Stabilizing brain-computer interfaces through alignment of latent dynamics

Abstract

Intracortical brain-computer interfaces (iBCIs) restore motor function to people with paralysis by translating brain activity into control signals for external devices. In current iBCIs, instabilities at the neural interface result in a degradation of decoding performance, which necessitates frequent supervised recalibration using new labeled data. One potential solution is to use the latent manifold structure that underlies neural population activity to facilitate a stable mapping between brain activity and behavior. Recent efforts using unsupervised approaches have improved iBCI stability using this principle; however, existing methods treat each time step as an independent sample and do not account for latent dynamics. Dynamics have been used to enable high performance prediction of movement intention, and may also help improve stabilization. Here, we present a platform for Nonlinear Manifold Alignment with Dynamics (NoMAD), which stabilizes iBCI decoding using recurrent neural network models of dynamics. NoMAD uses unsupervised distribution alignment to update the mapping of nonstationary neural data to a consistent set of neural dynamics, thereby providing stable input to the iBCI decoder. In applications to data from monkey motor cortex collected during motor tasks, NoMAD enables accurate behavioral decoding with unparalleled stability over weeks- to months-long timescales without any supervised recalibration.

Using adversarial networks to extend brain computer interface decoding accuracy over time

“…Although earlier attempts to achieve alignment via KLDM were only moderately successful, a recent approach using KLD to align neural latent dynamics identified using Latent Factor Analysis through Dynamical Systems (LFADS) (33, 34) was much more successful (32). In their evaluation, the resulting “Nonlinear Manifold Alignment with Dynamics” (NoMAD) also outperformed ADAN.…”

“…Other approaches to address iBCI decoder instability include supervised techniques that aim at stabilizing iBCI performance by recalibrating the decoder during ongoing iBCI control by relying on access to the task output variables (28)(29)(30), as well as unsupervised methods that do not require to re-estimate decoder parameters and only need neural data, with no provided task output variables or task labels (9,16,21,31,32). We restricted our comparison to GAN-based aligners and PAF for several reasons.…”

Universals for coding action in motor cortical networks: conserved neural ensemble latent space geometry for reach and grasp tasks

“…Other algorithms also do not implement shapepreserving alignment strategies. For example, NoMAD (Karpowicz, 2022), uses a deep-learning based approach aimed at modeling underlying neural dynamics by combining data across sessions. Both decoding methods DAD (Dyer, 2017) and CEBRA (Schneider, 2022) rely on the common kinematic space to align neural data across sessions.…”

Monkey-to-human transfer of brain-computer interface decoders

“…These aligned latent signals maintain a remarkably stable relation to behavior over months and even years 12 . As a consequence, a fixed decoder that uses these aligned signals as inputs remains accurate across long periods without supervised recalibration 11,12,23,24 .…”

“…Other techniques have been used to align the statistics of two clouds of points, independently of the time course of the associated signals 23,24,50 . These unsupervised methods have been tested on neural recordings corresponding to structured, trial-based behaviors, although they have been successful even without the use of any information about the behavior itself.…”