Maharshi Gor

NLP Researcher | Engineer

profile_pic.png

Iribe 4108

8125 Paint Branch Dr

College Park, MD 20742

mgor[at]cs[dot]umd[dot]edu

I am a third year Computer Science PhD student at the University of Maryland, College Park, currently working with Tianyi Zhou and Jordan Boyd-Graber. I am a member of the Computational Linguistics and Information Processing (CLIP) Lab at UMIACS.

My interests span across the areas of Natural Language Understanding and Information Retrieval, with a focus on Question Answering (QA) and Interpretability. Currently, my research is concentrated in two primary areas:

  • Analyzing Large Language Model (LLM)-based agents in terms of their skills and knowledge, alongside the complexity of questions, to curate a minimal yet comprehensive set for profiling agents within a Mixture of Experts (MoE) framework.
  • Crafting Retrieval-Augmented Language Models that are both aware of their environment and their inherent knowledge to ensure reliable and useful output generation.

I’ve spent some time working at Google Research where I collaborated with Brain and several Language Research teams, with focus on Model Interpretation and Analysis for Question Answering, Semi-Structured Text Understanding and Retrieval-augmented Language models for long-context understanding.

In the past, I have also worked on some of the Computer vision and Machine Learning problems like Human motion sequence modeling, Generative and Representation Learning and Adversarial Machine Learning.

In my free time, I like to engage in social deception board games. I also very much like to play Magic the Gathering :)


🗞️ News

May 23, 2023 I will be spending my summer at Cohere as ML Research Intern, and will be working on Attribution evaluation for Retrieval-augmented Generative (RAG) models.
Sep 14, 2022 Our work “Toward Efficient Robust Training against Union of Lp Threat Models” is accepted at NeurIPS 2022 and also for oral presentation at ADVML FRONTIERS @ ICML 2022.
Sep 12, 2022 I am starting as part-time Student Researcher at Google and will primarily be working on Long Context Text Understanding.
May 23, 2022 I am spending my summer at X, the moonshot factory (formerly Google X) as PhD Research Resident, and will be working on Program Synthesis.
Nov 22, 2021 I will be serving as a Program Committee member for SUKI: Workshop on Structured and Unstructured Knowledge Integration and DADC: Workshop on Dynamic Adversarial Data Collection at NAACL 2022

📝 Selected Publications

  1. Do great minds think alike? Investigating Human-AI Complementarity for Question Answering
    Maharshi GorTianyi Zhou, III Daumé, and Jordan Boyd-Graber
    2024
  2. Toward Efficient Robust Training against Union of Lp Threat Models
    Gaurang Sriramanan, Maharshi Gor, and Soheil Feizi
    In Advances in Neural Information Processing Systems, 2022
  3. MATE: Multi-view Attention for Table Transformer Efficiency
    In Empirical Methods in Natural Language Processing, 2021
  4. Toward Deconfounding the Influence of Entity Demographics for Question Answering Accuracy
    Maharshi GorKellie Webster, and Jordan Boyd-Graber
    In Empirical Methods in Natural Language Processing, Nov 2021