AI Systems Research

Qianli Ma

Portrait of Qianli Ma

I work on efficient LLM systems, semantic state transfer, multi-candidate reasoning, and security-oriented machine learning.

Research

Efficient LLM inference, reasoning reliability, and trustworthy machine learning.

Efficient LLM Inference

KV-cache reuse, semantic cache distillation, state transfer, selective recomputation, and bandwidth-aware serving.

Reasoning Pipelines

Candidate construction, fixed-verifier reranking, coverage-conversion gaps, and answer-mode compatibility.

Trustworthy ML

Backdoor attacks, frequency-domain robustness, model behavior under perturbation, and security evaluation.

Research Engineering

Reproducible experiments, GPU resource management, and AI-assisted research workflows.

Selected Publications

Five accepted papers: four at CCF A conferences and one at a CCF B conference.

  1. Semantic Cache Distillation: Efficient State Transfer via Reuse and Selective Patching

    Qianli Ma, Zhiqing Tang, Hanshuai Cui, Zhi Yao, Weijia Jia

    Forty-third International Conference on Machine Learning (ICML), 2026. (CCF-A)

  2. Correct but Unselectable: The Hidden Interface Tax in Multi-Candidate Reasoning

    Qianli Ma, Zhiqing Tang, Fanshuai Meng, Zhi Yao, Hanshuai Cui, Weijia Jia

    Fortieth Annual Conference on Neural Information Processing Systems (NeurIPS), 2026. (CCF-A)

  3. Prefill-Guided Trace Allocation for Sample-Efficient Test-Time Scaling

    Zhi Yao, Zhiqing Tang, Hanshuai Cui, Qianli Ma, Fanshuai Meng, Weijia Jia

    Fortieth Annual Conference on Neural Information Processing Systems (NeurIPS), 2026. (CCF-A)

  4. Predict to Skip: Linear Multistep Feature Forecasting for Efficient Diffusion Transformers

    Hanshuai Cui, Zhiqing Tang, Qianli Ma, Zhi Yao, Weijia Jia, Wei Zhao

    34th ACM International Conference on Multimedia (ACM MM), 2026. (CCF-A)

  5. CogGuard: Cognitive and Operational Profiling for Proactive Warning in Edge Intelligent Services

    Zhi Yao, Weihao Chen, Zhiqing Tang, Hanshuai Cui, Qianli Ma, Weijia Jia, Wei Zhao

    IEEE International Conference on Web Services (ICWS), 2026. (CCF-B)

Technical Stack

Tools and workflows for efficient inference, reliable evaluation, and reproducible research.

vLLM and reusable model state

KV-cache reuse, semantic cache distillation, selective patching, and ongoing vLLM integration for efficient inference.

Performance and reliability evaluation

Controlled comparisons and ablations for inference performance, candidate selection, and model safety.

Remote experiments and AI-assisted development

Linux/SSH, Conda, GPU experiment deployment and resource management, and Codex/MCP-assisted research workflows.

Contact

Open to research collaboration.

mql16899@163.com

Zhuhai, China