Anisha Gunjal

I am a graduate student in Computer Science at UT Austin advised by Prof. Greg Durrett at TAUR Lab. My research focuses on fact verification of text generated by large language models. This summer, I researched with the Foundational ML team of Scale AI where I worked on multimodal RLHF aiming to mitigate hallucinations and enhance fidelity in large vision-language models.

Prior to this, I was a Visiting Scholar at Cognitive Learning for Vision and Robotics Lab (CLVR) at University of Southern California, advised by Prof. Joseph Lim. I also worked as the Lead Machine Learning Engineer of Documents AI at HyperVerge, Inc. I did my undergrad in Computer Science at Pune Institute of Computer Technology.

Beyond the realms of research, I love globetrotting, salsa dancing and capturing the adventures of my cannine companions!

Email  /  CV  /  Scholar  /  Twitter  /  Github  /  LinkedIn  /  Blog

profile photo
May 2023 Machine Learning Research Intern at Scale AI
Aug 2022 Started MS in Computer Science at UT Austin
Jan 2022 Task-Induced Representation Learning accepted at ICLR 2022
Dec 2021 Poster Presentation at Deep RL Workshop, NeurIPS 2021
Aug 2021 Participant at 5th Summer School on Artificial Intelligence, IIIT Hyderabad
Jul 2021 Volunteer at WiML Workshop @ ICML 2021
Jun 2021 Joined CLVR Lab at USC as a Visiting Researcher
Sep 2020 Published a blogpost The ECCV Experience
Jul 2020 Reviewer at JupyterCon2020
Jul 2020 Published blogpost Document Visual Question Answering
Jun 2020 4th Position in DocVQA Challenge at CVPR 2020

My research focus revolves around ensuring that machine-generated text maintains fidelity, accuracy, and a strong connection between visual and language elements. I address these challenges through investigations into fact verification, Reinforcement Learning with human feedback (RLHF), and the effective grounding of text in visual contexts.

Detecting and Preventing Hallucinations in Large Vision Language Models
Anisha Gunjal, Jihan Yin, Erhan Bas
arxiv preprint 2023

Reducing Hallucinations in LVLMs using a novel benchmark dataset M-HalDetect which is used to train fine-grained reward models capable of detecting unfaithful text generations for a given image context.

Drafting Event Schemas using Language Models
Anisha Gunjal, Greg Durrett
arxiv preprint 2023

Generating and automatically evaluating lightly structures complex event schemas using large language models.

Task-Induced Representation Learning
Jun Yamada, Karl Pertsch, Anisha Gunjal, Joseph Lim
ICLR 2022
project page / paper

Investigation of using task information for learning representations for RL in visually complex scenes.

Document Visual Question Answering
Anisha Gunjal, Vipul Gupta, Moinak Bhattacharya, Digvijay Singh
CVPR 2020, Leaderboard Rank: 4
code / blog / workshop

Joint modeling of text and layout information using transformers for Visual Question Answering on unstructured documents.

Diabetic Retinopathy Grading using Deep Siamese Network
Anisha Gunjal
ICML 2018, Poster Presentation
paper / workshop

Contrastive Learning on retinal images for determining the stage of Diabetic Retinopathy disease progression on a small sized medical image dataset.

  • Document Visual Question Answering
  • The ECCV Experience
  • Listicles
  • Awesome Resource List for NeuroSymbolic Visual Reasoning

  • Clone this to overcome good website design FOMO!