Abstract

Real world applications and especially those dealing with complex data mining ones make quickly appear the insufficiency of academic (called also sometime theoretical) approach in solving such categories of problems. The difficulties appear since definition of the “problem’s solution” notion. In fact, academic approaches often begin by problem’s constraints simplification in order to obtain a “solvable” model (here, solvable model means a set of mathematically solvable relations or equations describing a processing flow, a behavior, a set of phenomena, etc…). If the theoretical consideration is a mandatory step to study a given problem’s solvability, for a very large number of real world dilemmas, it doesn’t lead to a solvable or realistic solution. Difficulty could be related to several issues among which:
-  large number of parameters to be taken into account making conventional mathematical tools inefficient, 
-  strong nonlinearity of the data (describing a complex behavior or ruling relationship between involved data), leading to unsolvable equations,
-  partial or total inaccessibility to relevant features (relevant data), making the model insignificant,
-  subjective nature of relevant features, parameters or data, making the processing of such data or parameters difficult in the frame of conventional quantification,
-  necessity of expert’s knowledge, or heuristic information consideration,
-  imprecise information or data leakage.

Examples illustrating the above-mentioned difficulties are numerous and may concern various areas of real world or industrial applications. As first example, one can emphasize difficulties related to economical and financial modeling (data mining, features’ extraction and prediction), where the large number of parameters, on the one hand, and human related factors, on the other hand, make related real world problems among the most difficult to solve. Another illustrative example concerns the delicate class of dilemmas dealing with complex data’s and multifaceted information’s processing, especially when processed information (representing patterns, signals, images, etc.) are strongly noisy or involve deficient data. In fact, real world and industrial applications, comprising image analysis, systems and plants safety, complex manufacturing and processes optimization, priority selection and decision,, classification and clustering are often those belonging to such class of dilemmas.
If much is still to discover about how the animal’s brain trains and self-organizes itself in order to process and mining so various and so complex information, a number of recent advances in “neurobiology” allow already highlighting some of key mechanisms of this marvels machine. Among them one can emphasizes brain’s “modular” structure and its “self-organizing” capabilities. In fact, if our simple and inappropriate binary technology remains too primitive to achieve the processing ability of these marvels mechanisms, a number of those highlighted points could already be sources of inspiration for designing new machine learning approaches leading to higher levels of artificial systems’ intelligence.
This plenary talk deals with machine learning based modular approaches which could offer powerful solutions to overcome processing difficulties in the aforementioned frame. It focuses machine learning based modular approaches which take advantage from self-organizing multi-modeling ("divide and conquer" paradigm). If the machine learning capability provides processing system’s adaptability and offers an appealing alternative for fashioning the processing technique adequacy, the modularity may result on a substantial reduction of treatment’s complexity. In fact, the modularity issued complexity reduction may be obtained from several instances: it may result from distribution of computational effort on several modules (mluti-modeling and macro parallelism); it can emerge from cooperative or concurrent contribution of several processing modules in handling a same task (mixture of experts); it may drop from the modules’ complementary contribution (e.g. specialization of a module on treating a given task to be performed).
One of the most challenging classes of data processing and mining dilemmas concerns the situation when no a priori information (or hypothesis) is available. Within this frame, a self-organizing modular machine learning approach, combining "divide and conquer" paradigm and “complexity estimation” techniques called self-organizing “Tree-like Divide To Simplify” (T-DTS) approach will be described and evaluated.

Bio

Graduated in fundamental physics in June 1985 from PARIS 7 – Jussieu University. He received his MSc. in Microelectronics and chip architecture from University PARIS 11 (PARIS-SUD), Orsay, France, in September 1986.
Received his Ph.D. in Electrical Engineering and Computer Sciences from University PARIS 11 (PARIS-SUD), Orsay, France, in February 1990.
From 1989 to 1990, he worked as assistant professor at Institut d’Electronique Fondamentale (Institute of Fundamental Electronics) of PARIS 11 University and CNRS (National Center of Scientific Research), Orsay, France.
In 1990, he joined Senart-FB Institute of Technology of University PARIS 12 – Val de Marne (Lieusaint, France), where he worked from 1990 to 1998 as assistant professor. In 1995, he received the DHDR Doctor Hab. degree (senior research doctorate degree) from University PARIS 12 – Val de Marne.
Since 1998 he works as Chair Professor in Electrical Engineering of Senart-FB Institute of Technology of University PARIS 12 – Val de Marne.
From 1992 to 2000 he has been head creator and head of DRN (Neural Networks Division) research group. From 2001 to 2004 he has been head of Intelligence in Instrumentation and Systems Laboratory of PARIS 12 – Val de Marne University located at Senart-FB Institute of Technology. Co-creator of Images, Signals and Intelligent Systems Laboratory (LISSI / EA 3956) of PARIS 12 – Val de Marne University in 2005, he is actually director of SCTIC research group, one of the two research groups of this laboratory.
He has worked on both digital and analog implementation of massively parallel processors arrays for image processing by stochastic relaxation, electro-optical random number generation, and both analog and digital Artificial Neural Networks (ANN) implementation.
Author and coauthor of more than 220 publications in international scientific journals, books (Springer, Kluwer, etc…), international conferences’ and symposiums’ proceedings, he has been regularly invited as key-note and invited lecture by international conferences and symposiums (IEEE, IFAC, etc…). His current research interests include
- large ANN structures behavior modeling and implementation,
- self-organizing, modular and hybrid neural based information processing systems and their software and hardware implementations,
- design and implementation of real-time neuro-control,
- humanoid robotics
- collective robotics
- neural based fault detection and diagnosis systems.

Since 1996 he is a permanent member (elected Academician) of International Informatization Academy. In 1997, he was also elected as permanent Academician of International Academy of Technological Cybernetics.
 

"HCI through the ‘HC Eye’ (Human-Centred Eye): Can Computer Vision Interfaces  Extract the Meaning of Human Interactive Behaviour?" By Dr. Claude C. Chibelushi, Faculty of Computing, Engineering & Technology, Staffordshire University, UK

Abstract

Some researchers advocating a human-centred computing perspective have been investigating new methods for interacting with computer systems. A goal of these methods is to achieve natural, intuitive and effortless interaction between humans and computers, by going beyond traditional interaction devices such as the keyboard and the mouse. In particular, significant technical advances have been made in the development of the next generation of human computer interfaces which are based on processing visual information captured by a computer. For example, existing image analysis techniques can detect, track and recognise humans or specific parts of their body such as faces and hands, and they can also recognise facial expressions and body gestures.

 This talk will explore technical developments and highlight directions for future research in digital image and video analysis which can enhance the intelligence of computers by giving them, for example, the ability to understand the meaning of communicative gestures made by humans and recognise context-relevant human emotion. The talk will review research efforts towards enabling a computer vision interface to answer the what, when, where, who, why, and how aspects of human interactive behaviour. The talk will also discuss the potential impacts and implications of technical solutions to problems arising in the context of human computer interaction. Moreover, it will suggest how the power of the tools built onto these solutions can be harnessed in many realms of human endeavour.

Bio

Dr. Claude C. Chibelushi is Reader in Digital Media Processing in the Faculty of Computing, Engineering and Technology, at Staffordshire University. He is Co‑Director of the Centre for Information Intelligence and Security Systems, and the Faculty Head of Postgraduate Research Studies. Prior to joining Staffordshire University, he was Senior Research Assistant at the University of Wales Swansea, after being Lecturer in the Department of Electrical and Electronic Engineering at the University of Zambia. He holds a Ph.D. in Electronic Engineering from the University of Wales Swansea, an M.Sc. in Microelectronics and Computer Engineering from the University of Surrey, and a B.Eng. in Electronics and Telecommunications with distinction from the University of Zambia. He was awarded a Beit Fellowship to undertake his Ph.D. studies. He is a Chartered Engineer, and a Member of the Institution of Engineering and Technology.

 His research interests include multimodal recognition, robust pattern recognition, medical image analysis, and image synthesis and animation. He is one of the pioneers of research on audio-visual speaker recognition targeted at applications such as human computer interfaces and biometrics.

Back