Kumar Krishna Agrawal
email: krishna@eecs.berkeley.edu
I am a PhD candidate at EECS, UC Berkeley. I am broadly
interested in algorithms and systems for efficient, reliable & human-centric machine learning.
Previously I was an AI resident at Google Brain, working on fast, simple algorithms for off-policy robot learning, generative models, program
synthesis. Before this, I studied mathematics and computer science at IIT Kharagpur, where I worked on multi-modal representation
learning for my master's thesis.
Research: I am broadly interested in empirics, theory which furthers our understanding of
learning algorithms in interactive, uncertain environments; and translates to building efficient,
scalable real-world systems. Some recent themes include:
- assessing trainability and model selection for self-supervised representation learning
- leveraging suboptimal data for offline, inverse reinforcement learning
- efficient systems for large-scale model training, inference
- structured and compositional generative models (for programs, vision, audio)
Research opportunities: If you are an undergraduate at Berkeley and interested in learning more
about neural-networks, machine learning; please send me an email.
about /
teaching /
writing /
reading /
github /
gscholar
|
|