Neuroscience and engineered machine intelligence. Combined.
We use mathematics and engineering to understand the brain as an engineered system. And develop next generation machine learning from the brain.
Our lab's research is focused at the intersection between neuroscience and new models of machine learning and computing. We are interested in mathematical models and algorithms capable of one-shot online learning that do not require training or prior exposure to data classes. In particular, models derived from functionally accurate neurophysiological mechanisms. At a foundational level, the key questions we are interested in involve understanding how and why online learning works in these models, and what classes of problems they are uniquely suited for that the existing state of the art in machine learning cannot solve. What classes of functions can these models operate on? How computationally robust are they? Under what constraints and conditions are convergent unique solutions guaranteed? And how can principles of neural plasticity (facilitation and depression) and neural morphology and architecture allow such models to adapt in contextually relevant ways (e.g. due to changing external queries or other considerations) in real time, without any training?
At the same time, we have a strong interest in how thinking about machine learning from a neuroscience perspective can inform neuroscience itself. On-going work in our group aims to understand what happens when there is a breakdown in signaling in neurodevelopmental disorders; in particular as Autism Spectrum Disorder. Much of our focus is in studying how network geometric and topological structure constrains and determines dynamic function. We also have an interest in the development of advanced brain machine interfaces and neural prosthesis that adapt and learn to new data in order to modulate their outputs.