Pumiao Yan

Fully-implantable intracortical brain-computer interfaces (iBCIs) have the ability to revolutionize neuroscience and medicine. However, the capability, performance, and robustness of iBCIs are limited by the current state of neural decoding algorithms. Machine Learning (ML) based algorithms outperform state of the art decoders, but new ML based hardware implementations of neural decoders must be far more energy efficient than any design that exists currently. I am investigating a variety of ML algorithms, with area and power constraints in mind, in both offline studies and human trials. My goal is to determine the computational limits of ML decoders given the constraints posed by hardware and demonstrate the potential of future fully implantable iBCIs. I hope my project not only produces optimized ML decoder hardware but also offers an algorithm/hardware co-design strategy, driven by data and applications, that can be applied to real-time embedded intelligent systems.