Hello, I’m Abhinav!

I am a Research Fellow at Microsoft Research India, working at AI Infrastructure Group under Dr. Ramachandran Ramjee, Dr. Nipun Kwatra and Dr. Sanjeev Krishnan.

I’m interested in building more efficient AI systems (specifically centred around LLM inference), ranging from exact methods like scheduling to approximate ones like compression (quantization, pruning etc.) to tweaking the architecture itself! Additionally, I am kneen on developing hardware-aware algorithms that can actually run faster on existing accelerators (for eg. most quantized LLMs actually run slower in practice!).

I am also interested in building more robust evaluations that can capture the nuanced effects of various efficiency techniques applied to LLMs. (too many works seemingly compromise the representative capability of LLMs without sacrificing anything on standard evaluations, but only some of them truly work)

Before joining MSR India, I was working with Prof. Ron Shamir at the ACTG Lab. I also worked as a Research Intern at McGill University under Prof Xujie Si as a MITACS GRI awardee.

Please get in touch with me via email if you would like to chat about research or collaboration!

Publications

• Accuracy Is Not All You Need
  Abhinav Dutta, Sanjeev Krishnan, Nipun Kwatra, Ramachandran Ramjee. Conference on Neural Information Processing Systems (NeurIPS), 2024.
  abstract | openreview | project page

• Parameterized syncmer schemes improve long-read mapping
  Abhinav Dutta, David Pellow, Ron Shamir. PLOS Computational Biology, 2022.
  paper | code | project page