Jerome A. Feldman

Connectionist Models and Their Properties

Much of the progress in the fields constituting cognitive science has been based upon the use of explicit information processing models, almost exclusively patterned after conventional serial computers. An extension of these ideas to massively parallel, connectionist models appears to offer a number of advantages. After a preliminary discussion, this paper introduces a general connectionist model and considers how it might be used in cognitive science. Among the issues addressed are: stability and noise-sensitivity, distributed decision-making, time and sequence problems, and the representation of complex concepts.

Four frames suffice: A provisional model of vision and space

This paper presents a general computational treatment of how mammals are able to deal with visual objects and environments. The model tries to cover the entire range from behavior and phenomenological experience to detailed neural encodings in crude but computationally plausible reductive steps. The problems addressed include perceptual constancies, eye movements and the stable visual world, object descriptions, perceptual generalizations, and the representation of extrapersonal space. The entire development is based on an action-oriented notion of perception. The observer is assumed to be continuously sampling the ambient light for information of current value. The central problem of vision is taken to be categorizing and locating objects in the environment. The critical step in this process is the linking of visual information to symbolic object descriptions; this is called indexing , from the analogy of identifying a book from index terms. The system must also identify situations and use this knowledge to guide movement and other actions in the environment. The treatment focuses on the different representations of information used in the visual system. The four representational frames capture information in the following forms: retinotopic, head-based, symbolic, and allocentric. The functional roles of the four frames, the communication among them, and their suggested neurophysiological realization constitute the core of the paper. The model is perforce crude, but appears to be consistent with all relevant findings.

On the Synthesis of Finite-State Machines from Samples of Their Behavior

The Nerode realization technique for synthesizing finite-state machines from their associated right-invariant equivalence relations is modified to give a method for synthesizing machines from finite subsets of their input-output behavior. The synthesis procedure includes a parameter that one may adjust to obtain machines that represent the desired behavior with varying degrees of accuracy and that consequently have varying complexities. We discuss some of the uses of the method, including an application to a sequential learning problem.

Embodied meaning in a neural theory of language.

Imagine that you are dining with a group of people and one of them asks you to pass the salt. Ignoring for now any social or emotional issues, what does it mean for you to carry out the request? You must locate the salt container, determine which grip to use, reach out and grasp the container, and move it to a location near the requester. If the salt is too far for you to easily reach, you should convey the request to someone who can pass it to you, possibly following a further request. Now think about your favorite theory of language and what it could describe of this scenario. In this paper, we outline an explicitly neural theory of language (NTL) that attempts to explain how many brain functions (including emotion and social cognition) work together to understand and learn language. The focus will be on the required representations and computations, although there will be some discussion of results on specific brain structures. In this approach, one does not expect to find brain areas specialized only for language or to find language processing confined to only a few areas. Our first sentence asked you to imagine yourself at some dinner, being asked to pass the salt. To understand this sentence you need to know what it means to pass the salt. The NTL assumption is that people understand narratives by subconsciously imaging (or simulating) the situation being described. When asked to grasp, we enact it. When hearing or reading about grasping, we simulate grasping, being grasped, or watching someone grasp. We will start with the meaning of words for simple actions like those involved in passing the salt. Consider the word ‘‘grasp’’. Everyone will agree that the meaning of the word grasp involves the motor action of grasping

High level programming for distributed computing

Programming for distributed and other loosely coupled systems is a problem of growing interest. This paper describes an approach to distributed computing at the level of general purpose programming languages. Based on primitive notions of module, message, and transaction key, the methodology is shown to be independent of particular languages and machines. It appears to be useful for programming a wide range of tasks. This is part of an ambitious program of development in advanced programming languages, and relations with other aspects of the project are also discussed.

Translator writing systems

A critical review of recent efforts to automate the writing of translators of programming languages is presented. The formal study of syntax and its application to translator writing are discussed in Section II. Various approaches to automating the postsyntactic (semantic) aspects of translator writing are discussed in Section III, and several related topics in Section IV.

Some decidability results on grammatical inference and complexity

The problem of grammatical inference is considered and a number of positive answers to decidability questions obtained. Conditions are prescribed under which it is possible for a machine to infer a grammar (or the best grammar) for even the general rewriting systems.

An ALGOL-based associative language

A high level programming language for large, complex associative structures has been designed and implemented. The underlying data structure has been implemented using a hash-coding technique. The discussion includes a comparison with other work and examples of applications of the language.

Integration of Form across Saccadic Eye Movements

To perceive a stable world, one must somehow be able to relate visual information from successive fixations. Little is known, however, about the nature of the integrative process. By using a task which requires the integration of spatial position information from different fixations, it is demonstrated that visual information from previous fixations is preserved in a world-centered representation which is precise enough to support judgements of geometric shape. It is also shown that successive views are aligned with respect to common visual features, indicating that visual stability may be normally accomplished by a visual matching strategy in combination with cancellation by an eye-position signal.

Decision theory and artificial intelligence: I. A semantics-based region analyzer

Abstract Mathematical decision theory can be combined with heuristic techniques to attack Artificial Intelligence problems. As a first example, the problem of breaking an image into meaningful regions is considered. Bayesian decision theory is seen to provide a mechanism for including problem dependent (semantic) information in a general system. Some results are presented which make the computation feasible. A programming system based on these ideas and its application to road scenes is described.