Research
Broadly, I am interested in improving the sample efficiency and out of distribution generalization abilities of Deep Learning models.
To this end, I have been recently working on investigating the use of generative models for generating high quality synthetic data for downstream use cases such as training new models and the evaluation of existing models.
I am also interested in and have previously worked on the several ideas related to reinforcement learning, and developing deep learning approaches based on human cognitive inductive biases.
|
AI-Assisted Generation of Difficult Math Questions
Vedant Shah,
Dingli Yu,
Kaifeng Lyu,
Simon Park,
Nan Rosemary Ke,
Michael Mozer,
Yoshua Bengio,
Sanjeev Arora,
Anirudh Goyal
MATH-AI Workshop, NeurIPS 2024
arXiv
We use frontier LLMs such as GPT-4 and Claude 3 to generate challenging mathematical questions by asking them to compose two domain skills at once.
Based on this, we present a new math eval where a large number of open source as well as proprietary LLMs show significant drops relative to a standard eval.
|
Towards DNA-Encoded Library Generation Using GFlowNets
Michał Koziarski,
Mohammed Abukalam,
Vedant Shah,
Louis Vaillancourt,
Doris Alexandra Schuetz,
Moksh Jain,
Almer van der Sloot,
Mathieu Bourgey,
Anne Marinier,
Yoshua Bengio
GEM Workshop, ICLR 2024
arXiv
We use PPI modulation task based reward models to train GFlowNets to address the combinatorially challenging task of generating DNA Encoded Libraries (DELs)
|
Efficient Causal Graph Discovery Using Large Language Models
Thomas Jiralerspong,
Xiaoyin Chen,
Yash More,
Vedant Shah,
Yoshua Bengio
How Far are We From AGI? Workshop, ICLR 2024
arXiv
We propose an approach to discover full causal graphs in real-world settings using LLMs such at GPT-4 by augmenting LLM querying with a BFS like procedure.
|
Unlearning via Sparse Representations
Vedant Shah,
Frederik Träuble,
Ashish Malik,
Hugo Larochelle,
Michael Mozer,
Sanjeev Arora,
Yoshua Bengio,
Anirudh Goyal,
Preprint
arXiv
We propose a nearly compute-free approach for class unlearning in multi-class classification settings.
|
Stateful Active Facilitator: Coordination and Environmental Heterogeneity in Cooperative Multi-Agent Reinforcement Learning
Dianbo Liu*,
Vedant Shah*,
Oussama Boussif*,
Cristian Meo,
Anirudh Goyal,
Tianmin Shu,
Michael Mozer,
Nicolas Heess,
Yoshua Bengio,
ICLR 2023
arXiv /
Talk
Investigating the problem of environmental heterogeneity and using behvioural priors to tackle the problems of coordination and heterogeneity in MARL
|
Neural Feature-Adaptation for Symbolic Predictions Using Pre-Training and
Semantic Loss
Vedant Shah*,
Aditya Agarwal*,
Lovekesh Vig,
Ashwin Srinivasan,
Gautam Shroff,
Tanmay Verlekar,
Preprint
arXiv
Using corrupted pretraining and semantic loss for obtaining explainable predictions in terms of human intelligible concepts by composing
a symbolic and a neural module together.
|
Forecasting Market Prices using DL with Data Augmentation and Meta-learning: ARIMA still wins!
Vedant Shah,
Gautam Shroff
I (Still) Can't Believe It's Not Better! Workshop, NeurIPS 2021
arXiv /
Poster
We train state of the art deep learning models on financial data using a data augmentation with synthetic data and meta learning.
ARIMA still beats the deep learning models in forecasting market data.
|
Adapting Deep Neural Networks for Pedestrian-Detection to Low-Light
Conditions without Re-training
Vedant Shah,
Anmol Agarwal,
Tanmay Tulsidas Verlekar,
Raghavendra Singh
Workshop on Traditional Computer Vision in the Age of Deep Learning, ICCV 2021
PDF /
Slides
We develop a real time pre-processing pipeline for helping deep neural networks in pedestrian detection in low-light conditions.
|
|
Sep 2023 - Present
Graduate Student Researcher
Advisor - Dr. Anirudh Goyal and Prof. Yoshua Bengio
Sep 2021 - July 2023
Research Intern
Advisors - Dr. Anirudh Goyal and Prof. Yoshua Bengio
|
|
Jan 2022 - Aug 2022
Undergraduate Researcher
Advisor - Prof. Ashwin Srinivasan
Jan 2021 - Aug 2021
Undergraduate Researcher
Advisor - Prof. Tanmay Verlekar
|
|
May 2021 - Sep 2021
Research Intern
Advisor - Dr. Gautam Shroff
|
|
May 2021 - Aug 2021
Student Developer
Organisation - GFOSS (Suborganisation - OpenDR)
|
|
Aug. 2020 - Nov. 2020
Research Intern
Advisor - Prof. GC Nandi
|
|