Mice LFP Data: GRU-Transformer

Overview

Abstract

Our experiments demonstrate that explicitly modeling temporal structure in band-limited local field potential (LFP) activity with a GRU-Transformer yields significantly better decoding of visual stimuli than a linear classifier trained on flattened power features. Improvements were consistent across multiple sessions, conditions, and class-imbalanced datasets, highlighting the importance of temporal dynamics in neural decoding.

Insight

Key Takeaway

Temporal dynamics carry crucial discriminative information; class imbalance can be mitigated with weighting and augmentation; GRU-Transformer significantly improves per-class recall and overall decoding accuracy (~10–15 percentage points). Findings support the translation of sequence modeling approaches from mouse LFP to human EEG applications, offering a methodological template for cross-species neural decoding.

Figures

Visuals

Resources

Links