OBJECT RECOGNITION AND SENSITIVE PERIODS - A COMPUTATIONAL ANALYSIS OF VISUAL IMPRINTING

Using neural and behavioral constraints from a relatively simple biolo gical visual system, we evaluate the mechanism and behavioral implicat ions of a model of invariant object recognition. Evidence from a varie ty of methods suggests that a localized portion of the domestic chick brain, the intermediate and medial hyperstriatum ventrale (IMHV), is c ritical for object recognition. We have developed a neural network mod el of translation-invariant object recognition that incorporates featu res of the neural circuitry of IMHV, and exhibits behavior qualitative ly similar to a range of findings in the filial imprinting paradigm, W e derive several counter-intuitive behavioral predictions that depend critically upon the biologically derived features of the model. In par ticular, we propose that the recurrent excitatory and lateral inhibito ry circuitry in the model, and observed in IMHV, produces hysteresis o n the activation state of the units in the model and the principal exc itatory neurons in IMHV. Hysteresis, when combined with a simple Hebbi an covariance learning mechanism, has been shown in this and earlier w ork (Foldiak 1991; O'Reilly and McClelland 1992) to produce translatio n-invariant visual representations. The hysteresis and learning rule a re responsible for a sensitive period phenomenon in the network, and f or a series of novel temporal blending phenomena. These effects are em pirically testable. Further, physiological and anatomical features of mammalian visual cortex support a hysteresis-based mechanism, arguing for the generality of the algorithm.