Processing Tools for Peta-Scale Neuroscience
We work extensively on large, peta-scale datasets collected by the US BRAIN Initiative, particularly in the space of Electron Microscopy and X-Ray Microtomography. I contribute to efforts to store (https://bossdb.org/) and process (https://github.com/aplbrain) these emerging datasets. We focus on the cloud infrastructure, software engineering, signal processing, and machine learning required to turn these highly exciting and valuable datasets into novel discoveries.
Neuroscience-Inspired Artificial Intelligence
Another major thrust of my work is developing algortihmic approaches inspired by neuroscience. Examples include research into spiking neural networks, novel network architectures inspired by the brain, and fusion of connectionist and symbolic architectures. Other ongoing thrusts include investigation of neuromorphic implementations of these approaches and incorporating insight from the neuroscience of sensory coding into brain-computer interface systems.
Biological Inspiration for Low-Complexity Robotics
I am currently interested in robotic agents which show the levels of autonomy and robustness observed in biological agents. We’re also interested in the extreme energy efficiency (relative to current robotic platforms) seen in the sensory and decision making systems of animal.
My previous work at Sprite Robotics involved modelling, analysis, and control of robotic systems. Our work was funded by an NSF SBIR grant to investigate motion of fast wheeled mobile robots through home and office environments. We developed models of small robots for rapid movement, online parameter learning, stochastic control techniques, and fusion with visual data.
Encoding by Sensory Neurons
During my thesis work, my primary research interest lies at the intersection of signal processing, information theory, and neuroscience. With my collaborators, we have developed a model of energy-constrained coding by sensory neurons, which predicts the neural activity required to minimize encoding error. This line of research aims to answer fundamental questions in neuroscience- how neurons represent and transmit information. Incorporating a metabolic energy-constraint, along with other biological constraints, is critical to understanding how neurons function in living organisms. Improving our understanding of neural encoding affects our entire understanding of the information processing chain in sensory systems had has important implications for artificial systems which interact with neurons.
Algorithms for Brain-Machine Interfaces
Another research interest is the design of brain machine interfaces and detection algorithms for brain machine interfaces. I am interested in improving stimulation and decoding in existing brain machine interfaces. Also, I’ve been interested in identifying new application areas where brain machine interfaces might be reasonable interface choices for any individual. This work involves neuroscience and engineering, but also questions of design and usability.
Signal Processing with Sparsity Constraints
Previously, I’ve also explored algorithms for sparse reconstructions and compressed sensing. My work focused on parameterized measurement models with continuous valued parameters. Such models are found in common applications like spectral analysis and array signal processing. I developed recovery algorithms for this case, and I am generally interested in signal processing and machine learning which incorporates sparsity constraints.