Sherin Muckatira
I am a PhD student at the University of Massachusetts, Lowell. I am advised by Dr. Anna Rumshisky. I am broadly interested in understanding and improving the generalization, interpretability, and efficiency of language models. My research spans training dynamics, representation learning, and the emergence of complex behavior in large models. I also explore methods for efficient pre-training to reduce computational cost while maintaining performance. Through my research, I aim to contribute to make LLM training in making large language model (LLM) training more transparent, efficient, and interpretable.
News
- [May 2025] Returning to Amazon for Applied Science Internship.
- [May 2024] Completed an Applied Science internship at Amazon.
- [April 2024] Our paper on Emergent Abilities accepted to NAACL Findings 2024!
- [March 2024] Our work on ReLoRA accepted to ICLR 2024!
Publications
- Emergent Abilities in Reduced-Scale
Generative Language Models
S Muckatira, V Deshpande, V Lialin, A Rumshisky
NAACL Findings, 2024 - ReLoRA: High-Rank Training Through
Low-Rank Updates
V Lialin, S Muckatira, N Shivagunde, A Rumshisky
ICLR, 2024 - Deconstructing In-Context Learning:
Understanding Prompts via Corruption
N Shivagunde, V Lialin, S Muckatira, A Rumshisky
LREC-Coling, 2024 - Let's Reinforce Step by
Step
S Pan, V Lialin, S Muckatira, A Rumshisky
NeurIPS WANT Workshop, 2023 - Properties Of Winning Tickets On Skin
Lesion Classification
S Muckatira
ECCV WiCV Workshop, 2020 - Image-level
and group-level models for Drosophila gene expression pattern annotation
Q Sun, S Muckatira, L Yuan, S Ji, S Newfeld, S Kumar, J Ye
BMC bioinformatics
YouTube
I run a dedicated YouTube channel where I provide overviews on cutting-edge research papers. Topics I cover include: Prompting, Alignment, Distillation, Pre-Training, and LLM Robotics
Check out my YouTube channel and subscribe for regular updates!