A compressed code for memory discrimination

Abstract

The ability to discriminate similar visual stimuli has been used as an important index of memory function. This ability is widely thought to be supported by expanding the dimensionality of relevant neural codes, such that neural representations for the similar stimuli are maximally distinct, or "separated." An alternative hypothesis is that discrimination is supported by lossy compression of visual inputs, efficiently coding sensory information by discarding seemingly irrelevant details. A benefit of compression, relative to expansion, is that it allows the individual to efficiently retain fewer essential dimensions underlying stimulus variation---a process linked to higher-order visual processing---without hindering discrimination. Under the compression hypothesis, memory discrimination is facilitated when more redundant information from similar stimuli can be discarded to emphasize distinct information. We test the compression versus expansion hypotheses by predicting performance on a memory discrimination task (combined n = 723). First, we train neural networks to compress perceptual and semantic factors of stimuli, and measure the lossiness of those representations using the mathematical framework underlying compression. Consistent with the compression hypothesis, we find that greater lossiness predicts the ease and performance of lure discrimination. We empirically confirm these predictions across image sets, four behavioral datasets, and alternative metrics of lossiness. Using representational similarity analysis and two task fMRI datasets (combined n = 91), we identify compressed representations in the higher-order visual stream inferotemporal cortex (IT) as well as hippocampal subregion dentate gyrus/CA3. Finally, we observe retrieval-related compression in biologically realistic neural network models of cortico-hippocampal pattern separation. These results suggest lossy compression may support memory discrimination.