Temel Kavramlar
MARBLE provides interpretable representations of neural dynamics based on local flow fields, enabling unsupervised comparison across conditions and systems.
Özet
The MARBLE framework introduces a novel approach to represent neural dynamics using local flow fields over manifolds. By leveraging geometric deep learning, MARBLE can infer latent representations that are highly interpretable and consistent across different conditions and systems. This unsupervised method outperforms current techniques by providing best-in-class decoding accuracy without the need for behavioral labels. Through extensive benchmarking, MARBLE demonstrates its ability to capture complex non-linear dynamics and reveal qualitative changes in dynamical landscapes. The method is robust to sparse and irregularly sampled data typical in neural recordings, offering a powerful tool for understanding neural computations and behavior.
İstatistikler
Using both in silico examples from non-linear dynamical systems and recurrent neural networks and in vivo recordings from primates and rodents, we demonstrate that MARBLE can infer latent representations that are highly interpretable in terms of global system variables such as decision-thresholds, kinematics or internal states.
Our results suggest that using the manifold structure in conjunction with temporal information of neural dynamics provides a common framework to develop better decoding algorithms and assimilate data across experiments.
Through extensive benchmarking, we show that unsupervised MARBLE provides best-in-class within- and across-animal decoding accuracy, comparable to or significantly better than current supervised approaches, yet without the need for behavioral labels.
Alıntılar
"Our results suggest that using the manifold structure in conjunction with temporal information of neural dynamics provides a common framework to develop better decoding algorithms." - Research Team
"MARBLE combines ideas from empirical dynamical modeling and statistical descriptions of collective systems to represent non-linear dynamics over manifolds." - Research Team
"We show that this data-driven metric is expressive enough to infer continuous and qualitative changes in the dynamical landscape of recurrent neural networks during sensory gain modulation or in decision-making tasks at the decision threshold." - Research Team