Advancing Capabilities: How do we build machines that assist us to better perceive, understand, and reason about the real world?
problems I work on
Understanding: How do representations and training dynamics emerge, and what principles drive efficiency?
learning dynamics: (tracing representation geometry, α‑ReQ)
efficient vision pretraining: (harnessing small projectors)
Systems: How do we enable efficient inference for long‑context, multimodal models?
sparse algorithms for efficient inference: (vAttention)
Previously
- Google Brain – robotics, generative models, music synthesis, program synthesis
- IIT Kharagpur – mathematics and computer science
- MILA – with Yoshua Bengio