Bareiss / Chandrasekaran Exemplar-Based Knowledge Acquisition

A Unified Approach to Concept Representation, Classification, and Learning
1. Auflage 2014
ISBN: 978-1-4832-1637-9
Verlag: Elsevier Science & Techn.
Format: PDF
Kopierschutz: 1 - PDF Watermark

E-Book, Englisch, 184 Seiten, Web PDF

Exemplar-Based Knowledge Acquisition: A Unified Approach to Concept Representation, Classification, and Learning covers the fundamental issues in cognitive science and the technology for solving real problems. This text contains six chapters and begins with a description of the rationale for the design of Protos Approach, its construction and performance. The succeeding chapters discuss how the Protos approach meets the requirements of representing concepts, using them for classification, and acquiring them from available training. These chapters also deal with the design and implementation of Protos. These topics are followed by a presentation of examples of the application of Protos to audiology and evaluate its performance. The final chapters survey related work in the areas of case-based reasoning and automated knowledge acquisition and the contributions of Protos approach. This book will be of great value to psychologists, psychiatrists, and researchers in the field of artificial intelligence.

Autoren/Hrsg.

Bareiss, Ray

Chandrasekaran, B.

Inhaltsverzeichnis

1;Front Cover;1
2;The Psychology of Learning and Motivation, Volume 30;4
3;Copyright Page;5
4;Contents;6
5;Contributors;10
6;Chapter 1. Perceptual Learing;12
6.1;I. Introduction;12
6.2;II. Classification of Learning Processes;13
6.3;III. Perceptual Learning: Beginning Anew;16
6.4;IV. Paradox of Perceptual Learning;17
6.5;V. Prism Adaptation: New Variants on a Classic Paradigm;19
6.6;VI. Hierarchy of Transformation Geometry;40
6.7;VII. Computer Mappings;45
6.8;VIII. The McCollough Effect;56
6.9;IX. Conclusion;65
6.10;References;68
7;Chapter 2. A Rational- Constructivist Account of Early Leaning About Numbers and Objects;72
7.1;I. Introduction;72
7.2;II. Different Accounts of Initial Concepts;74
7.3;III. On Variability;83
7.4;IV. Conclusions;102
7.5;References;104
8;Chapter 3. Remembiring, Knowing, and Recontructing the Past;108
8.1;I. Introduction;108
8.2;II. Disparate Effects of Repeated Testing;111
8.3;III. Interference and Reconstruction;120
8.4;IV. Remembering and Knowing Past Events;133
8.5;V. Conclusions;140
8.6;References;141
9;Chapter 4. The Long- Term Retention of Knowledge and Skills;146
9.1;I. Introduction;146
9.2;II. Features of Our Research Program;147
9.3;III. Specificity of Training;148
9.4;IV. Guidelines for Improving Long-Term Retention;151
9.5;V. Summary and Conclusions;172
9.6;References;174
10;Chapter 5. A Comprehension-Based Approach to Learning and Understanding;176
10.1;I. Introduction;176
10.2;II. A Psychological Process Model of Discourse Comprehension;177
10.3;III. Text Structure: On the Page or in the Head?;181
10.4;IV. Diagnosing and Circumventing Comprehension Problems;188
10.5;V. The Role of Active Inferencing in the Construction of Text Meaning;196
10.6;VI. Learning from Text;201
10.7;VII. A Tutor for Word Algebra Problems;211
10.8;VIII. Conclusion;220
10.9;References;221
11;Chapter 6. Separating Causl Laws from Casual Facts: Pressing the Limits of Statistical Relevance;226
11.1;I. Introduction;226
11.2;II. Evaluating Criticisms of Statistical Accounts;232
11.3;III. Some Problems and Limits of Statistical Relevance;265
11.4;IV. Conclusion;271
11.5;References;272
12;Chapter 7. Categories, Hierarachies, and Induction;276
12.1;I. Introduction;276
12.2;II. The Nature of Categories;277
12.3;III. Entrenchment and Induction: The Goodman Position;282
12.4;IV. Categories, Entrenchment, and Induction: A Psychological Perspective;287
12.5;V. Prior Experimental Work on Induction;293
12.6;VI. Discussion;303
12.7;VII. Summary;307
12.8;References;309
13;Index;314
14;Contents of Recent Volumes;322

Leseproben

A Rational-Constructivist Account of Early Learning About Numbers and Objects

Rochel Gelman

I Introduction

This article features my rational-constructivist account of cognitive development. The rationalist side of the theory captures the assumption that our young bring a skeletal outline of domain-specific knowledge to their task of learning the initial concepts they will share with others. The constructivist side of the theory captures the assumption that, from the start, our young actively join in their own cognitive development. Even as beginning learners, skeletal principles motivate them to seek out and assimilate inputs that nurture the development of these structures. To develop these assumptions I consider work on two topics: (1) conceptions of objects during infancy and (2) numerical concepts in infants and beginning language users. Special attention is given to the need to consider whether differences in performance levels across tasks are due to limits on the conceptual competence under investigation or to limits on the procedural and interpretative competences needed for successful performance.

There is no a priori reason to assume that the rational and the constructivist positions are inconsistent or contradictory. I join Marler in his challenge of those who still “think of learning and instinct as being virtually antithetical … [that] behavior is one or the other, but not both” (p. 37, 1991). In the history of science, key terms shifted their meaning when understanding of the phenomena to which they refer changed. For example, developments in physics, mathematics, and biology led to changes in the meaning of movement, zero, and alive (see Carey, 1985; Kitcher, 1982; Kuhn, 1970; Mayer, 1982; and Wiser, 1987). Likewise, recent advances in neuroscience, animal learning, and ethology are producing shifts in the meaning of phrases or terms like biological underpinnings and innate. The more we understand about the acquisition of complex actions, the more we appreciate that they depend on organisms’ opportunities to interact with and assimilate relevant environments. To say that genetic history contributes to the development of some classes of behavior is not to say that these will appear full-blown at a given point in time. Without opportunities to engage with and learn about the kinds of environments that nurture the potential given by the genetic history, development either will be abnormal or will fail to occur. Normal development is intricately tied to opportunities to interact with and process relevant environments. The story of how the young male white-crowned sparrow comes to learn his species-specific adult song provides an elegant example of these points.

The male white-crowned sparrow is born with a template specifying basic features of his adult song. However, he must hear examples of the correct conspecific song during a critical period early in development. If he is reared in an environment that does not include examples of his adult dialect, he is able to learn a nonpreferred song. Therefore, the learning process that supports acquisition of the conspecific song can yield unexpected or inappropriate outcomes, given sufficient atypical experience. Nevertheless, “errors” seldom occur because learning typically takes place in a supporting ecology.

Similarly, the opportunity to interact with the environment supports another key step in song development. The initial song, which is first produced well after the above critical period, is far from the adult song. Production of the adult crystallized song is preceded by a lengthy trial-and- error period. Although the bird does not have to hear any further inputs from other birds during this period, it appears that he does have to hear himself produce what are called subsongs and plastic songs. It seems as if the remembered song provides birds with a standard against which to compare their output, much as memories of recordings or performances can aid music students as they practice.

Such examples help to illustrate how the meaning of terms and phrases like learn, innate, and biological contributions are changing. Indeed, Marler (1991) now writes of “the instinct to learn” and refuses to pit terms like practice, trial-and-error, variability, and learn against ones like constraint, innate, biological, genetic, and so on. Parallel shifts in meaning can be found in writings about animal learning (Gallistel, 1990; Rozin & Schull, 1988) and cognitive development (e.g., Carey & Gelman, 1991; Karmiloff-Smith, 1992; Keil, 1981). These developments serve as the backdrop for my rational–constructivist account of knowledge acquisition. I have been especially concerned with the specification of the nature of relevant inputs and the laws of learning that apply for such an account.

II Different Accounts of Initial Concepts

A ASSOCIATION THEORIES OF LEARNING

There have been important developments in associative accounts of learning, especially regarding the need for the conditioned stimulus (CS) to predict the unconditioned stimulus (UCS) (Rescorla & Wagner, 1972). Still, the empiricists’ assumptions about the acquisition of knowledge remain as core assumptions in modern associationist accounts of concept development and learning. These assumptions are that all knowledge can be traced to our ability to process sensory inputs and to form associations between these sensations (S-S connections) and/or to our responses to these sensations (S-R connections). In the case of the infant, what is given is the ability to receive punctate sensations of light, or sound, or pressure, and so forth, and to form associations between these according to the laws of association (frequency and proximity). Sensations and responses that occur close together (in time or space) and repeatedly are more likely to be associated than those that are infrequent and far apart. As associations between sensations and responses are impressed on the infant’s blank mental slate, these too become associated with incoming data or each other and lay the groundwork for knowledge of the world at a sensory and motor level. These further associations in turn support knowledge acquisition at the perceptual level. Experiences at the perceptual level provide the opportunity for cross-modal associative learning and thus for the eventual induction of abstract concepts that cut across concepts about particular perceptual information.

B DEVELOPMENTAL THEORIES OF LEARNING

Developmental textbooks often pit learning theoretic (read as “associationist”) accounts against developmental ones. In this context, the idea is that development involves more than “mere” learning. Cognitive development proceeds through stages, and the way learners interpret inputs of a given kind is influenced by the stage they have achieved. For example, during the first two years of life, Piaget’s sensorimotor infants can build schemes relating actions to what they see, hear, touch, and so on; however, they will not be able to represent a set of objects in terms of class inclusion until they reach concrete operations at about 6 to 9 years of age.

Paired with the assumption of stages is the related assumption that learners actively interpret inputs with reference to their available knowledge and mental structures. In the Piagetian framework, the construction of the “correct” interpretation of the transformations performed on quantities must await the child’s advance to concrete operational thought. The younger child’s belief that the amount of water in a glass changes as it is poured into another, different-shaped glass, reflects reliance on perceptual information (Piaget, 1952).

As we shall see, there are important differing foundational assumptions of the associationist and developmental accounts of infant-knowledge acquisition. Still, the two classes of characterizations of an infant’s initial world are more similar than not. For example, Piaget limits an infant’s initial knowledge to a level that is controlled reflexively. The active practice of these reflexes leads to their adaptation into sensorimotor schemes. The active use of the consequent scheme leads to the development of intercoordinated schemes of action. The more such intercoordinations, the more likely that the infant builds a world of three-dimensional objects in a three-dimensional space.

In the associationist account, infants gradually build up a notion of an individual object by associating the primitive sensations generated by different objects. Somehow, by forming associative clusters for many different objects, young learners eventually produce the concept of an object as something that exclusively occupies a volume of space at a particular time and that has properties such as color, shape, weight, and so on.

To be sure, associations are not Piaget’s fundamental building blocks of cognition; sensorimotor schemes are. Nevertheless, his infant must have repeated interactions with objects in order to achieve more coordinated memories among those sensorimotor schemes that are used with a given object. These action-based representations help move the infant from a state of out-of-sight, out-of-mind to states that lead to a concept of an object. Only then does the infant finally know that an object persists over time and in space, whether or not it is covered...