Visual Categorization: How the Monkey Brain Does It

The computational processes underlying object categorization in cortex are still poorly understood. In a recent experiment, Freedman et al. recorded from inferotemporal (IT) and prefrontal cortices (PFC) of monkeys performing a “cat/dog” categorization task ([1] and Freedman, Riesenhuber, Poggio, Miller, Soc. Neurosci. Abs.). In this paper we analyze the tuning properties of view-tuned units in our HMAX model of object recognition in cortex [2, 3] using the same paradigm and stimuli as in the experiment. We then compare the simulation results to the monkey neuron population data. We find that the view-tuned model units’ tuning properties are very similar to those of IT neurons observed in the experiment, suggesting that IT neurons in the experiment might respond primarily to shape. The population of experimental PFC neurons, on the other hand, shows tuning properties that cannot be explained just by stimulus tuning. These analyses are compatible with a model of object recognition in cortex [4] in which a population of shape-tuned neurons responding to individual exemplars provides a general basis for neurons tuned to different recognition tasks. Simulations further suggest that this strategy of first learning a general representation as input to a classifier simplifies the learning task. Indeed, the physiological data are compatible with the notion of PFC neurons performing a simple classification based on the thresholding of a linear sum of the inputs from examplar-tuned units. Such a strategy has various computational advantages, especially with respect to transfer across recognition tasks.