Kumar Krishna Agrawal

I'm a PhD student in Computer Science at UC Berkeley, where I'm affiliated with Berkeley AI Research and UCSF. I am incredibly fortunate to be advised by Prof. Adam Yala and Prof. Trevor Darrell. Previously, I was an AI resident at Google Brain, where I worked on fast, simple algorithms for off-policy robot learning and deep generative models for music and program synthesis.

My research focuses on data and compute-optimal algorithms and systems for human-centric AI, with particular emphasis on deep multimodal representation learning and its applications in healthcare and robotics.

Before Berkeley, I studied mathematics and computer science at IIT Kharagpur, and was advised by Pabitra Mitra and Somesh Kumar. I have also been fortunate to spend time at MILA, advised by Yoshua Bengio.

Attribute diversity determines the systematicity gap in VQA
Ian Berlot-Attwell, Annabelle M. Carrell, KKA, Yash Sharma, Naomi Saphra
Harnessing small projectors and multiple views for efficient vision pretraining
KKA, Arna Ghosh, Shagun Sodhani, Adam Oberman, Blake Richards
Falcon: Live video analytics without profiling
Gur-Eyal Sela, KKA, Bharath Balaji, Joseph Gonzalez, Ion Stoica
On Different Faces of Model Scaling in Supervised and Self-Supervised Learning
Matteo Gamba, Arna Ghosh, KKA, Blake Richards, Hossein Azizpour, Mårten Björkman
selected publications
Neural population geometry across model scale: a tool for cross-species functional comparison of visual brain regions
Arna Ghosh, KKA, Zahraa Chorghay, Arnab Kumar Mondal, Blake Richards
Cosyne, 2023
Assessing representation quality in ssl by measuring eigenspectrum decay
Arna Ghosh, KKA, Arnab Kumar Mondal, Blake Richards
Neurips, 2022
Octopus : Low-latency & adaptive perception pipelines
Gur-Eyal Sela, Ionel Gog, Justin Wong, KKA, Sukrit Kalra, Peter Schafhalter, Xiangxi Mo, Xin Wang, Bharath Balaji, Ion Stoica, Joseph Gonzalez
ECCV, 2022
Learning from an exploring demonstrator: Optimal reward estimation for bandits
Wenshuo Guo, KKA, Aditya Grover, Vidya Muthukumar, Ashwin Pananjady
Discrete flows: invertible generative models for discrete data
Dustin Tran, Keyon Vafa, KKA, Laurent Dinh, Ben Poole
Neurips, 2019
Gansynth : Adversarial Neural Audio Synthesis
Jesse Engel, KKA, Shuo Chen, Ishaan Gulrajani, Chris Donahue, Adam Roberts
ICLR, 2019
Discriminator actor critic: Addressing sample inefficiency and reward bias in adversarial imitation learning
Ilya Kostrikov, KKA, Debidatta Dwibedi, Sergey Levine, Jonathan Tompson
ICLR, 2019