Michael McCloskey

Catastrophic Interference in Connectionist Networks: The Sequential Learning Problem

Publisher Summary Connectionist networks in which information is stored in weights on connections among simple processing units have attracted considerable interest in cognitive science. Much of the interest centers around two characteristics of these networks. First, the weights on connections between units need not be prewired by the model builder but rather may be established through training in which items to be learned are presented repeatedly to the network and the connection weights are adjusted in small increments according to a learning algorithm. Second, the networks may represent information in a distributed fashion. This chapter discusses the catastrophic interference in connectionist networks. Distributed representations established through the application of learning algorithms have several properties that are claimed to be desirable from the standpoint of modeling human cognition. These properties include content-addressable memory and so-called automatic generalization in which a network trained on a set of items responds correctly to other untrained items within the same domain. New learning may interfere catastrophically with old learning when networks are trained sequentially. The analysis of the causes of interference implies that at least some interference will occur whenever new learning may alter weights involved in representing old learning, and the simulation results demonstrate only that interference is catastrophic in some specific networks.

Misleading Postevent Information and Memory for Events. Arguments and Evidence Against Memory Impairment Hypotheses

The claim that a persons memory for an event may be altered by information encountered after the event has been influential in shaping current conceptions of memory. The basis for the claim is a series of studies showing that subjects who are given false or misleading information about a previously witnessed event perform more poorly on tests of memory for the event than subjects who are not misled. In this article we argue that the available evidence does not imply that misleading postevent information impairs memory for the original event, because the procedure used in previous studies is inappropriate for assessing effects of misleading information on memory. We then introduce a more appropriate procedure and report six experiments using this procedure. We conclude from the results that misleading postevent information has no effect on memory for the original event. We then review several recent studies that seem to contradict this conclusion, showing that the studies do not pose problems for our position. Finally, we discuss the implications of our conclusions for broader issues concerning memory.

Cognitive mechanisms in numerical processing: Evidence from acquired dyscalculia

This article discusses cognitive neuropsychological research on acquired dyscalculia (i.e., impaired numerical processing resulting from brain damage), surveying issues of current interest, and illustrating the ways in which analyses of acquired deficits can contribute to an understanding of normal processing. I first review the logic whereby inferences concerning normal cognition are drawn from patterns of impaired performance. I then consider research exploring the general functional architecture of the cognitive numerical processing mechanisms, and finally turn to studies aimed at probing the internal structure and functioning of individual processing components.

Cognitive mechanisms in number processing and calculation: Evidence from dyscalculia

This article presents a framework for the cognitive analysis of number processing and calculation. Within this framework the primary objective is the development of a model that is sufficiently detailed to serve as a basis for explaining the number-processing/calculation performance of both normal and cognitively impaired subjects. First a general model of the cognitive mechanisms for number processing and calculation is outlined. It is shown that patterns of impairments observed in brain-damaged patients support the major assumptions of the model and that the model provides a theoretically motivated framework for interpreting the deficits. A single case is then discussed in some detail, to demonstrate that through detailed analyses of impaired performance the preliminary model can be elaborated to specify not only the general architecture of the number-processing and calculation systems, but also the inner workings of specific components and the consequences of damage to these components. The article concludes with a discussion of several general issues arising from the presented arguments.

Amygdala and orbitofrontal reactivity to social threat in individuals with impulsive aggression

BACKGROUND Converging evidence from animal and human lesion studies implicates the amygdala and orbitofrontal cortex (OFC) in emotional regulation and aggressive behavior. However, it remains unknown if functional deficits exist in these specific brain regions in clinical populations in which the cardinal symptom is impulsive aggression. We have previously shown that subjects diagnosed with intermittent explosive disorder (IED), a psychiatric disorder characterized by reactive aggressive behavior, perform poorly on facial emotion recognition tasks. In this study we employed a social-emotional probe of amygdala-OFC function in individuals with impulsive aggression. METHODS Ten unmedicated subjects with IED and 10 healthy, matched comparison subjects (HC) underwent functional magnetic resonance imaging while viewing blocks of emotionally salient faces. We compared amygdala and OFC reactivity to faces between IED and HC subjects, and examined the relationship between the extent of activation in these regions and extent of prior history of aggressive behavior. RESULTS Relative to controls, individuals with IED exhibited exaggerated amygdala reactivity and diminished OFC activation to faces expressing anger. Extent of amygdala and OFC activation to angry faces were differentially related to prior aggressive behavior across subjects. Unlike controls, aggressive subjects failed to demonstrate amygdala-OFC coupling during responses to angry faces. CONCLUSIONS These findings provide evidence of amygdala-OFC dysfunction in response to an ecologically-valid social threat signal (processing angry faces) in individuals with a history of impulsive aggressive behavior, and further substantiate a link between a dysfunctional cortico-limbic network and aggression.

Naive beliefs in “sophisticated” subjects: misconceptions about trajectories of objects

Abstract University students were asked to solve simple problems about the trajectories of falling objects. A majority of the students revealed a variety of misconceptions about motion. However, the few basic patterns of responses produced by the subjects suggest considerable commonality in the types of naive physical “laws” people develop on the basis of everyday experience with the world.

Cognitive processes in verbal-number production: inferences from the performance of brain-damaged subjects.

This article presents a model of the cognitive processes involved in the spoken production of verbal numbers (e.g., thirteen thousand four hundred two). On the basis of single-case studies of two brain-damaged subjects with number production deficits, we argue that verbal-number production involves the generation of a syntactic frame that constitutes a plan for the production of the appropriate sequence of words. The syntactic frame specifies each to-be-retrieved word in terms of a number-lexical class (i.e., ones, teens, or tens) and a position within that class. These class/position-within-class specifications guide the retrieval of lexical representations from a production lexicon that is partitioned into functionally distinct ones, teens, and tens classes. We conclude with a brief discussion of the rationale for, and advantages of, using patterns of impaired performance as a basis for drawing inferences about normal cognition.

The case for single-patient studies

Abstract It is argued that only single-patient studies allow valid inferences about normal cognitive processes from the analysis of acquired cognitive disorders. The arguments in favour of this position support the further claim that clinical classifications such as agrammatism and deep dyslexia are theoretically useless.

Networks and Theories: The Place of Connectionism in Cognitive Science

This article considers how connectionist modeling can contribute to understanding of human cognition. J argue that connectionist networks should not be thought of as theories or simulations of theories, but may nevertheless contribute to the development of theories.

Naive Physics: The Curvilinear Impetus Principle and Its Role in Interactions with Moving Objects.

Several recent studies in which subjects solved pencil-and-paper problems concerning the behavior of moving objects have shown that many people have incorrect beliefs about motion. The present study considers the question of whether these naive beliefs are manifested in situations where people observe and interact with moving objects. Several findings in the problem-solving literature suggest that abstract or unrealistic tasks may fail to tap knowledge and reasoning abilities that are routinely used in more concrete or realistic situations. Thus, most people may have accurate knowledge about the behavior of moving objects, knowledge that they use in their everyday interactions with objects in motion. However, this knowledge may not be activated in the context of abstract, static problems, and as a result people attempting to solve such problems may resort to naive beliefs. Three experiments examine this possibility in the context of one specific naive belief, the curvilinear impetus belief. Contrary to expectations, results suggest that the curvilinear impetus belief is used not only on pencil-and-paper problems but also in situations where people observe and interact with moving objects. Implications of these findings are discussed.