An experimental study of some algorithms for unsupervised learning.

Three well-known algorithms for unsupervised learning using a decision-directed approach are the random labeling of patterns according to the estimated a posteriori probabilities, the classification according to the estimated a posteriori probabilities, and the iterative solution of the maximum likelihood equations. The convergence properties of these algorithms are studied by using a sample of about 10 000 handwritten numerals. It turns out that the iterative solution of the maximum likelihood equations has the best properties among the three approaches. However, even this one fails to yield satisfactory results if the number of unknown parameters becomes large, as is usually the case in realistic problems of pattern recognition.

A knowledge based system for analysis of gated blood pool studies.

A system for obtaining a complete diagnostic description of an image sequence taken in nuclear medicine from the human heart has been developed, implemented, and tested. The knowledge about these images is represented in a semantic net, conclusions are drawn by a production rule approach, and scoring of alternative diagnoses is based on fuzzy membership functions. On the low level, image pixels are smoothed and organ contours are extracted; these are the input for the high level processing. Tests with several image sequences gave correct descriptions as compared to the diagnosis of a physician.

An architecture for distributed visual memory.

The development of autonomous as well as situated robots is one of the great remaining challenges and involves a number of different scientific disciplines. In spite of recent dramatic progress, it remains worthwhile to examine natural systems, because their abilities are still out of reach. Motivated by research work done in the fields of cognitive systems, visual perception, and psychology of memory we designed and implemented a memory architecture for visual tasks. Structural and functional concepts of the memory architecture were modeled on the ones found in natural systems. We present an efficient implementation based on parallel programming techniques. The memory module is integrated into a distributed system for speech and image analysis, which is currently developed in the Sonderforschungsbereich (SFB) 360, Situated Artificial Communicators, where a hybrid vision system combining neural and semantic networks is used.

Estimation and filtering of potential protein-protein docking positions.

MOTIVATION: Software systems predicting automatically whether and how two proteins may interact are highly desirable, both for understanding biological processes and for the rational design of new proteins. As a part of a future complete solution to this problem, a bundle of programs is presented designed (i) to estimate initial docking positions for a given pair of docking candidates, (ii) to adjust them, and (iii) to filter them, thus preparing more detailed computations of free energies. RESULTS: The system is evaluated on a test set of 51 co-crystallized complexes aiming at redocking the subunits. It works completely automatically and the evaluation is performed using one single set of parameters for all complexes in the test set. The number of solutions is fixed to 50 positions with a median CPU time of 26 min. For 30 complexes, these contain a near-correct solution with root mean square deviation ( RMSD )

Protein docking combining symbolic descriptions of molecular surfaces and grid-based scoring functions.

With the growing number of known 3D protein structures, computing systems, that can predict where two protein molecules interact with each other is becoming of increasing interest. A system is presented, integrating preprocessing like the computation of molecular surfaces, segmentation, and searching for complementarity in the general framework of a pattern analyzing semantic network (ERNEST). The score of coarse symbolic computations is used by the problem independent control strategy of ERNEST to guide a more detailed analysis considering steric clash and judgements based on grid-based surface representations. Successful examples of the docking system are discussed that compare well with other approaches.