About Me

I’m an ML Engineer and an Independent Researcher. Previously I collaborated with Sara Hooker, Ferdinando Fioretto and Beyza Ermis at Cohere For AI as a Community Researcher.

My current research interests are based on overcoming the deficiencies in current Foundational models. More specifically, my research interests are in:

• Robustness: The real world is not a fixed static distribution [2]. How can we train models that learn robust features, generalize Out-of-Distribution and are Adversarially Robust as well?
• Trustworthy and Fairness: For ML models to be widely adopted without much human intervention, they need to align with human values, be truthful, and be fair [1] even if the training dataset is biased.

Working towards this goal I have studied recently, how hardware choice affects fairness in ML systems (ICML 2024), how LLMs handle distrubtion shift in continual pre-training and using LLMs to generate currcicula for OOD generalization in RL (NeurIPS 2023).

Recent Updates

• May ‘24: We’re thrilled to have our work accepted at ICML 2024! Our work demonstrates how the choice of GPU affects the fairness of underrepresented groups from a theoretical perspective supported by empirical experiments. We propose a mitigation solution that helps alleviate the unfairness caused. This is my first first-author paper in a major conference!

• Jan ‘24: I completed my Master in Computer Science at UNB, and would be joining as an ML Engineer at a startup in Canada.

• Sept ‘23: Our work on a new version of Neural MMO got accepted at NeurIPS 2023 D&B. I worked on automated curriculum generation for RL agents, to help them learn new skills using In-Context Learning, LLMs and QD Algorithms.

Publications

(* - Denotes equal contribution)

• On The Fairness Impacts of Hardware Selection in Machine Learning
  Shree Harsha Nelaturu*, Nishaanth Kanna Ravichandran*, Cuong Tran, Sara Hooker, Ferdinando Fioretto
  (to appear) ICML ‘24
  [arXiv]

• Investigating Continual Pretraining in Large Language Models: Insights and Implications
  Çağatay Yıldız, Nishaanth Kanna Ravichandran, Prishruit Punia, Matthias Bethge, Beyza Ermis
  (submitted to) CoLLAs ‘24
  [arXiv]

• Neural MMO 2.0: A Massively Multi-task Addition to Massively Multi-agent Learning
  Joseph Suarez, David Bloomin, Kyoung Whan Choe, Hao Xiang Li, Ryan Sullivan, Nishaanth Kanna Ravichandran, Daniel Scott, Rose Shuman, Herbie Bradley, Louis Castricato, Phillip Isola, Chenghui Yu, Yuhao Jiang, Qimai Li, Jiaxin Chen, Xiaolong Zhu
  NeurIPS 2023 Datasets and Benchmarks
  [neurips.cc]