Mapping the relationship between neural activity and motor behavior is a central aim of sensorimotor neuroscience and neurotechnology. While most progress to this end has relied on restricting complexity, the advent of foundation models instead proposes integrating a breadth of data as an alternate avenue for broadly advancing downstream modeling. We quantify this premise for motor decoding from intracortical microelectrode data, pretraining an autoregressive Transformer on 2000 hours of neural population spiking activity paired with diverse motor covariates from over 30 monkeys and humans. The resulting model is broadly useful, benefiting decoding on 8 downstream decoding tasks and generalizing to a variety of neural distribution shifts. However, we also highlight that scaling autoregressive Transformers seems unlikely to resolve limitations stemming from sensor variability and output stereotypy in neural datasets.

Intracortical brain-computer interfaces (iBCIs) can restore movement and communication abilities to individuals with paralysis by decoding their intended behavior from neural activity recorded with an implanted device. While this activity yields high-performance decoding over short timescales, neural data is often nonstationary, which can lead to decoder failure if not accounted for. To maintain performance, users must frequently recalibrate decoders, which requires the arduous collection of new neural and behavioral data. Aiming to reduce this burden, several approaches have been developed that either limit recalibration data requirements (few-shot approaches) or eliminate explicit recalibration entirely (zero-shot approaches). However, progress is limited by a lack of standardized datasets and comparison metrics, causing methods to be compared in an ad hoc manner. Here we introduce the FALCON benchmark suite (Few-shot Algorithms for COnsistent Neural decoding) to standardize evaluation of iBCI robustness. FALCON curates five datasets of neural and behavioral data that span movement and communication tasks to focus on behaviors of interest to modern-day iBCIs. Each dataset includes calibration data, optional few-shot recalibration data, and private evaluation data. We implement a flexible evaluation platform which only requires user-submitted code to return behavioral predictions on unseen data. We also seed the benchmark by applying baseline methods spanning several classes of possible approaches. FALCON aims to provide rigorous selection criteria for robust iBCI decoders, easing their translation to real-world devices. https://snel-repo.github.io/falcon/

In human-machine interfaces, decoder calibration is critical to enable an effective and seamless interaction with the machine. However, recalibration is often necessary as the decoder off-line predictive power does not generally imply ease-of-use, due to closed loop dynamics and user adaptation that cannot be accounted for during the calibration procedure. Here, we propose an adaptive interface that makes use of a non-linear autoencoder trained iteratively to perform online manifold identification and tracking, with the dual goal of reducing the need for interface recalibration and enhancing human-machine joint performance. Importantly, the proposed approach avoids interrupting the operation of the device and it neither relies on information about the state of the task, nor on the existence of a stable neural or movement manifold, allowing it to be applied in the earliest stages of interface operation, when the formation of new neural strategies is still on-going. In order to more directly test the performance of our algorithm, we defined the autoencoder latent space as the control space of a body-machine interface. After an initial offline parameter tuning, we evaluated the performance of the adaptive interface versus that of a static decoder in approximating the evolving low-dimensional manifold of users simultaneously learning to perform reaching movements within the latent space. Results show that the adaptive approach increased the representational efficiency of the interface decoder. Concurrently, it significantly improved users' task-related performance, indicating that the development of a more accurate internal model is encouraged by the online co-adaptation process.