Connecting neural dynamics to behavior through multi-scale analysis of neural systems

Abstract:

Neural and behavioral data span multiple spatial and temporal scales and are heterogeneous (e.g., spike data as point processes vs. local field potentials (LFPs) data as continuous signals; categorical decisions vs. continuous exploration of eye movement). However, existing methods are, modality-specific, and lack integration with behavior. Thus, to facilitate the joint analysis of multi-modal brain and behavioral data, in this talk, we will discuss a set of tools we recently developed (e.g., Safavi et al. PLoS CB 2023) for multi-scale analysis of neural data. These methods include methods for multivariate spike-LFP coupling, as well as a method to connect transient events in LFPs to coordinated dynamics at the level of whole-brain. Lastly, we will discuss our recent developments that allow us, to identify characteristic neural dynamics in (almost) a modality-agnostic fashion, that connects these dynamics to behaviorally meaningful phases. We assess the last method based on neural recordings from macaque monkeys during an attention task, as well as neural dynamics in artificial neural networks (more precisely, recurrent neural networks or RNNs).