معرفی
Karan Singh serves as an Assistant Professor of Operations Research at Carnegie Mellon University's Tepper School of Business, where he develops theoretically rigorous algorithms for machine learning systems with emphasis on reinforcement learning and control theory. His work synthesizes techniques from online learning, optimization, and statistics to address complex interactive learning challenges.
His academic journey includes:
- PhD in Computer Science from Princeton University under Elad Hazan
- Postdoctoral research at Microsoft Research (Redmond)
- Bachelor's degree in Computer Science from Indian Institute of Technology (IIT) Kanpur
Singh's research program centers on three interconnected pillars:
- Algorithmic Reductions: Creating efficient methods to solve complex learning problems (e.g., reinforcement learning) using solvers for simpler tasks, yielding breakthroughs in online boosting and RL with concave rewards
- Nonstochastic Control: Establishing an algorithmic foundation for control theory through provably efficient instance-optimal algorithms that extend online learning to stateful systems
- Privacy-Preserving Online Learning: Investigating fundamental limits of regret minimization under differential privacy constraints while maintaining performance
Analysis of his 15 most recent publications reveals a clear evolution toward integrating algorithmic reductions with nonstochastic control frameworks. Recent work (2023-2025) demonstrates increasing focus on sample efficiency in agnostic boosting, privacy-aware optimization without smoothness assumptions, and competitive ratio analysis in online control. A unifying thread is the development of regret-optimal algorithms for linear dynamical systems under adversarial disturbances.
His contributions have earned significant recognition:
- Best Paper Award at OptRL workshop (NeurIPS 2019)
- Spotlight Prize from New York Academy of Sciences' ML Symposium (2018)
- Multiple oral presentations at NeurIPS/ICML (acceptance rate <0.5%)
Though specific student advisees aren't listed, Singh's extensive publication record with junior co-authors indicates active mentorship. His research has secured substantial support including a US patent (11,138,513 B2) for dynamic learning systems and collaborations through CMU's Machine Learning and Optimization group. Current projects involve interdisciplinary work on differentiable control libraries (Deluca) and medical applications like mechanical ventilation control.
Singh leads research within CMU's Machine Learning and Optimization ecosystem, collaborating across computer science and engineering departments. His team develops foundational tools like the Deluca differentiable control library while pursuing real-world applications in healthcare systems, demonstrating strong cross-disciplinary integration.
