A key translational goal in the lab is to extract “algorithms” from the biological networks we study, study them computationally, and ultimately embed them into neuromorphic hardware. The overall goal of this neuromorphic computing strategy is to enable us to build artificial systems with capabilities approaching those of their biological counterparts. This means artificial noses, of course, but we have found that the structure of the olfactory problem is appropriate for a much broader range of signal restoration and source separation problems.
Our first olfactory bulb-based neuromorphic algorithm, implemented on Intel’s Loihi processor, was one of the first such algorithms with real-world utility, enabling one-shot odor learning and subsequent identification of learned odors despite substantial, unpredictable background interference. Ongoing development on this Sapinet project is utilizing additional brain-inspired strategies including analogue similarity and category learning, based in part on our geometric theory of odor learning.
The lab is a member of the Intel Neuromorphic Research Community, and we port key neuromorphic algorithms to Intel’s Loihi and Loihi2 processors.