Research

← Home

During my PhD, my research focused on techniques for learning with limited labeled data - unsupervised learning, few-shot learning, and semi-supervised learning. Consequently, related areas such as data and model debiasing, spurious correlation mitigation, and model explainability also opened up as interesting research directions.

Recently, I have begun exploring more fundamental aspects of representation learning, such as sparsity and hierarchy in concept bottleneck models, and temporal reasoning in video-language models.