GRAFT: Gain-Recalibrated Adapters for Transformer-Based Neural Population Activity Modeling

2026-06-09

AI Summary

This work addresses the challenge of cross-day variability in brain–computer interfaces, where traditional neural population activity models struggle due to fixed mappings between input/output layers and specific neurons, which cannot accommodate day-to-day changes in neuron identity, count, or response statistics. To overcome this limitation, the authors propose GRAFT, a novel architecture that decouples a recalibratable neural interface from a shared temporal dynamics backbone. By integrating gain modulation, positional encoding, and lightweight adapters, GRAFT enables efficient cross-day transfer while updating only 9.21% of its parameters. On the NLB'21 MC_Maze benchmark, the model achieves a state-of-the-art co-bps score of 0.3866 and demonstrates robust performance across Large, Medium, and Small cross-day datasets with scores of 0.3749, 0.3112, and 0.3152 co-bps, respectively.

Abstract

Neural population activity models can recover rich temporal structure from binned spikes, but their read-in and readout layers often remain tied to a fixed set of recorded neurons. This coupling limits reuse in long-term brain-computer interfaces, where recorded neuron identities, counts, and response statistics can change across days. We introduce GRAFT, a Transformer-based neural population activity model that separates reusable temporal dynamics from a recalibratable neuron interface. The neuron interface controls how recorded neurons enter and leave the shared backbone, and auxiliary gain and positional mechanisms support neural activity modeling inside the Transformer. On MC Maze under the standard NLB'21 protocol, GRAFT reaches 0.3866 co-bps as an ensemble, setting a new state of the art on the primary co-bps metric among public and reported NLB'21 results. In a cross-day protocol constructed from the NLB'21 MC Maze dataset series, GRAFT recalibrates from MC Maze to the scaled MC Maze datasets (Large/Medium/Small) by updating only 9.21% of parameters, reaching 0.3749, 0.3112, and 0.3152 co-bps with restricted target-day support sets. These results show that the same interface-backbone separation supports both strong Transformer-based neural population activity modeling and data-efficient cross-day recalibration.

Problem

• neural population activity modeling
• brain-computer interface
• cross-day recalibration
• neuron identity shift
• temporal dynamics

Innovation

• Gain Recalibration
• Adapter
• Transformer
• Neural Population Modeling
• Cross-day Adaptation