Publications

2026

1. WWW

   EcoThink: A Green Adaptive Inference Framework for Sustainable and Accessible Agents

   Linxiao Li^* and Zhixiang Lu^*†

   In Proceedings of the ACM Web Conference (WWW), 2026

   Oral (Selection Rate: 4.6%) Abs Bib PDF

   EcoThink is a green adaptive inference framework that addresses LLM overthinking in agent systems. It achieves an average of 40.4% energy savings (up to 81.9%) without significant performance loss, making AI agents more sustainable and accessible.

   @inproceedings{li2026ecothink,
     title = {EcoThink: A Green Adaptive Inference Framework for Sustainable and Accessible Agents},
     author = {Li, Linxiao and Lu, Zhixiang},
     booktitle = {Proceedings of the ACM Web Conference (WWW)},
     year = {2026},
     url = {https://dl.acm.org/doi/10.1145/3774904.3792995}
   }

2. ICML

   Deterministic Component Mining for Multi-framework UI2Code Generation

   Zixiong Yang^*, Linxiao Li^*, and 4 more authors

   In Proceedings of the International Conference on Machine Learning (ICML), 2026

   Poster Abs Bib PDF

   We propose a deterministic component mining approach for multi-framework UI2Code generation, enabling accurate and framework-agnostic code synthesis from UI designs.

   @inproceedings{yang2026dcm,
     title = {Deterministic Component Mining for Multi-framework UI2Code Generation},
     author = {Yang, Zixiong and Li, Linxiao and Lin, Jiaye and Wu, Binrui and Kang, Xiaoyu and Gao, Jiechao},
     booktitle = {Proceedings of the International Conference on Machine Learning (ICML)},
     year = {2026},
     url = {https://icml.cc/virtual/2026/poster/64878}
   }

3. arXiv

   Zeta: Dual Whitening for Matrix Optimization via Coordinate-Adaptive Preconditioning

   Kaiwen Chen^*, Shuhai Zhang^*, and 8 more authors

   arXiv preprint arXiv:2606.14187, 2026

   Abs arXiv Bib PDF Code

   Zeta is a dual whitening optimizer for matrix-aware neural network training. It combines coordinate whitening and spectral whitening to address coordinate-wise scale imbalance in momentum matrices, improving conditioning and orthogonalization reliability across language modeling, MoE, and vision workloads.

   @article{chen2026zeta,
     title = {Zeta: Dual Whitening for Matrix Optimization via Coordinate-Adaptive Preconditioning},
     author = {Chen, Kaiwen and Zhang, Shuhai and Liu, Zimo and Li, Linxiao and Sun, Ying and Li, Yuchen and Zhang, Yifan and Han, Bo and Tan, Mingkui and Chen, Qiuwu},
     journal = {arXiv preprint arXiv:2606.14187},
     year = {2026},
     url = {https://arxiv.org/abs/2606.14187v2}
   }

4. NeurIPS

   Toward Executable Multi-framework Front-end Code Generation with Self-Correction

   Linxiao Li^*, Haoran Ma^*, and 4 more authors

   In Advances in Neural Information Processing Systems (NeurIPS), 2026

   Poster Abs Bib Code

   Generating codes from UI screenshots has recently benefited from MLLMs, yet existing methods centered on HTML targets and transfer poorly to multi-frameworks such as React, Vue, and Angular. However, unlike HTML, which does not entail compilation issues, generating executable code in multi-framework settings is more challenging due to framework-specific differences in syntax and failure modes, as well as the need to maintain global consistency across files. To address this problem, we study executable multi-framework front-end code generation and propose MESCoder, a tree-structured self-corrective generation framework. Our method first extracts a structural scaffold from the screenshot and then constructs sub-trees and assembly relations step by step, transforming large project generation into a sequence of localized decisions. On top of this scaffold, we introduce a unified Project Agent that continuously revises its own partial project under runtime sandbox feedback and dynamically chooses between local repair and cluster level repair. For training, we adopt a two-stage strategy that first initializes the policy offline to learn basic multi-framework generation and repair priors, and then refines it with online reinforcement learning to optimize long horizon executability and final page quality. The resulting framework models multi-framework executability as a conditioned project state transition process and improves the stability and quality of complex front end generation through self correction. Experiments show that our method improves executability, while narrowing the visual-fidelity gap to HTML-based generation.

   @inproceedings{li2026selfcorrection,
     title = {Toward Executable Multi-framework Front-end Code Generation with Self-Correction},
     author = {Li, Linxiao and Ma, Haoran and Wang, Chenyue and Lin, Jiaye and Sui, Haochen and Gao, Jiechao},
     booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
     year = {2026},
   }

5. NeurIPS

   Phase-wise MLLM Tuning for Multi-framework WebUI Code Generation

   Haoran Ma^*, Linxiao Li^*, and 3 more authors

   In Advances in Neural Information Processing Systems (NeurIPS), 2026

   Poster Abs Bib

   Multimodal large language models (MLLMs) achieve strong performance in translating WebUI screenshots into HTML/CSS. However, when they face multiple frontend frameworks (React/Vue/Angular), they often suffer from negative transfer that leads to compilation failures. These failures commonly stem from mixing framework specific syntax and violating framework constraints. This paper studies supervised fine tuning for multi-framework WebUI code generation and proposes a phase-wise MLLM tuning method driven by compatibility and heterogeneity signals across frameworks. We first apply a normalization mapping that preserves each framework’s native syntax while suppressing noise from variable naming and literal content. We then estimate cross framework heterogeneity and compatibility offline, and automatically derive training groups and their ordering by maximizing a phased objective. Finally, we perform phase-wise adapter tuning to reduce cross framework interference. Experiments on multiple benchmarks show that the proposed method improves both generation quality and compilation success rates compared with baselines.

   @inproceedings{ma2026phasewise,
     title = {Phase-wise {MLLM} Tuning for Multi-framework {WebUI} Code Generation},
     author = {Ma, Haoran and Li, Linxiao and Wang, Chenyue and Sui, Haochen and Gao, Jiechao},
     booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
     year = {2026},
   }

6. NeurIPS

   Tree Rotary Positional Encoding for Extreme Length Extrapolation from Scratch

   Qiuwu Chen, Ziteng Huang, and 8 more authors

   In Advances in Neural Information Processing Systems (NeurIPS), 2026

   Poster Abs Bib

   Large language models rely on positional representations for long context modeling, yet are typically trained on short sequences and evaluated at much longer lengths. Bridging this gap requires positional encodings remaining distinguishable at large distances while generalizing beyond the training length. In this paper, we identify a structural limitation of existing rotary-based positional encodings: all positional dimensions are controlled by a single global index, which couples training-time positional variation with long-range extrapolation. To address this, we propose Tree Rotary Positional Encoding (TPE), a structured multi-scale positional encoding designed for training from scratch that represents absolute position through combinations of different components, avoiding single-index extrapolation. We further devise an Offset-based Positional Training (OPT) strategy to improve learning interactions across TPE components, enabling better generalization to longer contexts. We theoretically analyze that its multi-scale decomposition enables effective combinatorial position coverage. Experimental results show strong gains under extreme length extrapolation: with a 1.2B model trained from scratch on length 512, our TPE achieves a 64× extrapolation to 32K tokens, reducing perplexity from 4532.21 (RoPE) to 149.96, with comparable inference-time efficiency.

   @inproceedings{chen2026treerope,
     title = {Tree Rotary Positional Encoding for Extreme Length Extrapolation from Scratch},
     author = {Chen, Qiuwu and Huang, Ziteng and Zhang, Shuhai and Liu, Zimo and Li, Linxiao and Sun, Ying and Li, Yuchen and Zhang, Yifan and Chen, Yaofo and Tan, Mingkui},
     booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
     year = {2026},
   }