DIMACS Theory Lecture Series


A Computational Model for Cognition


Leslie G. Valiant
Harvard University


Lunch 11:45 AM, Room 402, Computer Science Building, 35 Olden Street
Talk will be held in the Small Auditorium, Room 105, 12 Noon
Princeton University


Lunch 11:45 AM, Talk 12 Noon
Friday, April 7, 1995


A model of computation is described that is designed to bridge the gap in biological systems between the neural level of computation and the level of cognitive functioning. A {\it neuroid} is a model neuron that is similar to a classical threshold element, but is augmented with states. The availability of states makes the model flexibly programmable and allows a variety of useful timing mechanisms to be exploited in computations. To specify a {\it neuroidal system} one needs to define the programs executed by each neuroid, the network that connects the neuroids together, as well as the interfaces of the system with any input, output or other auxiliary or peripheral devices. Given such a specification one can attempt to analyze the behavior of the system.

A primary goal is to study algorithms that implement idealizations of some of the most basic cognitive functions. An important constraint is that the number of steps taken by the algorithms, the number of neuroids needed, and the interconnectivity among them that is assumed, should all be biologically plausible. We study a class of functions, that we call {\it random access tasks,} that test these constraints most severely. These functions are characterized as those that, at each invocation, may potentially access any item already stored in memory. The class includes the tasks of memorizing a new item that is related to previously memorized ones, forming an association between two items, and inductive learning. More complex functions, such as those involving relational information, or those that formalize simple reasoning processes are also studied. It is emphasized that all these functions have to be supported compatibly in a single system, so that a long sequence of interactions with the world will result in the accumulation of competence by the system.

The material to be described is from a monograph {\it Circuits of the Mind,} Oxford University Press, November 1994.

Document last modified on March 28, 1995