignment. The monotonicity of relations between the dependent and explanatory variables is very usual as a prior knowledge form in data classification and it should be exploited to obtain more robust models.

Scope and Topics

This special session aims to cover a wide range of works and recent advances on ordinal and monotonic classification. We hope that this session can provide a common forum for researchers and practitioners to exchange their ideas and report their latest finding in the area. In particular we encourage submissions addressing the following issues:

  • Extensions of standard classification methods to ordinal and monotonic classification (support vector machines, Gaussian processes, discriminant analysis, etc).
  • Extensions of deep learning techniques to ordinal and monotonic classification.
  • Threshold models and decomposition methods for ordinal classification.
  • Non-standard predictive problems with ordering relation or monotonic constraints: Imbalanced classification, semi-supervised, multi-label or multi-instance learning.
  • Clustering and pre-processing methods for ordinal and monotonic data (data cleaning techniques, feature selection, over-sampling, under-sampling etc).
  • Evaluation measures for ordinal and monotonic classification.
  • Preference learning.
  • Data preprocessing (feature selection, noise filtering, etc…) for ordinal and monotonic classification.
  • Applications in medicine, information retrieval, recommendation systems, risk analysis… and any other real-world problems.
IJCNN-23 Deep Learning and Reinforcement Learning for Games


Organized by Yuanheng Zhu (yuanheng.zhu@ia.ac.cn), Dongbin Zhao, Risto MiikkulainenGames have always been one of the most attractive fields for computational intelligence. Researchers in this field are making effort to let computer beat human experts, and a lot of games have been conquered by computers, such as Chess, Atari, Go, No-limited Texas Hold’em poker. Games provide a convenient and comparative platform to test performance of methods in various areas, such as perception, strategy, planning, multi-agent, adaptation, etc. Nowadays with the development of deep learning and reinforcement learning, computers have gained more and more power in game intelligence. But for more complicated games like StarCraft that involves more game rules, more characteristics, more strategies, computers are still far from intelligence. New theories, algorithms and experiments need to be exploited to promote the community.

Scope and Topics

The aim of this special session is to provide a forum for the presentation of the latest theories, algorithms, experiments, and future research directions that use deep learning and reinforcement learning to solve problems that occur in games. The special session invites submissions in any of the following areas:Deep/reinforcement learning for one-character games;

  • Deep/reinforcement learning for multi-character games;
  • Deep/reinforcement learning for real-time strategy games;
  • Deep/reinforcement learning for zero-sum games;
  • Deep/reinforcement learning for game perception;
  • Deep/reinforcement learning for state representation in games;
  • Deep/reinforcement learning for human-computer interaction in games;
  • Deep/reinforcement learning for multi-agent operation in games;
  • Deep/reinforcement learning for long-term strategy and planning in games;
  • Deep/reinforcement learning for a wide range of games.
IJCNN-24 Neural Techniques for Artificial and Natural Locomotions


Organized by Zhijun Yang (Z.Yang@mdx.ac.uk), Vaibhav Gandhi, Mehmet Karamanoglu, and Felipe França

There are evidences showing that walking with adaptive gait patterns may be an acquired characteristic possessed by legged animals and humans. A baby animal or infant usually experiences an inept process to learn walking before becoming fully adaptive to a complex terrain. This learning process starts with reflexes reflecting involuntary responses to stimuli. It may involve a complex sequence of activities for sensorimotor integration, synchronization and coordination of cortical neurons and muscles. After the relevant cortical regions are well acquainted with the
external world, the animals are considered as trained and represent the most capable walking machine in nature.

Many theoretical and experimental approaches have been proposed intending to decipher the mechanisms underlying the natural locomotion while presenting its artificial intelligence (AI) counterpart. For instance, the finite state machine (FSM), in both deterministic and probabilistic variants, are traditionally used to model the gait pattern generation and transition. Recent years have seen the interest in this area of research growing rapidly thanks to the emergence of new computing methodologies using spiking neurons and neuronal populations, with the neuron complexity rangin from the simplest integrate-and- firing type to the classic Hodgkin-Huxley type. A special
Izhikevich neuron can display an abundant spectrum of real neuron activities. These methods show great potential in modelling the natural locomotion models in this relatively new research field. On the other hand, the modern technology provides us means of implementing the theoretic models by using high performance computing (HPC) techniques such as dedicated neuromorphic circuits, GPUs, FPGAs as well as deep learning tools.


Scope and Topics


This special session brings together the new research works from academics and industry related researchers in this prevalent area. The workshop aims to promote the applications of multidisciplinary methods in investigating and exploiting the neural mechanisms of natural locomotion. We invite papers on both theory and applications of the broad area of neural control for natural locomotion. The artificial locomotion systems, built upon the mechanisms underlying the natural locomotion systems, are particularly welcome. The topics of interest include, but are not
limited to, the following.

  • Neuroscience studies of natural locomotion and control
  • Novel mathematic models for gait pattern generation and transition
  • Central pattern generation models and applications
  • Somatosensory system, sensorimotor interaction and impacts on locomotion
  • Machine learning and deep learning methods applicable for motion control
  • Neuromorphic hardware implementation, parallel computing platforms using state-of- the-art
    hardware such as GPU or FPGA for neural control of locomotion
  • Software frameworks, such as robot operating system (ROS), applicable for robot control
  • Bayesian inference and hidden Markov models for decision making on robot motion
  • Novel finite state machine methods, implementation and applications in robotics
  • State-of- the-art robot projects using leading edge hardware and/or software
IJCNN-25 Multi-agent Reinforcement Learning and Adaptive Dynamic Programming Designs


Organized by Xiangnan Zhong (Xiangnan.Zhong@unt.edu), Haibo He, Huaguang Zhang

Reinforcement learning (RL) and adaptive dynamic programming (ADP) have been widely recognized as one of the “core methodologies” to achieve optimal control for intelligent systems. Recently, novel distributed RL and ADP approaches and architectures are proposed, including integrating game-based RL theory, biologically inspired ADP algorithms, deep RL, among others. They are widely applied on complex networked multi-agent systems to efficiently achieve optimal designs as well as relaxing the impractical requirement about actual knowledge of system dynamics. Intelligent distributed RL and ADP based data-driven control, learning, and optimization techniques
are vital towards the control modernization in complex dynamic environment.


Scope and Topics


This special session will provide a unique platform for researchers from different societies to share and publish innovative research in the field of multi-agent reinforcement learning and also provide an international forum to discuss their latest scientific developments and effective applications, to assess the impact of the approach, and to facilitate technology transfer. The special session will also enhance the discussion among different communities, including machine learning, computational intelligence, networked control systems, intelligent control, neuroscience, communications, among others, to explore more challenge cross-discipline topics along this direction.

We are particularly interested in but not limited to the following topics:

  • Intelligent multi-agent consensus control based on RL and ADP
  • Distributed RL and ADP based networked control systems
  • Distributed RL and ADP for multi-agent systems
  • Robust RL and ADP for uncertain multi-agent systems
  • Distributed RL and ADP based event-triggered/self- triggered control
  • Game-based RL and ADP for differential games
  • Applications in cooperative adaptive control for robotics with RL and ADP
  • New RL or ADP methods for human-robot/agent interaction
  • New RL or ADP methods for multi-robot systems
  • Theoretical foundation of multi-agent system designs based on RL and ADP
  • Data communication security in multi-agent reinforcement learning
IJCNN-26 Data Mining and Knowledge Discovery in Cyber-Physical Systems


Organized by Seiichi Ozawa(ozawasei@kobe-u.ac.jp), Bo Tang, Cesare Alippi, Haibo He

Special session website: https://my.ece.msstate.edu/faculty/tang/IJCNN2018CPS.html

The integration of embedded computation, communication, sensors and actuators has led to the mergence and development of Cyber-Physical Systems (CPS). Such systems cover vast application areas such as those referring to power grids, transportation, healthcare, manufacturing, structure health monitoring just to name the few.
Thanks to their ability to interact with the environment they are deployed in, the sensor platform associated with CPS collects a large amounts of data. By embedding intelligence in the application, researchers and engineers can enable new functions not previously possible, leading to many smart- X systems such as smart grid, smart healthcare, smart elderly care, smart agriculture, smart transportation, and smart building, among others. Computational intelligence and machine learning- based data mining techniques constitute the basis of intelligence, by handling the uncertainty coming from the physical world and support automatic decision making such that the smart systems are more robust, adaptive and fault tolerant to the dynamically changing environments.
The large-scale and heterogeneous nature of data in CPS raises a number of challenges for data mining and knowledge discovery. Firstly, the ubiquitous deployment of sensors produces massive amounts of data. In contrast to traditional big data challenges, mining from massive data streams in CPS usually needs be efficient in order to support real-time decision making. Meanwhile, data become easily obsolete, due to the dynamically changing environment (time variance). Learning from limited labeled data –this is mostly the case- requires advanced semi-supervised learning and unsupervised learning techniques for big data analysis. Moreover, heterogeneous data are usuallycollected from multiple sources or systems, and information fusion at data/feature/model levels is usually needed. The goal of this special session is to unveil these challenges and present the state- of-the- art research activities and results on all facets of data mining and knowledge discovery in CPS.


Scope and Topics


The goal of this special session is to unveil these challenges and present the state-of- the-art research
activities and results on all facets of data mining and knowledge discovery in CPS. The special
session invites submissions in any of the following areas:

  • Learning with limited or inaccurate supervision
  • Data and information fusion in smart environments
  • Data analytics with distributed computing
  • Data mining in Smart-X environments and the Internet of Things
  • Mobile computing and data mining
  • Intelligent transportation systems
  • Computational intelligence in structure health monitoring
  • CPS in smart city
  • CPS in smart healthcare and elderly care
  • CPS in smart agriculture
  • Big Data model in CPS
  • Reinforcement learning in CPS
  • Reliability and sustainability in CPS
  • Performance optimization in CPS
  • Application and implementation of CPS
IJCNN-27 Extreme Learning Machines


Organized by Guang-Bin Huang (egbhuang@ntu.edu.sg), Bao-Liang Lu, Jonathan Wu, Donald C. Wunsch II


Over the past few decades, conventional computational intelligence techniques faced bottlenecks in learning (e.g., intensive human intervention and time consuming). With the ever increasing demand of computational power particularly in areas of big data computing, brain science, cognition and reasoning, emergent computational intelligence techniques such as extreme learning machines (ELM) offer significant benefits including fast learning speed, ease of implementation and minimal human intervention.

Extreme Learning Machines (ELM) aim to break the barriers between the conventional artificial learning techniques and biological learning mechanism. ELM represents a suite of machine learning techniques for hierarchical neural networks (including but not limited to single and multi- hidden layer feedforward neural networks) in which hidden neurons need not be tuned: inherited from their ancestors or randomly generated. From ELM theories point of view, the entire networks are structured and ordered, but they may be seemingly ‘‘messy’’ and ‘‘unstructured’’ in a particular layer or neuron slice. ‘‘Hard wiring’’ can be randomly built locally with full connection or partial connections. Coexistence of globally structured architectures and locally random hidden neurons happen to have fundamental learning capabilities of compression, sparse coding, feature learning, clustering, regression and classification. ELM theories also give theoretical support to local receptive fields in visual systems.

ELM learning theories show that hidden neurons (including biological neurons whose math modelling may be unknown) (with almost any nonlinear piecewise activation functions) can be randomly generated independent of training data and application environments, which has recently been confirmed with concrete biological evidences. ELM theories and algorithms argue that “random hidden neurons” capture the essence of some brain learning mechanism as well as the intuitive sense that the efficiency of brain learning need not rely on computing power of neurons. This may somehow hint at possible reasons why the brain is more intelligent and effective than computers. ELM offers significant advantages such as fast learning speed, ease of implementation, and minimal human intervention. ELM has good potential as a viable alternative technique for large-scale computing and artificial intelligence.

The need for efficient and fast computational techniques poses many research challenges. This special session seeks to promote novel research investigations in ELM and related areas.Scope and Topics

All the original papers related to ELM technique are welcome. Topics of interest include but are not limited


• Universal approximation, classification and convergence, robustness and stability
• Biological learning mechanism and neuroscience
• Machine learning science and data science


• Real-time learning, reasoning and cognition
• Sequential/incremental learning and kernel learning
• Clustering and feature extraction/selection/learning
• Random projection, dimensionality reduction, and matrix factorization

• Closed form and non-closed form solutions
• Hierarchical solutions, and combination of deep learning and ELM
• No-Prop, Random Kitchen Sink, FastFood, QuickNet, RVFL, Echo State Networks
• Parallel and distributed computing / cloud computing


• Time series prediction, smart grid and financial data analysis
• Social media and video applications
• Biometrics and bioinformatics, security and compression
• Human computer interface and brain computer interface
• Cognitive science/computation
• Sentic computing, natural language processing and speech processing
• Big data analytics


• Lower power, low latency hardware / chips
• Artificial biological alike neurons / synapses

IJCNN-28 Adversarial machine learning in information security


Organized by Yun Li (liyun@njupt.edu.cn), Tao Li

Machine learning (ML) is related to a number of emerging security and privacy problems. Firstly, ML algorithms have been widely applied into critical security-related infrastructures, such as healthcare, automotive, finance and economics, etc. ML is increasingly important for autonomous real-time analysis and decision-making in security-sensitive areas with high-dimensional data. The use of learning methods in security-sensitive domains creates many frontiers for information security research, wherr adversaries may attempt to mislead or evade intelligent machines.

Additionally, ML techniques induce a wealth of privacy issues, due to the overabundance and accessibility of data. Thus, modern adversarial machine learning techniques require to be proposed for solving these problems related to information security and privacy preserving.


Scope and Topics


The goal of this special session to provide a forum for focused discussion on adversarial meachine learning techniques and their applications into security-sensitive tasks, such as malware detection, spam filtering, etc., improving security, reliability and the defense ability of known attacks, especially unknown ones. The main topics of this special session include, but are not limited to, the following:

Theoretical topics related to secure machine learning:

• Adversarial Learning
• Differential Privacy and Privacy-Preserving Learning.
• Sophisticated New Learning Algorithms in Security
• Vulnerability Testing Through Intelligent Probing (e.g. Fuzzing)
• Techniques and Methods for Crafting Synthetic Datasets
• Transferability in Machine Learning
• Evasion Attacks and Poisoning Attacks
• White-Box and Black-Box Attacks
• Adversarial Example Generation and Defense
Information Security Applications:
• Intrusion Detection and Response
• Anomaly Behavior Detection
• Phishing detection and prevention
• Malware and Authorship Identification
• Spam Message Filtering
• Biometric Recognition
• Traffic Sign Recognition

IJCNN-29 Machine Learning for Massive and Complex Urban Data Analytics


Organized by Allou Samé (allou.same@ifsttar.fr), Latifa Oukhellou, Luis Moreira-Matias, Michel Verleysen

In recent decades, the development of smart technologies and the advent of new observation capabilities have increased the availability of massive and complex urban data that can greatly benefit advanced machine learning approaches. For example, a large amount of urban data is collected by various sensors, such as smart meters, or provided by GSM, Wi-Fi or Bluetooth records, ticketing data, geo-tagged posts on social networks, etc. In conjunction with this digital transition, demographic growth, urban sprawl, increasing road congestion, and the desire to reduce environmental pollution have been responsible for the emergence of sustainable urban policies in a variety of areas such as mobility, energy, water and air quality. In such a context, several research works are carried out to develop Data-driven approaches in order to provide urban stakeholders with decision-making tools thus allowing them to better monitor urban systems.


Scope and Topics


This special session aims to gather researchers that develop advanced analytics, learning and visualization methods dedicated to complex data from the applicative fields of Transport, Mobility, Urban Planning, Energy and Environment. The challenges posed by the analysis of these urban data lies in the fact that they can be heterogeneous, noisy, incomplete, temporal, massive and spatiotemporal. With regard to operational issues, mining and visualizing such data can help to better understand and optimize the working of cities, offer the possibility of managing city infrastructures and their interactions with citizens differently, better meet future needs through improved forecasting and better match the supply of urban services to citizen demands.
The main application areas concerned by this session are

  • Intelligent Transport Systems and Mobility analysis (public transportation, sharing mobility systems,
    traffic analysis)
  • Logistics
  • Urban planning/Land use
  • Energy management and smart grids
  • Smart environment and resource management (water, air quality, waste management)
  • Smart urban health care

Methodological topics linked to these application domains include but are not limited to

  • Unsupervised ML for clustering and dimensionality reduction
  • Supervised ML for classification and regression
  • Online ML
  • Novelty and anomaly detection
  • ML forecasting models
  • Time series modeling and forecasting
  • Reinforcement learning
  • Urban data interactive visualization
IJCNN-30: Machine learning for big data: scalable algorithms and applications


Organized by Ahmad Taher Azar (ahmad_t_azar@ieee.org), Robi Polikar, Edwin Lughofer

Technology is advancing overtime and sizes of data sets generated are considered big and complex. The current database management tools and methods used to process data are inadequate; this paves way for big data analytics evolution and innovation. There is a growing need to develop big data tools and techniques to build capabilities to solve problems better than ever before. In current practices a number of industries are readily leveraging big data to their benefit.

Big data research has empowered the success of many applications in urban computing, social science, e-commerce, computer vision, natural language processing, speech recognition, bioinformatics, education, physics, chemistry, biology, and engineering. On the other hand, in order to enable learning with big data, scalable algorithms have attracted much attention in machine learning and data mining. Big data computing needs advanced technologies or methods to solve the issues of computational time to extract valuable information, in a realistic and practical time frame, without compromising the models’ quality. Numerous computational techniques for Big Data have been proposed, including stochastic optimization, parallel and distributed optimization, randomization, and GPU computing.


Scope and Topics


The aim of this special session is to provide an opportunity for international researchers to share and review recent advances in the emerging topic of machine learning with big data, with an emphasis on applications and scalable algorithms. The special session aims to solicit original, full length original articles on new findings and developments from researchers, academicians and practitioners from industries, in the area of machine learning with big data and scalable algorithms.

The topics of interest include, but are not limited to:

  • Active Learning with Big Data
  • Algorithmic paradigms, models, and analysis of Big Data
  • Big Data Clustering
  • Cloud computing for Big Data models and paradigms
  • Data Stream Mining Techniques for Big Data
  • Dimensional Reduction Techniques
  • Evolving modeling Techniques for Big Data
  • Large or Sequence Data Processing
  • Machine Learning Algorithms for Big Data
  • Meta-Heuristics Algorithms
  • Multi-dimensional Big Data
  • Non-linear Pattern Analysis
  • Optimization on Manifolds
  • Scalable and Incremental Algorithms
  • stochastic optimization
  • Visualization of Big Data Technology
IJCNN-31 Neural Models for Behavior Recognition

Organized by Pablo Barros (barros@informatik.uni-hamburg.de), Bruno Fernandes and Stefan Wermter

Website: https://www2.informatik.uni-hamburg.de/wtm/OMG-EmotionChallenge/?special_session=True


Categorizing and, to a certain degree, understanding human behavior is an important skill for autonomous systems. A robot able to identify whether a person needs help, an autonomous car to perceive whether a pedestrian is crossing the road, and a surveillance camera to localize someone in danger need a robust and reliable human behavior understanding. Although many different solutions have appeared in the past years, most of them do not take longer contextual information into consideration, usually only relying on instantaneous or short-context scenarios which limits the applicability of such systems in a real-world scenario.

The potential of having neural models that take into consideration contextual information is substantial and would be beneficial for this field.To have a system which is capable of learning, understanding and generalizing human behavior within a longer contextual scene is a challenging task. With recent neural computing technologies, such as training and understanding deep neural networks, life-long learning and unsupervised approaches, to name a few, we believe that autonomous systems which are capable of learning from long-context human behavior are feasible. This kind of processing would provide contextual information which would be beneficial to understanding and acting in real-world scenarios.

Scope and Topics
The aim of the special session is to gather a community to discuss human behavior analysis with emphasis on longer contextual information processing, moving forward to a real-world application of intelligent systems. We expect to provide a platform for discussion for students, starting and senior researchers from computer science, psychology, and neuroscience.

The special session invites submissions in any of the following areas:

  • Gesture processing, recognition and/or generation
  • Models for emotional behavior analysis and/or understanding
  • Personality trait analysis and/ generation
  • Action recognition and/or generation
  • Crossmodal human behavior learning
  • Temporal architectures for human behavior learning
  • Emotional feedback and modulation in decision-making processes
  • Life-long learning for human behavior processing


IJCNN-32 Biologically-inspired Neural Networks for Robotics and Mechanics

Organized by Chaomin Luo (luoch@udmercy.edu)

Biologically-inspired intelligence technique, an important embranchment of series on computational intelligence, plays a crucial role for robotics. The autonomous robot and vehicle industry has had an immense impact on our economy and society, and this trend will continue with biologically inspired neural networks techniques. Biologically-inspired intelligence, such as biologically-inspired neural networks (BNN), is about learning from nature, which can be applied to the real world robot and vehicle systems. Recently, the research and development of bio-inspired systems for robotic applications is increasingly expanding worldwide. Biologically-inspired algorithms contain emerging sub-topics such as bio-inspired neural network algorithms, brain-inspired neural networks, swam intelligence with BNN, ant colony optimization algorithms (ACO) with BNN, bee colony optimization algorithms (BCO), particle swarm optimization with BNN, immune systems with BNN, and biologically-inspired evolutionary optimization and algorithms, etc. Additionally, it is decomposed of computational aspects of bio-inspired systems such as machine vision, pattern recognition for robot and vehicle systems, motion control, motion planning, movement control, sensor-motor coordination, and learning in biological systems for robot and vehicle systems.

This special session seeks to highlight and present the growing interests in emerging research, development and applications in the dynamic and exciting areas of biologically-inspired algorithms for robot and vehicle systems (autonomous robots, unmanned underwater vehicles, and unmanned aerial vehicles).

Scope and Topics

Original research papers are solicited in related areas of biologically-inspired algorithms for robotics. Submissions to the Special Session should be focused on theoretical results or innovative applications of computational intelligence of biologically-inspired algorithms (such as BNN) for robot and vehicle systems. Specific topics for the special session include but are not limited to:

  •  Biologically-inspired neural networks for robotics
  •  Deep neural networks and learning systems for robotics such as motion planning, navigation, mapping, localization, image processing, etc.
  •  Bio-inspired system on computer vision and image progressing for robotics
  •  Human-like learning for robotics
  •  Theory, design, and applications of neural networks and related learning systems for robotics and vehicles
  •  Neuro-dynamics based models for robot and vehicle systems
  •  Evolutionary optimization, machine vision, pattern recognition for robot and vehicle systems
  •  Brain-inspired neural networks for robotics
  •  Swarm intelligence with BNN for robotics
  •  Evolutionary neuro-computing for robot and vehicle systems
  •  Bio-inspired system on machine learning, intelligent systems design for robotics
  •  Cellular automata for robotics
  •  Immune systems with BNN for robotics
  •  Ant colony optimization algorithms (ACO) with BNN for robotics
  •  Bee colony optimization algorithms (BCO) with BNN for robotics


IJCNN-33 Neural Intelligence After Tomorrow

Organized by Ivan Tyukin (I.Tyukin@le.ac.uk), Danil Prokhorov, and Alexander N. Gorban

Neural Intelligence, as a common theme across Neural Networks (NN) theory, including artificial, natural, and neuromorphic, and Artificial Intelligence (AI) as a broader theoretical filed, has seen major and unprecedented progress in the last two decades. Re-surfacing of Deep Learning, Convolutional Neural Networks coupled by widely accessible and evolving open-source development platforms as Tensorflow, Caffe, Torch etc and vast amounts of readily available data for training and testing gave rise to a novel reality: the world of mass-produced and growing AIs. The question, however, is if such a growth has limits, and if so then what these limits are?

Scope and Topics

The session is envisioned as a forum for communicating novel ideas, vision, and radically new approaches in the area from leading experts and practitioners. The focus of the session is to discuss fundamental challenges of modern AI and NN theory and practice, clarify ways to overcome these challenges, and lay out a vision for future AI & NN technologies and their development. The special session invites and welcomes submissions in any of the following areas:

  • Non-destructive AI error correction and guaranteed knowledge transfer
  • Evolving networks with self-esteem and collective AI
  • Neuromorphic intelligence and computation
  • Randomized learning
  • High-dimensional optimization, approximation, and quasiorthogonal dimension
  • The Depth of Deep: advantages and limitations of Deep Learning
  • Trusting AI with life


IJCNN-34 Ensemble models for Pattern Recognition and Data Mining

Organized by Brijesh Verma (b.verma@cqu.edu.au), Lipo Wang and Mohammed Bennamoun

Brief Description

There is a great interest of ensemble models among the pattern recognition and data mining researchers. Many ensemble models have successfully demonstrated the capability of solving real world problems in document analysis, speech recognition and medical diagnosis. The purpose of the special session on ensemble models for pattern recognition is to address the latest developments of ensemble learning algorithms for numerous applications in pattern recognition and data mining.

This session aims to bring together ensemble learning and pattern recognition researchers to demonstrate latest progress, emphasize new research questions and collaborate for promising future research direction.

Scope and Topics

The theme of the session is the application of ensemble models to pattern recognition and data mining. The
list of topics includes and is not restricted to the following:

  • Ensemble models
  • Ensemble of classifiers
  • Fusion of classifiers
  • Ensemble of neural learning
  • Ensemble of deep learning
  • Ensemble of neural-evolutionary learning
  • Hybrid learning
  • Fusion of neural learning
  • Fusion of feature learning
  • Fusion of probabilistic graphical models
  • Fusion of supervised, unsupervised and reinforcement learning
  • Ensemble learning applications in image processing, document processing, video processing,
    medical imaging, object recognition, face recognition and other pattern recognition and data mining


IJCNN-35 Evolutionary Computation for Neural Networks

Organized by Yeh Wei-Chang Yeh (weichang.yeh@gmail.com), Vera Y.Y. Chung (vera.chung@sydney.edu.au)

This Special Session on “Evolutionary Computation for Neural Networks” mainly focus on the research of exploring Evolutionary Computation and Swarm Intelligence methodologies for Artificial Intelligence and multimedia applications. Although a significant amount of research have been done in neural networks, there still remain many open issues and intriguing challenges in optimizing neural networks architectures, especially Deep Learning networks for Artificial Intelligence and Multimedia applications.

Scope and Topics

In recent years, evolutionary computation based Neural Networks has attracted the interest of many scientists. Evolutionary computation involves the study of collective behaviour of individuals in a population, and how it interacts with one another and their environments. The emerging multimedia computing such as multimedia refereeing expression, multimedia knowledge extraction and reasoning tasks have gained high attention recently. The directly borrow the models from the neural networks cannot fit the requirements of most Artificial Intelligence (AI) and multimedia applications. Therefore, we need to consider how to use optimization algorithms to train the neural network or how to use the optimization algorithm to design the novel neural network architectures for emerging AI theories, technologies, simulations and applications. The goal of this special session is to stimulate researchers in providing novel evolutionary computation methods for Neural Networks, AI and applications.The topics of interest for this special session include, but are not limited to, the following:

  • Particle swarm optimization and neural networks
  • Evolutionary computation based neural network for multi-objective optimization algorithms
  • Survey and comparative studies of swarm intelligence and neural network techniques
  • Real-world problem solving using swarm intelligence based neural network methods
  • New services and applications based on evolutionary computation and neural networks
  • Deep neural networks for content based analysis and understanding
  • Enhanced deep neural networks for multimodal data perception and reasoning
  • Emerging applications of deep neural networks in multimedia search and retrieval


IJCNN-36 Deep, Transfer and Reinforcement Learning for Robotics and Intelligent Agents

Organized by Abdulrahman Altahhan (a.altahhan@leedsbeckett.ac.uk),Vasile Palade

Deep Learning has been under the focus of neural network research and industrial communities due to its proven ability to scale well into difficult problems. Recent developments meant that intelligent agents can benefit from previous experiences of other agents/systems and build on top of it by utilising Transfer Learning. Hence, an agent can reschedule a set of End-To- End learning tasks that differ in purpose and are similar in nature to accomplish a set of tasks that encompasses a full story of events, hence achieving the ultimate goal of a fully Autonomous Intelligent Agent AIA.

Reinforcement learning (RL) is considered the model of choice for problems that involve learning from interaction. Typically these applications involve processing a stream of data coming from different sources such as pervasive and smart sensors. Currently RL algorithms still takes longer than can be afforded by physical systems, however with the usage of Transfer Learning, difficult application that needs a lot of training can become attainable. Furthermore, considering recent developments of RL algorithms that can take advantages of deep learning, the realisation of a truly Autonomous AI system becomes more auspicious.

Robots, whether physical or virtual, with the help of Deep Learning and Reinforcement Learning are promising more than ever to solve important problems in domestic, industrial and military domains, helping humans to achieve their maximum potential and eliminate or mitigate risks and problems that can occur due to aging, mental and physical illness, social constrains, natural disaster or hazardous and search and rescue situations.

Scope and Topics

This special session will provide a unique platform for researchers to showcase the synergies that can be created by instilling Deep Learning and/or Reinforcement Learning into Robotics and the emerging techniques that arise from the challenges of embedding deeply trained and reinforced robotics in our daily life activities and/or rare emergencies. It will allow communities to share their research experience and proposed solutions to problems that used to be considered not yet practical for robotics, in order to allow this multidisciplinary area to flourish further. It will focus generally on the potential benefits of the different approaches of combining Robotics, RL and/or DL.

In addition, the session aims at bringing more focus onto the potential of infusing transfer and reinforcement learning frameworks within the physics of a robotic system to allow it to deeply represent its environment through its sensory information and to make its experience easily accessible to other agents.
Topics of interest include, but are not limited to the following:

  • Novel RL algorithms for Real and Simulated Robotics
  • Novel DL architectures suitable for Real time systems and Robotics
  • Transfer Learning and/or Curriculum Learning for Robotics
  • Autonomous agents/cars and Multi-agents Applications
  • Reinforcement Learning and Deep Learning for Hazardous Robotics Applications
  • Deep Robotics Learning for the Elderly
  • Deeply Hierarchical and other Deep RL algorithms for Robotics
  • Deep RL for Sensor Fusion and Uncertainty
  • Other Robotics and Intelligent Agents application


IJCNN-37 Advances in Document Analysis and Recognition

Organized by Umapada Pal, Michael Blumenstein and Nabin Sharma (Nabin.Sharma@uts.edu.au)

Document image analysis, particularly, Optical Character Recognition (OCR) have been the primary research areas in the field of pattern recognition, artificial intelligence and image processing, for many decades. There are many commercial applications available related to OCR, automatic form processing, postal automation, to mention a few. OCR and document image analysis is still a challenging research area due to advancement in the digital technology and availability of low price imaging devices, which is in the verge to replacing traditional digital scanners. Mobile

phones and other portable digital cameras have given birth to new challenges related to camera and video based document processing. The document images and videos captured using a digital camera suffers from low resolution, contrast, blur, distortion, noise, etc., to mention a few. This pose a fresh challenge to the document analysis community.

Increasing popularity of neural networks aka. Deep Leaning have shown remarkable results in image classification, object detection and segmentation tasks. Hence, use to deep neural network has become an obvious choice to explore their potential in solving the complex tasks in document image analysis and OCR.

Scope and Topics

This special session aims to bring together the current research progress on Machine learning and Deep learning theories and its applications in Document Image Analysis. Special attention will be devoted to handle advanced issues of network architecture design, real-time performance criteria for various applications to Document Image Analysis. The special session invites high quality submissions in various research areas in Document Image Analysis.

The main topics of this special session include, but are not limited to the following:

  • Document image analysis and systems
  • Handwriting recognition
  • Writer identification
  • Text classification/analysis
  • Camera-based document processing
  • Graphics recognition
  • Multi-lingual character recognition
  • Historical document processing
  • Document Forensics
  • Human-Document interaction
  • Deep Learning in Document Analysis


IJCNN-38 Neural Approaches for Natural Language

Organized by Marco Pota (marco.pota@icar.cnr.it), Massimo Esposito

Mining information from textual data is a frontier of increasing interest in recent research. While the usefulness of such information is very valuable, due to the huge amount of written and spoken available textual sources, mining approaches are still not established, due to difficulties related to the unstructured and variable nature of language. Moreover, systems are often tested for English or few other languages, and their application is restricted to some narrow domain, while multilingual and reusable approaches could be more powerful. The last trends imply using deep neural approaches for different issues related to text mining, going from Natural Language Processing, to
Entity Recognition, till the development of Question Answering paradigm and Conversational Systems.

Scope and Topics

This Special Session provides an interdisciplinary forum for researchers and developers to present and discuss experiences, ideas, and research results in this area. Original contributions are sought, covering the whole range of theoretical and practical aspects, technologies and systems for text mining. In particular, the Session will focus on deep learning methodologies, algorithms and techniques, addressing various issues of this research field. Topics are encouraged, but not limited to, at least one of the following areas:

  • Word embedding and character embedding for numerical representation of text;
  • Part of Speech tagging of words;
  • Parsing of sentences;
  • Entity Recognition;
  • Question Answering;
  • Conversational Systems;
  • Application and comparison of deep neural architectures for textual data;
  • Innovative neural architectures for textual data;
  • Hybrid approaches for Natural Language Processing.


IJCNN-39 Industrial Applications

Organized by Catherine Huang (Catherine.huang@intel.com)

Recent years deep learning has been a powerful tool in a wide variety of applicaions. The most advanced models are used in diverse data modalities, such as images and speeches, and to complex task, such as games and genrating molecules. Industrial applications, such as robotics and autonomous driving, have pushed the deep learning and reinforcement learning to the front end. It is important to understand the needs and challenges on how to deploy the cutting-edge research from research labs to real-world environments.

Scope and Topics

The aim of the Industrial Special Session is to provide a forum for researchers and developers from academia, industry, and government to interact and share their research ideas and results in the field. The researchers will present the latest findings in all aspects, including algorithms, applications, new research prototypes, and proof of concepts. The papers may highlight the main contributions and challenges in applying artificial intelligence in real-world applications. The special session invites submissions in any of the following areas:

  • Conventional machine learning
  • Deep learning
  • Reinforcement learning
  • Apprenticeship learning
  • Learning to learning from a few examples
  • Collaborative learning
  • Brain computing
  • Applications in transportation, finance, health, robotics, education, social media and


IJCNN-40: Advanced Machine Learning Techniques for Computational Biology

Organized by Anirban Mukhopadhyay (anirban@klyuniv.ac.in), Ujjwal Maulik, Sanghamitra Bandyopadhyay

Website: https://sites.google.com/site/wcci2018mlcb/

With technological advances, the field of computational biology is facing challenges from explosion of molecular biology and cell profiling data. In the context of huge data availability, many computational biology problems, such as sequence analysis, gene identification, protein structure and function prediction, microarray analysis, biological network analysis and rational drug design can be modeled as supervised, semi-supervised and unsupervised learning problems. Advanced machine learning techniques like deep neural networks, probabilistic graphical models, support vector machines, semi-supervised learning, convolutional neural networks, hidden Markov models, Bayesian learning, random forests, association rule learning, reinforcement learning and multiobjective optimization have already been proved to be very effective in learning from large data sets in many research fields. Thus it is of great interests to see how these advanced machine learning techniques perform in computational biology problems.

Scope and Topics

The main aim of this special session is to bring together the scientists and researchers of this field to exchange the latest advances in theories and experiments in this field of research. Researchers are invited to submit original and unpublished works that deal with theoretical and experimental results of advanced machine learning techniques in the following and other related areas.

  • Sequence analysis including next-generation sequencing
  • RNA and protein structure prediction
  • Protein function prediction
  • Microarray analysis
  • Bio-marker identification
  • Bio-molecule ordering
  • Protein sub-cellular location prediction
  • MicroRNA target prediction
  • Biological motif finding (sequence and network motifs).
  • Systems biology and network analysis


IJCNN-41 Machine Learning for Encoding‐Decoding the Brain neural activityOrganized by Mahboobeh Parsapoor (mahboobeh.parsapoor@mail.mcgill.ca)

One of the exciting research fields in neuroscience domain is neural coding that encompasses neural encoding (understanding of how information of sensory stimuli is represented as action potentials by individual neurons or a network of neurons) and neural decoding (understanding the process of retrieving information of sensory stimuli from action potentials of neurons). Even statistical methods such as multivariate pattern analysis (MVPA) have shown reasonable performance in analyzing neural data (e.g. MEG and EEG data), the advance of powerful machine learning (ML) tools (e.g., deep structure) bring excellent opportunity for developing high-performance neural coding tools.


Scope and Topics


The goal of the special session is to discuss challenges and present successful ML algorithms (e.g., neural networks, deep learning) as encoding and decoding tools. The session also aims to discuss how we can improve traditional neural coding methods by taking inspiration from cognitive functions of the brain; how new ML tool can be developed for neural coding problems; how preprocessing methods can be developed to increase the accuracy of traditional neural coding methods; how available toolbox such as “Braindecide 0.1.7”, “Princeton-mvpa-toolbox” can be utilzed for new neural coding problems. The special session invites submissions in any of the following areas:


  • Machine Learning for Neural Encoding-Decoding
  • Deep Learning for Neural Encoding-Decoding
  • Bio-inspired Techniques for Neural Encoding-Decoding
  • Analyzing EEG and MEG Data for Neural Encoding-Decoding