Efficient & Reliable ML
Language and generative models: compression, inference, reliability, and representations.
- Compression & inference. LowRankArena, FlashSVD, and BudgetDraft study compression quality, practical speed, and efficient decoding.
- Reliability & agents. DPQ preserves uncertainty after quantization; related work examines agent decisions and execution utility.
- Representations. DecodeShare studies shared decoding subspaces, alongside visual generation and graph community learning.
Publications & preprints (13)
Compression & efficient inference
The Functional Structure of Post-Compression Recovery in Low-Rank LLMs
Preprint · 2026PreprintPDF
BibTeX
@misc{shao2026functional,
author = {Shao, Zishan and
Tian, Liang and
Zemlevskiy, Georgiy and
Cui, Kangning and
Zhang, Lixun and
Wang, Yixiao and
Jiang, Ting and
Kim, Jinhee and
Chen, Yixuan and
Wang, Rui-Feng and
Yang, Fan and
Li, Hai and
Chen, Yiran},
title = {The Functional Structure of Post-Compression Recovery in Low-Rank {LLMs}},
year = {2026},
url = {https://arxiv.org/abs/2610.05504v1},
eprint = {2610.05504},
archiveprefix = {arXiv},
version = {1},
note = {arXiv preprint, version 1},
}Joint Branch-Space Transform Coding for Diffusion Activation Quantization with Classifier-Free Guidance
Preprint · 2026PreprintPDF
BibTeX
@misc{jiang2026joint,
author = {Jiang, Mingrun and
Liu, Yuejia and
Shao, Zishan and
Jiang, Ting and
Wang, Qinsi and
Ye, Hancheng and
Wang, Yixiao and
Wang, Rui-Feng and
Cui, Kangning and
Chen, Yixuan and
Yang, Fan and
Cheng, Xiang and
Li, Hai and
Chen, Yiran},
title = {Joint Branch-Space Transform Coding for Diffusion Activation Quantization with Classifier-Free Guidance},
year = {2026},
url = {https://arxiv.org/abs/2610.00930v1},
eprint = {2610.00930},
archiveprefix = {arXiv},
version = {1},
note = {arXiv preprint, version 1},
}You Cannot Pick a Provider From the Price List: Market-Aware Routing for Open-Weight LLM Inference
Preprint · 2026PreprintPDF
BibTeX
@misc{he2026provider,
author = {He, Liang and
Wen, Jingbo and
Chen, Yixiong and
Yang, Yue and
Lan, Qizhen and
Cui, Kangning and
Wang, Xilu},
title = {You Cannot Pick a Provider From the Price List: Market-Aware Routing for Open-Weight {LLM} Inference},
year = {2026},
url = {https://arxiv.org/abs/2609.37902v1},
eprint = {2609.37902},
archiveprefix = {arXiv},
version = {1},
note = {arXiv preprint, version 1},
}NeurIPS'26, ED Track, PosterPaperPreprintCodePDF (preprint)
BibTeX
@inproceedings{shao2026lowrankarena,
author = {Shao, Zishan and
Zhang, Lixun and
Cui, Kangning and
Wu, Wenhao and
Kim, Jinhee and
Wang, Yixiao and
Jiang, Ting and
Ye, Hancheng and
Wang, Qinsi and
Yang, Fan and
Zhuo, Danyang and
Li, Hai and
Chen, Yiran},
title = {{LowRankArena}: A Standardized Evaluation Platform for {SVD}-Based {LLM} Compression},
booktitle = {The Fortieth Annual Conference on Neural Information Processing Systems},
year = {2026},
url = {https://neurips.cc/virtual/2026/poster/139091},
note = {Accepted conference paper, Evaluations and Datasets Track},
}Not All Visual Tokens Are Equally Safe to Remove: Consequence-Sensitive Visual Token Compression
Preprint · 2026PreprintPDF
BibTeX
@misc{wen2026consequence,
author = {Wen, Jingbo and
He, Liang and
Cao, Mingyu and
Wang, Haoyu and
Hu, Minxuan and
Cui, Kangning and
Wang, Xilu},
title = {Not All Visual Tokens Are Equally Safe to Remove: Consequence-Sensitive Visual Token Compression},
year = {2026},
url = {https://arxiv.org/abs/2608.09176v1},
eprint = {2608.09176},
archiveprefix = {arXiv},
version = {1},
note = {arXiv preprint, version 1},
}BudgetDraft: Acceptance-Aware Multi-View Training for Sparse-KV Speculative Decoding
Preprint · 2026PreprintPDF
BibTeX
@misc{he2026budgetdraft,
author = {He, Liang and
Wen, Jingbo and
Zhan, Qishi and
Chen, Yixiong and
Cui, Kangning and
Lan, Qizhen and
Wang, Xilu},
title = {{BudgetDraft}: Acceptance-Aware Multi-View Training for Sparse-{KV} Speculative Decoding},
year = {2026},
url = {https://arxiv.org/abs/2606.00144v1},
eprint = {2606.00144},
archiveprefix = {arXiv},
version = {1},
note = {arXiv preprint, version 1},
}FlashSVD v1.5: Making Low-Rank Transformers Inference Actually Fast
Preprint · 2026PreprintPDFCode
BibTeX
@misc{wu2026flashsvd,
author = {Wu, Wenhao and
Shao, Zishan and
Cui, Kangning and
Kim, Jinhee and
Wang, Yixiao and
Ye, Hancheng and
Zhuo, Danyang and
Chen, Yiran},
title = {{FlashSVD} v1.5: Making Low-Rank Transformers Inference Actually Fast},
year = {2026},
url = {https://arxiv.org/abs/2605.08314v1},
eprint = {2605.08314},
archiveprefix = {arXiv},
version = {1},
note = {arXiv preprint, version 1},
}Reliability, agents & representations
EMNLP'26, FindingsPaperPreprintCodePDF
BibTeX
@inproceedings{yang-etal-2026-target,
author = {Yang, Zhen and
Hou, Sizai and
Zheng, Kaiwen and
Liu, Yaofang and
He, Liang and
Chen, Yixuan and
Cui, Kangning},
title = {Target-Aware Calibration Data Selection for Preserving Uncertainty in Quantized Language Models},
booktitle = {Findings of the Association for Computational Linguistics: {EMNLP} 2026},
editor = {Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond},
year = {2026},
pages = {14073--14093},
publisher = {Association for Computational Linguistics},
address = {Budapest, Hungary},
month = {October},
isbn = {979-8-89176-462-0},
url = {https://aclanthology.org/2026.findings-emnlp.670/},
}From Relevance to Execution Utility: Reward-Aware Dynamic Execution Gating for Skill-Based LLM Agents
Preprint · 2026PreprintPDF
BibTeX
@misc{he2026execution,
author = {He, Liang and
Wen, Jingbo and
Gu, Hongyu and
Li, Hao and
Wang, Haoyu and
Chen, Yixiong and
Cui, Kangning and
Wang, Xilu},
title = {From Relevance to Execution Utility: Reward-Aware Dynamic Execution Gating for Skill-Based {LLM} Agents},
year = {2026},
url = {https://arxiv.org/abs/2608.09168v1},
eprint = {2608.09168},
archiveprefix = {arXiv},
version = {1},
note = {arXiv preprint, version 1},
}Auditing Provenance Sensitivity in LLM Agent Action Selection
Preprint · 2026PreprintPDF
BibTeX
@misc{liao2026auditing,
author = {Liao, Junchi and
Huang, Qiang and
Qian, Rui and
Mao, Xutao and
Cheng, Juntao and
Cui, Kangning and
Li, Yu and
Sun, Jingyu and
Hou, Yuchao},
title = {Auditing Provenance Sensitivity in {LLM} Agent Action Selection},
year = {2026},
url = {https://arxiv.org/abs/2607.20827v1},
eprint = {2607.20827},
archiveprefix = {arXiv},
version = {1},
note = {arXiv preprint, version 1},
}ICML'26, SpotlightPreprintPaperPDFCode
BibTeX
@inproceedings{pmlr-v306-shao26j,
author = {Shao, Zishan and
Zhang, Lixun and
Cui, Kangning and
Wang, Yixiao and
Jiang, Ting and
Ye, Hancheng and
Wang, Qinsi and
Du, Zhixu and
Fu, Yuzhe and
Yang, Fan and
Zhuo, Danyang and
Chen, Yiran and
Li, Hai Helen},
title = {{DecodeShare}: Tracing the Shared Subspace of {LLM} Decode-Time Decisions},
booktitle = {Proceedings of the 43rd International Conference on Machine Learning},
editor = {Zhang, Tong and
Dudik, Miroslav and
Jaggi, Martin and
Agarwal, Alekh and
Li, Sharon and
Schuurmans, Dale and
Zhu, Jerry and
Berkenkamp, Felix and
Dong, Hanze and
Bietti, Alberto},
year = {2026},
volume = {306},
pages = {110338--110373},
series = {Proceedings of Machine Learning Research},
publisher = {PMLR},
month = {July},
url = {https://proceedings.mlr.press/v306/shao26j.html},
}Beyond Text Prompts: Visual-to-Visual Generation as A Unified Paradigm
Preprint · 2026PreprintPDFProject
BibTeX
@misc{liu2026visual,
author = {Liu, Yaofang and
Cui, Kangning and
Chu, Meng and
Li, Zhaoqing and
Zhang, Suiyun and
Morel, Jean-Michel and
Cun, Xiaodong and
Che, Haoxuan and
Liu, Rui and
Chan, Raymond H.},
title = {Beyond Text Prompts: Visual-to-Visual Generation as A Unified Paradigm},
year = {2026},
url = {https://arxiv.org/abs/2605.12271v2},
eprint = {2605.12271},
archiveprefix = {arXiv},
version = {2},
note = {arXiv preprint, version 2},
}Structured learning
A Local Perspective-based Model for Overlapping Community Detection
IEEE SMC'25PaperPreprint
BibTeX
@inproceedings{zhou2025community,
author = {Zhou, Gaofeng and
Wang, Rui-Feng and
Cui, Kangning},
title = {A Local Perspective-based Model for Overlapping Community Detection},
booktitle = {2025 {IEEE} International Conference on Systems, Man, and Cybernetics ({SMC})},
year = {2025},
pages = {2965--2970},
publisher = {IEEE},
month = {October},
doi = {10.1109/smc58881.2025.11343612},
url = {https://doi.org/10.1109/smc58881.2025.11343612},
}
