Computational Intelligence Laboratory, Department of Informatics, UMK

List of old/finished/aborted projects in Neural and Cognitive Modeling

Some projects are continued with other people and under different names.

  1. Title: Comparison of classifcation trees and neural network methods

  2. Objective: compare decision trees with neural classificators.
    Participants: F. Kamieniecki, W. Duch
    Time: 1996-1997
    Results: MSc thesis by Filip Kamieniecki. Only a few comparisons done so far.
  3. Title: Approximations to the biologically motivated neuron transfer functions

  4. Objective: Try to solve some optimization problems using modifications of SOM/MDS
    Participants: W. Duch, A. Ludwiczewski
    Time: 1996-1997
    Results: Mgr Ludwiczewski left this project because of the slow progress. The problem is still open and is quite suitable for a PhD.
  5. Title: SOM and MDS for optimization

  6. Objective: Try to solve some optimization problems using modifications of SOM/MDS
    Participants: W. Duch, Z. Kaczmarek
    Time: 1996-1997
    Results: Mgr Kaczmarek left because of health problems
     
  7. Title: Approximations to the recurrent biologically motivated networks.

  8. Objective: introduce stochastic dynamics to the adaptive vector quantization method, try to obtain the same dynamics in the feature space containing the input data clusters as in the recurrent network or other dynamical systems; use simple Chua oscillators as an example.
    Participants: W. Duch, MSc students
    Time: 1995-1997
  9. Title: Neural networks and genetic algorithms in time series analysis.

  10. Objective: compare neural models and genetic optimization with standard statistical methods used in econometrics; make general comparison between statistical and neural models. Use fuzzy function approximation capabilities of FSM, investigate the influence of the number of autoregressive variables on accuracy of predictions.
    Participants: W. Duch, P. Iwinski (PhD project)
    Time: 1995-1998
  11. Title: Dynamical neural models of neurophysiological phenomena.

  12. Objective: Use simplified models of neurons with biologically realistic properties to model simple ganglia. Negotiations on the scope of this project with the neurophysiology group are under way.
    Participants: W. Duch, P. Skudlarski (PhD project)
    Time: 1995-99
  13. Title: Neural networks and Hidden Markov Models (HMM) for RNA and protein structure prediction.

  14. Objective: compare recent results obtained with HMM programs with results obtained using neural networks. Implement the HMM programs on a PC.
    Participants: W. Duch, T. Rohn (MSc project)
    Time: 1995-1996
     
  15. Title: Neurofuzzy systems.

  16. Objective: Develop a universal neurofuzzy system capable of neural-like adaptation and fuzzy expert system reasoning. The Feature Space Mapping (FSM) system software is being written and integrates results from other projects on new learning procedures and on self-organization. FSM is modular, with each module specializing in a description of particular type of feature space. Applications include classification problems (molecular and astronomical spectra, psychograms), approximation problems (fuzzy function approximation) and modeling of intuitive behavior.
    Participants: R. Adamczak, W. Duch, N. Jankowski, A. Naud
    Time: 1994-1997
    Results: Duch W (1994) Floating Gaussian Mapping: a new model of adaptive systems Neural Network World 4:645--654; Duch W and Diercksen GHF (1995) Feature Space Mapping as a universal adaptive system Comp.Phys.Comm. 87: 341-371; Duch W, Jankowski N, Naud A, Adamczak R (1995) Feature Space Mapping: a neurofuzzy network for system identification in: Engineering Applications of Neural Networks, Helsinki 21-23.08.1995
     
  17. Title: Local adaptive coordinate systems.

  18. Objective: replace functional approach using neural networks by a formalism based on differential geometry describing the data by a set of local coordinate systems with adaptive parameters tuned to the incoming data. Formulate learning theory and show equivalence between such systems and neural models.
    Participants: W. Duch and M. Buczkowski (MSc project)
    Time: 1994-97
    Results: in preparation.
     
  19. Title: New learning algorithms for adaptive systems.

  20. Objective: Development of a new learning algorithms for adaptive systems, especially those based on localized learning. Generalization of the radial basis function algorithm to biradial functions and non-radial localized functions, development and testing of learning algorithms for such functions. Applications in classification and approximation problems.
    Participants: R. Adamczak, W. Duch, N. Jankowski
    Time: 1994-1997
    Results: Duch W, Jankowski N, Naud A, Adamczak R (1995) Feature Space Mapping: a neurofuzzy network for system identification, in: Engineering Applications of Neural Networks, Helsinki 21-23.08.1995; Duch W (1993) UMK-KMK-TR 6/93 report On the optimal processing functions for neural network elements; papers in preparation.
     
  21. Title: Self-organization via multidimensional scaling.

  22. Objective: Replace Kohonen's SOM procedure with global minimization that guarantees the best mapping of highly dimensional data on a low dimensional target space. Compare results with original SOM and use it for classification and visualization of psychometric data.
    Participants: W. Duch, A. Naud (PhD project)
    Time: 1994-96
    Results: Duch W (1995) Quantitative measures for the self-organized topographical mapping, Open Systems and Information Dynamics 2:295--302
     
  23. Title: Models of mind.

  24. Objective: define series of approximations leading from neurophysiological events related to perception and described using neural models and dynamical systems, to mind events, described in psychological spaces. Work out the theory of psychological spaces and apply it to construct new neurofuzzy systems. Show that such models of mind are philosophically sound and that consciousness may be included as a bonus.
    Participants: W. Duch
    Time: 1994-97
    Results: book in preparation, Duch W (1993) UMK-KMK-TR 3/93 report Floating Gaussian Mapping for Modeling of Human Conceptual Space; Duch W (1994) UMK-KMK-TR 1/94 report A solution to the fundamental problems of cognitive sciences; Duch W (1994) Mind space approach to neurofuzzy systems, in: Proc. of the Japanese Neural Networks Society 1994, Tsukuba, 9-11.11.1994,Japan, pp. 173-174; Duch W (1994) Towards Artificial Minds, in: ``First national conference on neural networks and their applications, Kule, 12-15.IV.1994, pp. 17--28; Duch W (1995) From cognitive models to neurofuzzy systems, in: Proc. from ``System, Modelling, Control''. Zakopane, 1-5.05.1995; Duch W (1995) Physics of consciousness. IV national conference on ``Modeling biological systems'', Krakow 2-3.06.1995, pp. 101-114; Duch W (1995) Psychophysics, in: European Summer School on Computing Techniques in Physics, Skalsky Dvur, 5-14.09.1995 (Lecture Notes); Duch W, (1995) Transparent theory of consciousness: is there a problem? Behavioral and Brain Sciences (submitted)
     
  25. Title: Applications of neural classifiers in psychometrics.

  26. Objective: Compare neural-based methods, such as SOM and FSM with statistical classificators for psychometric data, create a fuzzy expert system for personality type diagnosis.
    Participants: W. Duch, J. Gomula, A. Naud, T. Kucharski
    Time: 1994-96
    Results: nice software for visualization of data, SOM and MDS programs adapted to the job, interesting results reported on various psychiatric and psychological conferences.
     
  27. Title: Automatic Theorem Proving.

  28. Objective: Mechanization of mathematical proofs using the system ISABEL.
    Participants: K. Grabczewski, L. Paulson (Cambridge University)
    Time: 1994-1995
    Results: papers in preparation, mechanization of theorems from a whole book finished.
     
  29. Title: Application of algorithmic complexity in machine learning.

  30. Objective: Create universal learning procedure that generalizes in completely unbiased way creating simplest possible programs to solve the problem. Programs with the lowest Levin's complexity are selected.
    Participants: N. Jankowski, J. Schmidhuber (TUM Munich)
    Time: 1994 (Tempus project)
    Results: N. Jankowski, Application of Levin's optimal search algorithm, in: Proc. from ``System, Modelling, Control''. Zakopane, 1-5.05.1995; Jankowski N and Schmidhuber J (in preparation).
     
  31. Title: Application of Kohonen's SOM to texture segmentation

  32. Objective: Use self-organizing properties of feature maps to define areas in image that have different texture.
    Participants: A. Naud, N. Bonnet (INSERM Reims)
    Time: 1994 (Tempus project)
    Results: Naud A, Proc. from ``System, Modelling, Control''. Zakopane, 1-5.05.1995 Application of Kohonen's Self-Organizing Maps to texture image segmentation
     
  33. Tile: Complex systems theory.

  34. Objective: research in complex systems, definition of information and graph-based complexity, applications to molecular biology and computer linguistics.
    Participants: W. Duch, E. Jankowska and A. Korcala
    Time: 1993-97
    Results: Duch W (1993) UMK-KMK-TR 1/93 report Towards statistical theory of complex systems; Duch W and Jankowski N (1994) UMK-KMK-TR 5/93 report and ``First national conference on neural networks and their applications, Kule, 12-15.IV.1994, pp. 224--230 Complex systems, Information Theory and Neural Networks. more papers in preparation, some software was written.
     
  35. Title: Applications of neural networks in physics and chemistry.

  36. Objective: To understand what neural networks are good for and what kind of results one may expect.
    Participants: W. Duch and G.H.F. Diercksen (Max Planck Inst. Astrophysics, Garching b. Munich, Germany)
    Time: 1993-94
    Results: Duch W (1993) Modeling neural networks - a physicist's point of view, in: Psychological and neurophysiological backgrounds of new computer technologies, Eds. Kulikowski J.L and Zmyslowski W (International Center of Biocybernetics, Warsaw 1993); Duch W (1994) UMK-KMK-TR 4/93 report Neural networks for approximation; Duch W (1994) Neural networks for approximation, in ``First national conference on neural networks and their applications, Kule, 12-15.IV.1994, pp. 218-223; Duch W and Diercksen GHF (1994) Neural networks as tools to solve problems in physics and chemistry, Comp.Phys.Comm. 82: 91-103
     
  37. Title: Neurolinguistics.

  38. Objective: Look at linguistics from the connectionist point of view, find appropriate associative memory model for distributed representation of lexica, use statistical theory (especially HMM) and simplified neural network models, correlation matrix approach, check usefulness for the intelligent data bases (automatic associations to the queries, search in very large textual bases and natural language interfaces).
    Participants: W. Duch, E. Jankowska, A. Korcala
    Time: 1992-1994
    Results: some software has been written, reports: Duch W (1993) UMK-KMK-TR 2/93 report Metody komputerowe w leksykografii; Mgr Jankowska left with her husband for France.
     
  39. Title: Harmony Theory.

  40. Objective: Development of a general theory of cognitive systems and of a program for harmony theory - a model of a cognitive system - working on parallel machines. Testing and developing simple models to check scalability of this approach; 1991/92 new harmony program was written.
    Participants: L. Cierniak, W. Duch
    Time: 1991-1992
    Results: some software was written, some reports, in general harmony theory and Boltzman machines are slow. Mgr Cierniak left for a commercial company before finishing this project.

Wlodzislaw Duch, May 1997