Many researchers interested in connectionist models accept that such models are "neurally inspired " but do not worry too much about whether their models are biologically realistic. While such a position may be perfectly justifiable, the present paper attempts to illustrate how biological information can be used to constrain connectionist models. Two particular areas are discussed. The first section deals with visual information processing in the primate and human visual system. It is argued that speed with which visual information is processed imposes major constraints on the architecture and operation of the visual system. In particular, it seems that a great deal of processing must depend on a single bottum-up pass. The second section deals with biological aspects of learning algorithms. It is argued that although there is good evidence for certain coactivation related synaptic modification schemes, other learning mechanisms, including back-propagation, are not currently supported by experimental data.

monkey To investigate the functions of the amygdala in visual information processing and in emotional and social responses, recordings were made from single neurons in the amygdala of the monkey. A population of neurons (40 of more than 1000 recorded in 4 monkeys) was investigated which responded primarily to faces. These neurons typically (1) responded to some human or monkey faces, which were presented to the monkey through a large aperture shutter so that response latencies could be measured, or were simply shown to the monkey, (2) responded to 2-dimensional representations of these faces, as well as to real 3-dimensional faces, (3) had no responses or only small (less than half maximum) responses to gratings, simple geometrical, other complex 3-D stimuli, or to arousing and aversive stimuli, (4) had response latencies of 110-200 ms, (5) were located in the basal accessory nucleus of the amygdala, (6) responded differently to different faces, as shown by measures ofd', and could thus over a population of such neurons code information useful for making different responses to different individuals, (7)could in some cases (9/11 tested) respond to parts of faces, and (8) in a few cases (4/19 tested) responded more to a face which produced an emotional response. A comparison made in three monkeys of the responses of these neurons with the responses of 77 neurons with face-selective responses recorded in the cortex of the superior temporal sulcus (STS) showed that the amygdaloid neurons had longer response latencies (110-200 compared to 90-140 ms), and were in some respects more selective in their responses to different faces. It is suggested that the deficits in social and emotional behavior produced by amygdala lesions could be due in part to damage to a neuronal system specialized in utilizing information from faces so that appropriate social and emotional responses can be made to different individuals.

In order to gather evidence on functional subdivisions of the temporal lobe neocortex of the primate, the activity of more than 2600 single neurons was recorded in 10 myelo- and cytoarchitecturally defined subdivisions of the cortex in the superior temporal sulcus (STS) and inferior temporal gyrus of the anterior part of the temporal lobe of 5 hemispheres of 3 macaque monkeys. First, convergence of different mo-dalities into each area was investigated. Areas TS and TAa, in the upper part of this region, were found to receive visual as well as auditory inputs. Areas TPO, PGa, and IPa, in the depths of the STS, received visual, auditory, and somato-sensory inputs. Areas TEa, TEm, TE3, TE2, and TE 1, which extend from the ventral bank of the STS through the inferior temporal gyrus, were primarily unimodal visual areas. Sec-ond, of the cells with visual responses, it was found that

Certain basal forebrain neurons encode the learned rein-forcement value of objects: they respond differentially to visual stimuli that signal availability of fruit juice (positively reinforcing) or saline (negatively reinforcing) obtained by lick responses in visual discrimination tasks. In this report we describe the rapid, learning-related changes in the re-sponses of these neurons during the acquisition and reversal of the reinforcement contingency of a visual discrimination reversal task. The same neurons also responded differen-tially to novel and familiar stimuli in 2 recognition memory tasks, in which monkeys applied the learned rule that lick responses to novel stimuli elicited saline and responses to familiar stimuli elicited juice. These differential responses to novel and familiar stimuli thus reflected the reinforcement value of the stimuli. A single presentation of a novel or a

Animals are motivated to choose environmental options that can best satisfy current needs. To explain such choices, this paper introduces the MOTIVATOR (Matching Objects To Internal VAlues Triggers Option Revaluations) neural model. MOTIVATOR describes cognitiveemotional interactions between higher-order sensory cortices and an evaluative neuraxis composed of the hypothalamus, amygdala, and orbitofrontal cortex. Given a conditioned stimulus (CS), the model amygdala and lateral hypothalamus interact to calculate the expected current value of the subjective outcome that the CS predicts, constrained by the current state of deprivation or satiation. The amygdala relays the expected value information to orbitofrontal cells that receive inputs from anterior inferotemporal cells, and medial orbitofrontal cells that receive inputs from rhinal cortex. The activations of these orbitofrontal cells code the subjective values of objects. These values guide behavioral choices. The model basal ganglia detect errors in CS-specific predictions of the value and timing of rewards. Excitatory inputs from the pedunculopontine nucleus interact with timed inhibitory inputs from model striosomes in the ventral striatum to regulate dopamine burst and dip responses from cells in the substantia nigra