Dr. Liu earned his Ph.D. of Neurobiology from Yi Wang lab at Institute of Biophysics, Chinese Academy of Science, where he studied the neural coding of multidimensional stimuli features in primary visual cortex. After Ph.D. he was trained in computational neuroscience in Osborne lab at The University of Chicago, investigating the efficient coding of motion information in MT area and smooth pursuit eye movement. Then he moved to Carmena lab at University of California, Berkeley to investigate neural adaptation during Brain-machine interface (BMI) learning. Recently he is working in Duke University to investigate the feasibility of implementing a BMI system with eye motion signals from FEF area.
How the brain communicates with the outside world is one of the most important questions in neuroscience, as well as in Brain-machine interface (BMI). Our previous researches showed that the neural information coding is not a passive representation, but an active building up process. This active process includes adaptation, learning, synergistic coding, etc, which demonstrating an efficient coding principle. Similarly, we asked how would the neural system actively learn to acquire a skillful control of a BMI system? Our studies showed that neural adaptation is crucial for facilitating BMI learning. Furthermore with closed-loop decoder adaptation (CLDA) algorithms, we investigated how to combine neural and decoder adaptations to boost BMI learning in this ‘two-learner’ system. We have shown that the initial assist of CLDA will yield improved early performance of BMI, but it does not improve the rates of skillful neuroprosthetic learning. Instead, the skillful BMI control and learning can only derive from neuron adaptation. These results provide the neural mechanisms of BMI learning, and also the potential design principles for the next generation BMI, in order to create a reliable and robust neuroprosthetic system.