Effects of Neuronal Correlations on Population Decoding and Encoding Models

In this thesis, we analyze the effect of the correlations in neural activity on the information that is encoded in and can be decoded from a population of neurons. Various noise models describing these correlations are considered in particular, we use models that take into account the pairwise correlations and other, simpler models that assume shared global additive and/or multiplicative noise factors. The performance of these models on firing rate prediction (encoding) and population decoding are studied. Our analyses show a significant beneficial effect of pairwise correlations on encoding models, with much of this benefit being explained by the global noise models. However, the effects of correlations on decoding vary among our datasets, providing an empirical justification to the theoretical results suggesting correlations can be either helpful or harmful to decoding. Thesis Supervisor: Tomaso Poggio Title: Eugene McDermott Professor