In my research, I aim to characterize principles of biological computation that are useful for building efficient, adaptive, and robust artificial systems. My thesis work focuses on the neural computation of invariant and equivariant representations in the domain of vision, using ideas from group representation theory.
There has been a recent swell of interest in this area in the machine learning community, but most approaches require hand-specifying the groups in the data. I tackle the problem of instead learning and factoring group structure with architectures that build in generic structures from group representation theory without specifying specific groups.
In other work, I build neuron models whose dynamics implicitly compute group representations. These suggest hypotheses for how such representations might be implemented in the visual system.
My research lies on the theoretical side of the computational neuroscience spectrum (as opposed to the data analysis side) and is mostly oriented towards machine learning applications, but I keep an eye toward empirical predictions.
Key interests: vision, motion perception, video processing, neuromorphic computing, dynamic vision sensors, spiking neural networks, dynamic neuron models, group representation theory, unsupervised learning, representation learning