Research
Optimization for efficient and reliable learning
I develop and analyze optimization algorithms for learning under communication, privacy, robustness, and memory constraints. My work connects convergence guarantees with distributed training, adaptive optimizers, and parameter-efficient fine-tuning.
Research directions
Communication-Efficient Federated Learning
- Local training
- Compression
- Client sampling
Learning across devices with fewer communication rounds and smaller messages.
Explore this topicPrivacy and Robustness
- Differential privacy
- Byzantine robustness
- Clipping
Protecting training data and learning reliably in the presence of faulty or malicious participants.
Explore this topicOptimization for Deep Learning
- Adaptive optimizers
- Low-rank adaptation
- Sparse training
Understanding and improving optimizers for training and adapting neural networks.
Explore this topicAcross these directions, I study the foundations of stochastic optimization, including convergence theory, variance reduction, and generalized smoothness.