Fig. 3: Robustness comparison of the context decoding performance in the PFC and the MD when different levels of noise are introduced to PFC units or the inputs. The MD outperformed the PFC in context decoding under various noise conditions, especially in the delay period when the cue stimuli were off.
From: Rapid context inference in a thalamocortical model using recurrent neural networks
A Comparisons of the context decoding performance with varying levels of PFC noise in the cue (top) and delay (bottom) periods. Shaded areas denote the standard deviations. B Comparisons of the context decoding performance with varying levels of input noise in the cue (top) and delay (bottom) periods. Shaded areas denote the standard deviations. C, D The MD context selectivity is robust to different model hyperparameters, e.g., the number of the MD neurons, the number of contexts. The MD can selectively encode context rather than rule with different numbers of the MD neurons and contexts in the experiment setups. Shaded areas denote the standard deviations.
