Research Areas
Federated Learning
The lab has made breakthrough progress in both the theory and application of federated learning. We have tackled key algorithmic challenges in client optimization, fairness guarantee, and privacy protection, including heterogeneous data client clustering, diverse sampling, and federated forgetting, promoting innovative applications of federated learning in cross-domain recommendation and heterogeneous data distribution scenarios. Multiple results have been accepted by top international conferences such as ICML, NeurIPS, AAAI, and leading journals including IEEE TKDE.
Large Models (Reasoning, MOE, Multimodal)
The lab has released five flagship works in large model architectures and cross-modal spatiotemporal reasoning: DynMoE (ICLR 2025) and TRACE (ICLR 2025) achieve dynamic MoE tuning and causal event chain long-video grounding, respectively; VTG-LLM (AAAI 2025) integrates timestamp knowledge into video LLMs, setting new zero-shot benchmarks; NLPrompt (CVPR 2025) proposes two key techniques, PromptMAE and PromptOT, significantly improving the robustness of vision-language models in prompt learning; Difficult Task Yes but Simple Task No (EMNLP 2024 Findings) constructs the LazyBench benchmark, revealing the "laziness" phenomenon of multimodal large models on simple tasks and preliminarily mitigating it via chain-of-thought reasoning. Multiple papers have been accepted by ICLR, AAAI, CVPR, and other top conferences. We welcome you to join us in pushing the boundaries of intelligence!
Intelligent EV Charging Optimization
The lab focuses on addressing charging system optimization challenges in the context of transportation electrification, with the mission to build an intelligent and efficient charging service ecosystem. Through key technological breakthroughs including EV user behavior modeling, scientific charging station siting, dynamic scheduling algorithm optimization, and charging service pricing strategy research, we systematically tackle core challenges arising from transportation electrification: ensuring grid stability under large-scale charging loads, mitigating traffic congestion caused by surging charging demand, and establishing a sustainable development framework that balances charging station profitability with user charging experience.