Researchers have developed BEEMO, a novel framework for neural decoding that decouples the learning of neural and behavioral representations. This approach allows for the incorporation of abundant, easily collected behavioral data alongside scarcer, more invasive paired neural-behavioral recordings. By training a behavior decoder on trajectories and using cross-attention for neural activity when available, BEEMO enables independent scaling of both data types. Experiments on nonhuman primate motor datasets demonstrate that using unpaired behavioral data improves decoding performance, especially for task transfer, and that behavior-only pretraining can rival models trained on larger neural datasets. AI
IMPACT This approach could significantly reduce the cost and complexity of developing neural decoding models, potentially accelerating advancements in brain-computer interfaces.
RANK_REASON Academic paper detailing a new method for neural decoding. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
