Connectionist neuropsychology: the breakdown and recovery of behavior in lesioned attractor networks

Recent success in modeling the cognitive deficits of brain-injured patients by "lesioning" a connectionist model of the normal process supports the claim that connectionist networks can provide insight into the neural implementation of cognitive processes. However, there is little understanding of what underlying principles are responsible for the results. This thesis investigates the effects of damage in networks that make familiar patterns of activity into stable "attractors", in order to identify the computational principles that enable them to reproduce specific neuropsychological phenomena. In a series of simulations, networks are trained to pronounce written words via a simplified representation of their semantics. Under damage, the networks produce a distribution of visual and semantic errors similar to that of patients with "deep dyslexia." Further simulations replicate other characteristics of these patients: additional error types, better performance on concrete vs. abstract words, preserved lexical decision, and greater confidence in visual vs. semantic errors. A range of network architectures and learning procedures produce qualitatively similar results, demonstrating that the layout of attractors giving rise to the error pattern depends more on the unstructured nature of the task than on the architectural details that enable the attractors to develop. Additional simulations address issues in relearning after damage: the speed of recovery, degree of generalization, and strategies for optimizing recovery. Relative differences in the degree of generalization for different lesion locations is due to the amount of structure in the subtasks performed by parts of the network. In the related domain of object recognition, a network augmented with short-term correlational weights is trained to generate semantic representations of objects from high-level visual representations. Under damage, the network exhibits the complex semantic and perseverative effects of patients with a visual naming disorder known as "optic aphasia." The greater structure in mapping visual to semantic representations for object vs. words explains why the errors of optic aphasics are predominantly semantic rather than visual. Taken together, the results of the thesis demonstrate that the breakdown and recovery of behavior in lesioned attractor networks reproduces specific neuropsychological phenomena by virtue of the way the structure of a task shapes the layout of attractors.

[1]  Carsten Peterson,et al.  A Mean Field Theory Learning Algorithm for Neural Networks , 1987, Complex Syst..

[2]  G. Kane.  Parallel Distributed Processing: Explorations in the Microstructure of Cognition, vol 1: Foundations, vol 2: Psychological and Biological Models , 1994 .

[3]  Geoffrey E. Hinton,et al.  Lesioning an attractor network: investigations of acquired dyslexia , 1991 .

[4]  M. Mozer,et al.  On the Interaction of Selective Attention and Lexical Knowledge: A Connectionist Account of Neglect Dyslexia , 1990, Journal of Cognitive Neuroscience.

[5]  James L. McClelland,et al.  Connections and disconnections: Acquired dyslexia in a computational model of reading processes. , 1989 .

[6]  James L. McClelland,et al.  A computational model of semantic memory impairment: modality specificity and emergent category specificity. , 1991, Journal of experimental psychology. General.

[7]  Geoffrey E. Hinton,et al.  Learning and relearning in Boltzmann machines , 1986 .

[8]  James L. McClelland,et al.  Parallel distributed processing: explorations in the microstructure of cognition, vol. 1: foundations , 1986 .

[9]  Sally Byng,et al.  Computer assisted remediation of a homophone comprehension disorder in surface dyslexia , 1989 .

[10]  James L. McClelland.  Stochastic interactive processes and the effect of context on perception , 1991, Cognitive Psychology.

[11]  John J. L. Morton,et al.  Interaction of information in word recognition. , 1969 .

[12]  Jerome A. Feldman,et al.  Connectionist Models and Their Properties , 1982, Cogn. Sci..

[13]  F. Lhermitte,et al.  A visual-speech disconnexion syndrome. Report of a case with optic aphasia, agnosic alexia and colour agnosia. , 1973, Brain : a journal of neurology.

[14]  James L. McClelland,et al.  Parallel Distributed Processing: Explorations in the Microstructure of Cognition : Psychological and Biological Models , 1986 .

[15]  A. Caramazza,et al.  The structure of orthographic representations in spelling , 1990 .

[16]  Sally Byng,et al.  A treatment for surface dyslexia , 1989 .

[17]  E K Warrington,et al.  Concrete word dyslexia. , 1981, British journal of psychology.

[18]  James L. McClelland,et al.  An interactive activation model of context effects in letter perception: I. An account of basic findings. , 1981 .

[19]  Eric L. Schwartz,et al.  Computational Neuroscience , 1993, Neuromethods.

[20]  Geoffrey E. Hinton.  Deterministic Boltzmann Learning Performs Steepest Descent in Weight-Space , 1989, Neural Computation.

[21]  Geoffrey E. Hinton.  Using fast weights to deblur old memories , 1987 .

[22]  Alfonso Caramazza,et al.  The logic of neuropsychological research and the problem of patient classification in aphasia , 1984, Brain and Language.

[23]  Marlene Behrmann,et al.  The rites of righting writing: Homophone remediation in acquired dysgraphia , 1987 .

[24]  Gregory V. Jones.  Deep dyslexia, imageability, and ease of predication , 1985, Brain and Language.

[25]  James L. McClelland,et al.  An interactive activation model of context effects in letter perception: part 1.: an account of basic findings , 1988 .