AI Insight
REALM is a retrospective distillation framework designed to enable real-time decoding of neural behavior from local field potentials (LFPs), a lower-bandwidth alternative to spike signals in brain-computer interfaces. The approach trains a bidirectional Mamba-2 teacher model using masked autoencoding, then transfers its representational knowledge to a compact causal student model through a combined objective of representation alignment and task supervision. REALM outperforms existing causal and non-causal LFP decoding methods while reducing parameter count by half and training time by a factor of ten.
Why it matters
As implantable brain-computer interfaces move toward wireless and high-channel-count designs, the energy and bandwidth demands of spike-based decoding become impractical, and REALM offers a scalable pathway to reliable real-time neural decoding using lower-cost signals. This could accelerate the development of clinically viable neuroprosthetics and assistive communication devices.
arXiv:2605.14867v1 Announce Type: cross
Abstract: Spike activity has been the dominant neural signal for behavior decoding due to its high spatial and temporal resolution. However, as brain-computer interfaces (BCIs) move toward high channel counts and wireless operation, the high sampling frequency of spike signals becomes a bottleneck due to high power and bandwidth requirements. Local field potentials (LFPs) represent a different spatial-temporal scale of brain activity compared to spikes, offering key advantages including improved long-term stability, reduced energy consumption, and lower bandwidth requirement. Despite these benefits, LFP-based decoding models typically show reduced accuracy and often rely on non-causal architectures that are unsuitable for real-time deployment. To address these challenges, we propose REALM: a retrospective distillation framework that enables causal LFP decoding. Inspired by offline-to-online distillation strategies in speech recognition, REALM transfers representational knowledge from a pretrained multi-session bidirectional LFP model to a causal version for real-time deployment. We first pretrain a bidirectional Mamba-2 teacher model using a masked autoencoding objective. We then distill this teacher model into a compact student model via a combined objective of representation alignment and task supervision. REALM consistently outperforms both causal and non-causal LFP-based SOTA methods for behavior decoding. Notably, our REALM improves decoding performance while achieving a $2times$ reduction in parameter count and a $10times$ reduction in training time. These results demonstrate that retrospective distillation effectively bridges the gap between offline and real-time neural decoding. REALM shows that LFP-only models can achieve competitive decoding performance without reliance on spike signals, offering a practical and scalable alternative for next-generation wireless implantable BCIs.
Source: REALM: Retrospective Encoder Alignment for LFP Modeling