Skip navigation
maChine Learning and quANtum computing research (CLAN)

Research Overview

The four current areas of research in CLAN Labs are listed below. Each theme links to the topic page with the corresponding publications.

Reinforcement Learning and Sequential Decision-Making

We develop reinforcement learning, bandit, and online optimization algorithms with provable sample-complexity and regret guarantees. The current themes are:

  1. Order-optimal sample complexity and regret for discounted-reward and average-reward reinforcement learning
  2. Policy gradient and actor-critic methods with general (neural) policy parameterization
  3. Constrained, concave-utility, multi-objective, and robust reinforcement learning
  4. Multi-armed bandits: combinatorial, submodular, and continuous-action bandits
  5. Online linearizable optimization and DR-submodular maximization
  6. Stochastic approximation with Markovian noise and nonconvex optimization

Foundation Models and Generative AI

We study how to align large models with human preferences and what guarantees generative models can achieve. The current themes are:

  1. Bi-level reinforcement learning as a foundation for reinforcement learning from human feedback (RLHF)
  2. Online and robust preference alignment of large language models
  3. Reliable evaluation of adaptive LLM systems and agents
  4. Sample-complexity guarantees for diffusion models and flow matching
  5. Alignment of text-to-image diffusion models

Quantum Machine Learning

We design quantum algorithms for learning and decision-making, and the tools to train and simulate them at scale. The current themes are:

  1. Quantum speedups for reinforcement learning and bandits
  2. Variational quantum circuits, quantum neural networks, and hybrid quantum-classical models
  3. Quantum search (Grover's search) and quantum optimization
  4. Quantum causal inference
  5. Quantum-accelerated imaging

Applications of Machine Learning

We apply these methods to problems in healthcare, imaging, and networked systems. The current themes are:

  1. Generative and foundation models for drug discovery and genomics
  2. Clinical decision support (e.g., sepsis) and video/sensor-based ergonomic risk monitoring
  3. Low-photon imaging via machine learning and physics-based optimization
  4. AI-native networking: service function chains, intent-based control, and learning-augmented traffic engineering