Welcome to T-lab

T-lab proposes G²RPO-A, an adaptive-guidance variant of Group Relative Policy Optimization that injects structured external guidance into small-model RL training, substantially improving math and reasoning performance where vanilla GRPO plateaus. The paper is accepted to ACL 2026 Main Conference.

T-lab develops an economic-analysis framework for the optimal timing of upgrading regular charging stations to fast chargers, jointly modeling demand growth and upgrade cost to quantify the best decision window for operators under varying grid and traffic conditions. Published in IEEE Transactions on Transportation Electrification (TTE).

T-lab proposes a bounded rationality Bayesian game framework, first applying prospect theory to EV highway charging decisions. Compared to traditional models, total social cost drops 44%, station daily procurement cost falls 29%, and peak queue length reduces up to 4x. Published in IEEE IoTJ (JCR Q1, IF 9.6).

Proposed by Tencent AI Lab with CUHK-Shenzhen, AutoDeco equips Transformers with lightweight heads to dynamically predict temperature and top-p per token, and introduces a differentiable soft top-p to enable end-to-end training. It achieves SOTA-level results across models/tasks with ~1.7% latency overhead, and shows early signs of natural-language controllable decoding. See coverage for details.

T-lab's research advances vehicle-to-grid synergy, enabling smart interactions between electric vehicles and the power grid. See the China News story for details.

T-lab's research achievements in battery swap station pricing strategy and charging optimization have garnered industry attention. The related paper was published in IEEE Transactions on Mobile Computing (TMC) journal, providing new insights for theoretical research and practical applications in the battery swap station market.

Shenzhen TV interviewed Prof. Xiaoying Tang, who discussed how Shenzhen's electricity consumption growth is closely related to the development of new quality productive forces. The construction of industrial computing centers, data centers, and R&D activities all require substantial power support, reflecting Shenzhen's rapid development in digital transformation.

The university's official WeChat account featured Prof. Xiaoying Tang's contributions in fundamental algorithm innovation. As the thirteenth scholar in the AI Rising Star series, the report showcases her exploration and practice in multi-domain applications.

V2G system algorithms developed by the T-lab team have been applied in this landmark project, providing key technical support for large-scale vehicle-to-grid interaction in China. See full coverage at CCTV News.

Machine Heart reported on T-lab's latest research published at The Thirteenth International Conference on Learning Representations (ICLR 2025). The TRACE model, through causal event modeling, provides a new technical approach for temporal localization tasks in video understanding.

Prof. Xiaoying Tang was invited by Shenzhen Municipal Talent Work Bureau to participate in the production of "Come to Shenzhen, Be a Shenzhener" as one of the young talent representatives from Shenzhen universities.