Generalized associative representations in parietal cortex

Associating sensory stimuli is a critical aspect of behavior. Neurons in the lateral intraparietal area (LIP) of Macaca mulatta have been shown to reflect learned associations between directions of moving visual stimuli. Individual LIP neurons might encode associations only for specific stimuli, such as motion directions; alternatively, they may generally encode associations whenever animals must decide between discrete alternatives. To test this, I asked whether LIP neurons reflect learned associations between pairs of arbitrarily-chosen static shapes. Additionally, I asked whether the same neurons encode associations between motion directions. The neurons robustly encoded the learned pair associations between shapes, and shape-pair selective neurons tended to also be selective for direction associations. These findings suggest that representing generic categorical outcomes may be a fundamental role of parietal neurons.;Analyses of neuronal populations typically consider the magnitude of associative encoding rather than the sign, that is, which group of stimuli is preferred. In lower visual cortical areas, preferred features are typically broadly distributed among neurons. But would the same unbiased representations hold for learned associations in parietal cortex? Surprisingly, the distribution of preferred associations was strikingly biased: nearly all neurons recorded from an animal had higher activity for the same motion-direction category or shape pair. In addition, I found a similar bias in a previous study from our lab and also in a perceptual decision-making study from another lab. In no case could we account for the population bias by behavioral factors, such as asymmetries in behavioral performance. At a computational level, a recent theoretical study proposed that periods of relatively stable firing in LIP are "one-dimensional," in the sense that the vector of firing rates of all neurons in the population is linearly related whenever firing rates change slowly, such as during spontaneous firing or slowly changing memory-delay activity. Because we found that different pair-associates or categories elicit different levels of delay activity in individual neurons, a linear relationship between the neuronal population's spontaneous and delay activity would predict a population-level bias in the preferred associations.