I work in several areas where there is a lot of (infinite amount?) work to be done by ambitious students. The topics below cover some of those fields, but I do not update this page frequently so it may happen that some topic may be outdated and some new topics should be added. Most of these topics are well suited for smaller projects (BSc, MSc) and may be later expanded into PhD projects. Their are divided roughly into the following categories:

• Applications of computational intelligence methods.
• Computational intelligence and knowledge discovery.
• Cognitive topics, mind models.
• Theoretical topics in computational intelligence and physics.

Description that follow are deliberately rather general. First, things are changing too quickly to provide more precise description, second I do not have time to update these pages and describe all precisely.
Please see also on-line publications list of my group and my own list of publications.

Cognitive science, modeling of mind.

Long-term goal: understand how mind functions arise from brain activity; implement computational models; used them to design and explain novel experiments. On the applied side, used mind models as a basis for humanized interfaces (HITs), our future companions in computers and phones.

• Humanized interfaces (HIT).

This is a larger project that is a part of large program started in the fall 2004 at Nanyang Technological University in Singapore. We are working on modular, extensible platform with talking head that would recognize the user (biometric functions), accept speech input in a robust (albeit limited) way, accept gestures, recognize emotional states on your faces, anticipate certain actions, have some natural language analysis capability, reasoning capability, and control behavior (speech, gestures, face expressions), find information in well-structured and free-text Internet databases and report it. We intend to have HIT platform both on computers and on portable devices (PDAs, telephones) and use it in variety of applications.

The graphics head is (at least initially) based on Haptek's "Virtual Friend" technology, and initially it was coupled with free Microsoft speech API. The emphasis is placed on integration of different technologies, identification of the functions that are difficult, narrowing the interaction to well-defined context to make it robust.

Within this general project we are interested in many sub-topics that should be integrated into the whole system:

• Design of the modular, extensible platform for HIT, integration of different functions, control mechanisms. This is central part
• Lower-level functions: signal analysis, filtering of noise, active noise control, blind signal separation, pre-processing signals from various sensors that may be distributed in some building or area, or microphones and cameras attached to the portable device.
• Auditory image understanding and sound localization, recognition of various natural sounds.
• Recognition of speech in narrow domains 10-50 words, and improvement of general speech recognition, recognition of prosody in speech.
• Visual understanding: face recognition and face tracking.
• Agents that try to understand user's questions, making it more precise by asking additional questions.
• Web agents that search for information in free-text sources and in structured databases.
• Emotions: recognition of emotions on people's faces and in their voices, see our DREAM demo.
• Speech synthesis and synchronization of sounds with lip movements of avatar.
• Behavioral simulation: emotional reactions of avatar.
• Memory models: recognition, semantic and episodic associative memory models that can be applied for large number of facts.
• Natural language analysis in question/answer systems, formulation of questions, knowledge representation.
• Mind models, integrating perception, memory and natural language processing, based on extension of general models such as ACT-R or SOAR.
• Cooperation of multiple agents, transition from simple neural systems with almost no knowledge to complex agents communicating their knowledge to form a competent multiagent entity.

There are also many applications, and each may be a separate project using whatever functionality we already have. In particular I am interested in word games played through the talking head interface. Main projects here are on the 20 question games and educational applications.

• Simulation of perceptual and cognitive brain functions.

The long term goal is to create a system that will capture most of neuroscience knowledge in a global simulator, and understand how the higher-level functions arise form brain neurodynamics. This is a subject of a large integrated EU project submitted in Sept. 2004, coordinated by John Taylor from King's College London. My group was mainly involved in modeling of perceptual systems, learning and memory, schemata learning and integration. Modeling will be done at the spiking neuron level. The project did not get funding.
I have developed a more ambitious approach to machine consciousness, involving three kinds of memory subsystems (frontal for working memory, parietal for associative memory and hippocampal for episodic memory), recreation of intermediate sensory information in the occipital lobes, commentator (both non-verbal and verbal) system in the temporal lobes and a reward system including midbrain VTA (Ventral Tegmental Area) for reinforcement learning. So far we do not have a large-scale simulator to implement this model, students are needed!

There are many subprojects in cognitive neuroscience connected with this main theme. I am interested very much in simplification of neurodynamics, leading to models that can be understood in psychological terms, for example via fuzzy symbolic dynamics. I have tried to analyze simple categorization experiments from this point of view, and work on neural realization of psychological spaces, but it is just a beginning of an attempt to bridge neuroscience and psychology, brain and mind.

I collaborate with philosophers, trying to grasp philosophical implications of cognitive sciences, understand basic concepts related to mind, their limitations, and relation to the brain dynamics. Difficulties in the philosophy of mind in understanding consciousness may be solved if we learn to approximate correctly, using symbolic language, real brain processes. A new vision of man is very different form traditional, painted by traditional philosophy grounded in old religious ideas. We need to understand better the origins of the concepts used to describe mind, and build a new vision of man in agreement with cognitive science, develop ecology of mind.

I try to explain mysteries of quantum mechanics from the cognitive perspective (as Lakoff and Nunez did with mathematics).

Topics in computational intelligence and knowledge discovery.

Long-term goal: create automatic systems capable of discovering knowledge in data, integrating different computational intelligence models. Our Ghostminer datamining package is developed in collaboration with the branch of Fujitsu company, called FQS Poland.

We have many projects aimed at improvement of visualization, classification, clusterization and approximation methods in Ghostminer. The main directions of our work are:

• Methods that can deal with objects of complex structure, going beyond vector space representation of knowledge.
• Heterogeneous systems, using different types of elements, for example different neural transfer functions in each node, different types of tests for nodes in decision trees, or different similarity functions for each prototype.
• New architectures and training methods for neural networks, distance-based MLPs, global minimization learning methods, numerical search based learning techniques.
• Separable function networks for classification and systematic reasoning.
• Prototype-based reasoning and logical rules as alternative to neurofuzzy systems.
• Information selection, feature aggregation and selection, creation of new features.
• Visualization methods, multidimensional scaling variants, non-linear mappings, visualization applied to understanding of neural network learning dynamics and mappings learned.
• Creation of general CI theory based on similarity concepts.

We also work on text databases, trying to improve search techniques, especially for medical applications. Information retrieval has been so far separated from work in cognitive science, at it seems that various memory models should be a basis for good information retrieval systems. We try to introduce recognition, semantic and episodic memory mores in information retrieval systems. .

Applications of computational intelligence.

In collaboration with Children's Hospital Research Foundation (CHRF) of Cincinnati (John Pestian, Jarek Meller, Rafal Adamczak) we work on the following topics:

• We do some work in bioinformatics, analyzing different data mostly from the point of view of understanding their structure.
• Applications of information retrieval to unstructured medical annotation data.
• Bioengineering work on cognitive toys for infants.
• A new topic that we plan to start is classification of medical data, and modeling of disease evolution on genomic and phenotypic level.

Some other applications we work on are related to HIT projects, and information retrieval. .

Theoretical topics.

I am interested in Bayesian approaches, inference, density estimation, mean field theory; some practical applications have been developed (from example TAP approach) but still this is a lot of work to be done.

• A list of some current and older topics.

Last modification 27.10.2008, by Wlodzislaw Duch; my local homepage