We are working on mapping multi-channel neural spike data, recordedfrom multiple cortical areas ofan owl monkey, to corresponding 3d monkey arm positions. In earlier work on this mapping task, we observed that continuous function approximotors (such as artificial neural networks) have diflculty in jointly estimating 36 arm positions for two distinct cases -namely, when the monkey's arm is stationary and when if is moving. Therefore, we propose a multiple-model approach thatfirst classifies neural spike data into two classes, corresponding to two states of the moneky's arm: (1) stationary and (2) moving. Then, the output of this classijier is used (IS a gating mechanism for subsequent continuous models, with one modelper class. In this paper. we first motivote and discuss our approach. Next, we present encouraging results for the class$er stage, based on hidden Markov models (HMMsJ, and also for the entire bimodal mapping system. Finally, we conclude with a discussion ofthe results and suggestfuture avenues of research.

No abstract

We propose two algorithms that decompose the joint likelihood of observing multidimensional neural input data into marginal likelihoods. The first algorithm, boosted mixtures of hidden Markov chains (BMs-HMM), applies techniques from boosting to create implicit hierarchic dependencies between these marginal subspaces. The second algorithm, linked mixtures of hidden Markov chains (LMs-HMM), uses a graphical modeling framework to explicitly create the hierarchic dependencies between these marginal subspaces. Our results show that these algorithms are very simple to train and computationally efficient, while also reducing the input dimensionality for brain-machine interfaces (BMIs).

Abstract-In this paper, we present a design for a wearable DSP system that is capable of processing various neural-tomotor translation algorithms. The system first acquires the neural data through a high speed data bus in order to train and evaluate our prediction models. Then via a widely used protocol, the low-bandwidth output trajectory is wirelessly transmitted to a simulated robot arm. This system has been built and successfully tested with real data.

Abstract. New architectures for Brain-Machine Interface communication and control use mixture models for expanding rehabilitation capabilities of disabled patients. Here we present and test a dynamic data-driven (BMI) Brain-Machine Interface architecture that relies on multiple pairs of forward-inverse models to predict, control, and learn the trajectories of a robotic arm in a real-time closedloop system. A method of window-RLS was used to compute the forwardinverse model pairs in real-time and a model switching mechanism based on reinforcement learning was used to test the ability to map neural activity to elementary behaviors. The architectures were tested with in vivo data and implemented using remote computing resources.