2025: 'RoSTE: An Efficient Quantization-Aware Supervised Fine-Tuning Approach for Large Language Models' – proposes rotated quantization as bias reduction, achieving SOTA performance in 4-bit weight-activation quantized LLM fine-tuning
2025: 'A Stochastic Approximation Approach for Efficient Decentralized Optimization on Random Networks' – introduces the first asynchronous primal-dual decentralized algorithm on random graphs with sparse communication
2025: 'Decentralized Stochastic Optimization over Unreliable Networks via Two-timescales Updates' – first algorithm supporting nonlinear noisy compression (e.g., quantization) on random gossip graphs
2024: 'EMC²: Efficient MCMC Negative Sampling for Contrastive Learning with Global Convergence' – applies MCMC to optimize global contrastive loss, outperforming baselines with small batch sizes
2024: 'A Two-timescale Primal-dual Algorithm for Decentralized Optimization with Compression' – analyzes convergence of compressed decentralized optimization with contractive compressors on nonconvex objectives
2023: 'Fully Stochastic Distributed Convex Optimization on Time-Varying Graph with Compression' – algorithm supporting time-varying graphs, compression, and asynchronous updates via primal-dual framework
2023: 'Network Effects in Performative Prediction Games' – studies equilibrium existence in networked performative prediction with dual-graph structure
2023: 'DoCoM: Compressed Decentralized Optimization with Near-Optimal Sample Complexity' – gradient-tracking-based algorithm supporting sparsification and quantization with near-optimal sample complexity