I am a Ph.D. candidate at Mila, co-advised by
Prof. Sarath Chandar and
Prof. Ross Goroshin.
My research interests are continual learning, lifelong learning,
and streaming reinforcement learning.
Previously, I completed my Master's at IIIT Hyderabad,
co-advised by Prof. Vineet Gandhi and
Prof. K. Madhava Krishna,
working on visual grounding, language-guided navigation, and multi-view perception.
An RL framework that trains a questioner agent to generate increasingly complex queries that automatically expose weaknesses in vision-language models without human intervention.
A data-efficient training framework for VLMs that dynamically selects samples based on the model's learning propensity.
CulturalVQA: a benchmark for evaluating VLMs' understanding of geo-diverse cultural elements, revealing performance disparities across regions and facets.
Hierarchical Ensembles (HiE) — a post-hoc strategy that leverages label hierarchy to reduce mistake severity in fine-grained classification at test time.
An instance-focused scene representation enabling language-based navigation that grounds references to specific instances within an environment.
Vision-and-Language Navigation for outdoor autonomous driving — explicitly grounds navigable regions from text and uses them as guidance for the navigation stack.
Existing multi-view detectors overfit to single scenes — we formalize three critical forms of generalization and propose evaluations for them.
A Transformer-based architecture capturing all forms of multi-modal interaction synchronously for referring image segmentation.
A visual-grounding-based approach to language-guided navigation that brings interpretability to Vision-Language Navigation.