ACCESSION NO: 1998-95022-079
       DOCUMENT TYPE: Dissertation Abstract
              AUTHOR: Kralik, Jerald Daniel
        AFFILIATION:  Harvard U, USA
               TITLE: From behavioral learning to neural learning: Computational 
                      modeling and empirical studies of nonhuman animals.
                YEAR: 1998
              SOURCE: Dissertation Abstracts International: Section B: The Sciences 
                      & Engineering, Vol 59(5-B), Nov 1998, 2447. 
           ISSN/ISBN: 0419-4217
       UMI ORDER NO.: AAM9832420
            LANGUAGE: English
            ABSTRACT: The thesis overall attempted to apply behavioral learning 
                      theories and experimental results to other important areas of 
                      learning, including computational learning theory and neural 
                      plasticity. In the introductory chapter, general 
                      computational theories of learning are considered. It is 
                      argued that current computational learning theories remain 
                      incomplete in their description of learning by biological 
                      organisms. In Chapter 2, an adaptive network model of 
                      category learning is developed. Two main modifications are 
                      proposed to the Rescorla-Wagner (1972) model as a learning 
                      rule. One modification is that the change of association 
                      strength is augmented if a cue is a better than average 
                      predictor of the target category, whereas the change of 
                      association strength is depressed if a cue is a below average 
                      predictor. The second main modification is based on the idea 
                      that absent cues still compete against existing cues if the 
                      absent cues are associated with the same target category as 
                      the existing cue, or if the absent cue is correlated with the 
                      existing cue. In Chapter 3, a model of a basic neural 
                      learning unit is developed using classical conditioning 
                      behavioral results as a benchmark. It is argued that 
                      conditioned response characteristics need to be taken into 
                      account during classical conditioning learning, because 
                      ultimately the goal of learning is to have optimal 
                      interactions with reinforcers. The model that is developed 
                      accounts for numerous classical conditioning phenomena such 
                      as blocking and overshadowing, and suggests an account for 
                      the variation between interstimulus interval curves. Chapter 
                      4 reports an experiment in which a nonhuman primate was 
                      tested on the abstract relation connectedness. Results show 
                      that the subjects did respond to the relation between 
                      objects; and the subjects generalized from the training 
                      images to novel images. The data also suggest a specific 
                      learning sequence for the acquisition of the connectedness 
                      concept. In the concluding chapter, Chapter 5, the learning 
                      sequence from the Connectedness Experiment is used to show 
                      that current computational theory needs a more complete 
                      theory of generalization. It is believed that theories and 
                      experiments in behavioral learning will continue to enlighten 
                      theories of computational learning as well as models of 
                      neural plasticity. ((c) 1999 APA/PsycINFO, all rights 
                      reserved):
          KEY PHRASE: From behavioral learning to neural learning: Computational 
                      modeling and empirical studies of nonhuman animals (classical 
                      conditioning)
          MAJOR DESC: Classical Conditioning; Learning Theory; Neural Plasticity
          MINOR DESC: Primates (Nonhuman)
         CLASS. CODE: 2340 Cognitive Processes; 2100 General Psychology
          POPULATION: Animal
        FORM/CONTENT: Empirical Study
        RELEASE DATE: 19990901