Kyle Daruwalla
NeuroAI Scholar at CSHL
⚠️ On the job market
I am on the faculty job market this fall! I am primarily looking for electrical and/or computer science departments with collaborations in neuroscience. If my research interests you, please reach out!

Welcome! I'm Kyle, a NeuroAI Scholar at Cold Spring Harbor Lab in Long Island, NY. I completed my PhD studying non-von Neumman computing under Dr. Mikko Lipasti at University of Wisconsin-Madison. My research is broadly organized into three areas:

  1. Applying ML models to neuroscience datasets: Neuroscience data is complex and high-dimensional but contains far fewer and noiser samples than standard ML benchmarks. I believe using existing ML methods in these contexts can help answer scientific questions while also providing unique insight into model failure modes. Currently, I collaborate with the Hou Lab at CSHL to apply 3D pose-tracking models to study facial expressions in mice.

  2. Building ML models inspired by neuronal development and evolution: Deep neural networks are trained on billions of data samples, while biological networks are able to learn within a few training examples by leveraging innate priors encoded in an organism’s genome via evolution. The importance of structural priors is well-known in ML, but a scalable mechanism for learning useful priors does not exist. Biology makes use of two processes for finding innate structure—neuronal development, which translates the information encoded in the genome into a functional network of neurons, and evolution, which mutates the genome to produce better networks. I study these processes through a computational lens to build AI models endowed with prior structure. In turn, these models are more sample-efficient and consume less energy to train.

  3. Studying neural processing using computer science tools: Even in a single brain, various sub-circuits employ different computing paradigms, memory systems, information encoding schemes, and signaling strategies. This diversity allows organisms to balance functional specialization and adaptability under various resource constraints. I study these circuits through the lens of a computer scientist to understand how to build heterogenous and distributed computing systems. I believe such systems will be necessary in the post-Moore's law era of computing.

Quick links: [ CV | Research Statement ]
Contact Info: [ daruwal /at/ cshl.edu | Github | Twitter | butterfly BlueSky | Mastodon ]

Recent Posts

New paper accepted at NeurIPS 2024!

2024-11-06

Our work on the Neural Tangent Ensemble is accepted as a spotlight at NeurIPS 2024!

New paper accepted at UniReps NeurIPS Workshop!

2024-10-10

Work led by CiCi Zheng is accepted as a workshop paper at UniReps!

New preprint: Cheese 3D!

2024-05-01

Our work on tracking facial movements in mice using 3D pose estimation is now on bioRxiv.