Dong Song’s research goals are to understand how the brain performs higher-order cognitive functions and to build cortical prostheses that can restore cognitive functions. Modeling, interface technology, and behavioral tasks will be required to carry out these goals.
The idea behind the hippocampal prosthesis is to mimic the signal processing function of the hippocampal circuit, bypassing a damaged region using multi-electrode arrays. This prosthetic is different from deep brain stimulation in several ways – it is lower intensity, multi-channel asynchronous signaling with both decoding and encoding stages. A new type of multiple input multiple output (MIMO) system needs to be developed, which is what Song is currently working on. He has already observed MIMO stimulation facilitating memory formation in humans.
Deep learning is used to develop the model to convert input signal into the correct output signal. Double layer pattern-matching helps validate the models accuracy. After only 100-200 trials, roughly 30-50% of the signal can be accurately decoded this way.
Neural interface technology has historically been limited, yet it is critical to prosthesis development. A new high density, chronical, parylene neural interface is being developed in collaboration with Ellis Meng. Song has a grant to test these interfaces in different species and different brain regions for large scale recording and stimulation.