BunDLe-Net: Learning minimal neural manifolds that represent only behaviourally relevant dynamics

Abstract

Neural manifold learning techniques represent high-dimensional neural dynamics in low-dimensional embeddings to reveal the intrinsic structure of neural manifolds. We introduce a novel neural manifold learning technique, BunDLe-Net, that learns a low-dimensional Markovian embedding of the neural dynamics which preserves only those aspects of the neural dynamics that are relevant for a given behavioural context. In contrast to other techniques that learn embeddings which allow a good reconstruction of the original data, BunDLe-Net eliminates neural dynamics that are irrelevant for decoding behaviour, effectively de-noising the data to reveal better the intricate relationships between neural dynamics and behaviour. We show that BunDLe-Net learns highly consistent manifolds across animals that reveal the building blocks of their neural manifolds on a variety of data sets, ranging from calcium imaging data recorded in the nematode C. elegans to spiking data from the rat hippocampus.