Our research lies at the intersection of high-dimensional statistics, optimization, and time-series analysis, with applications to neuroscience and multimedia signal processing. Recently, we have taken a keen interest in the connection between artificial neural networks/deep learning and sparse signal processing, as a means to understand the principles of hierarchical representations of sensory signals in the brain, and to develop explainable AI. This research aims to explain artificial neural networks as inference algorithms in biologically-plausible, generative, statistical models. It enables the principled design of model-based, explainable artificial neural networks, and their theoretical study via a study of the properties of the generative model associated with a given architecture. When the architecture arises from popular models of hierarchical sensory processing in the brain, they can give novel insights on the limitations of these models and how to improve them. Finally, the design of architectures tailored to generative models lets us leverage GPUs, and the computational infrastructure that has been developed to train artificial neural networks, to solve inference and parameter estimation problems that rely on generative models of data.
You can find sample projects related to two areas of research we’ve been interested in recent years here: