Chinmaya Kausik

• Mathematical problems motivated by tangible, real-world questions. These days, my work focuses on sequential decision making under various settings - offline-to-online transfer, partial observability/latent information and non-standard feedback and reward models. I also have side projects in deep learning. On the other hand, a lot of my undergraduate background was in geometry, topology and dynamics, with work in computer-assisted topology and geometry.
• Increasing accessibility to and in higher mathematics and creating communities where ideas cross pollinate and people pull each other up. I have started the Stats, Physics, Astronomy, Math (SPAM) graduate student social initiative at the University of Michigan. I also co-founded and co-organize Monsoon Math Camp. I have also been involved in building and expanding other mathematical communities, like platforms for the PolyMath REU, DRP programs and the undergraduate math organization at IISc, etc.

• Collaborating with Yonathan Efroni (Meta), Aadirupa Saha (Apple), Nadav Merlis (ENSEA) on algorithms for bandit and reinforcement learning algorithms with feedback at varying costs and accuracies, also called multi-fidelity feedback.
• Working on unifying various reward and problem frameworks in reinforcement learning and bandits with Aldo Pacchiano (Broad Institute of MIT and Harvard) and Mirco Mutti (Politecnico di Milano)
• Designing optimal algorithms for offline-to-online transfer in latent bandits with Kevin Tan (University of Pennsylvania).
• Extending my work with Rishi Sonthalia (UCLA) and Kashvi Srivastava (UMich) to more complex denoising models.
• Working on algorithms for offline policy evaluation (OPE) in linear bandits, and the role of the geometry of action sets.
• Continuing work on our project from LOGML 2022! I was a participant in Dr. Eli Meirom’s group, planning to work on using RL for graph rewiring in GNNs to prevent oversquashing for long range problems.
• Thinking about extensions of De Finetti’s theorem to decision processes.
• Organizing an interdepartmental social initiative, SPAM (Statistics, Physics, Astronomy, Mathematics).
• Fleshing out ideas for more academic communities like Monsoon Math.

primary goals

• Work on preference-based variants of sequential decision making problems.
• Using the multi-step inverse kinematics perspective for designing algorithms that work outside of Markovian assumptions and have strong empirical performance.
• Explore multi-objective decision making.
• Start maintaining my progress log again.
• Learn about safe RL and think about techniques beyond primal-dual ones, perhaps using model-based RL with uncertain models.
• Watch lectures from the Data Driven Decision Processes program at the Simons Institute this semester.

side-quests

• Causal inference and its interaction with sequential decision making and RL.
• Using insights from machine learning for biology. In a specific example, learning a hierarchical or causal structure from genomics.
• Dive deeper into the theory behind GNNs and deep learning in general.
• Algorithmic fairness.