Learning predictive latent structure from unclustered neural spike trains
Neural activity patterns enable perception, thought, and action. A powerful approach to understanding these abilities involves identifying low-dimensional, interpretable latent variables that capture the relationship between activity patterns, the information they represent, and the computations they support. Yet current methods often yield variables that are difficult to interpret and require…
Neural activity patterns are crucial for perception, thought, and action. Understanding these abilities often involves discovering low-dimensional, interpretable latent variables that represent the information in activity patterns and the computations they support. However, existing methods frequently produce variables that are hard to understand and necessitate data preprocessing, which reduces their precision and temporal resolution.
In this study, researchers present Contrastive Prediction of Point Process Observations (C3PO), a novel unsupervised technique that discovers a predictive and comprehensible latent variable representation directly from continuous-time neural spiking data, without the need for spike-sorting or binning. C3PO uncovers structure that aligns with known latent variables across diverse datasets, even when those variables are not available during training.
It identifies dynamics spanning timescales ranging from tens of milliseconds to hours and uncovers new representations and their associated dynamics. These findings demonstrate the effectiveness of C3PO as a robust tool for extracting interpretable structure from high-dimensional point-process data.
Rewritten by urgent.news; may contain errors — check the original.
Source: biorxiv.org