Task-Conditioned Latent Alignment for Cross-Session Neural Decoding
A new framework called Task-Conditioned Latent Alignment (TCLA) has been introduced by researchers to enhance cross-session neural decoding, particularly when data from the target session is scarce. TCLA employs an autoencoder to derive low-dimensional neural representations from a data-rich source session, subsequently aligning the target's latent representations to the source in a manner conditioned by the task. When tested on datasets from macaque motor and oculomotor center-out tasks, TCLA consistently surpassed baseline techniques that relied exclusively on target-session data, leading to improved decoding performance across various datasets and conditions.
Key facts
- TCLA is a framework for cross-session neural decoding with limited target-session data.
- It uses an autoencoder to learn low-dimensional neural representations from a source session.
- Target latent representations are aligned to the source session in a task-conditioned manner.
- Evaluated on macaque motor and oculomotor center-out datasets.
- TCLA consistently improves decoding performance compared to baselines trained only on target-session data.
- The paper is available on arXiv with ID 2601.19963.
- The announcement type is replace-cross.
- The framework addresses the challenge of limited data in recording sessions.
Entities
Institutions
- arXiv