Sarthak Mittal
Papers
1
Total Citations
5
H-Index
1
About
Sarthak Mittal is a researcher at the intersection of causality, reinforcement learning, and representation learning. His work tackles a fundamental challenge in AI: how autonomous agents can discover causal structures directly from raw sensory data, without being given pre-defined causal variables. In his highly cited work, "Systematic Evaluation of Causal Discovery in Visual Model Based Reinforcement Learning" (2021), Mittal systematically investigates the integration of causal discovery into model-based RL, demonstrating how agents can learn to infer cause-effect relationships from visual observations alone. This research bridges the gap between classical causal inference—which assumes variables are given—and the messy, high-dimensional reality faced by embodied AI systems like robots. By showing that causal structures can be induced from pixels and used to improve planning and generalization, Mittal’s contributions have opened new pathways for building more interpretable and sample-efficient reinforcement learning agents. His work is foundational for researchers aiming to create AI that not only perceives the world but understands its underlying mechanisms.
Research Focus
Key Achievements
Top Papers
- 1