Xin Liu刘鑫

Ph.D. student · School of Computer Science and Information Technology · Adelaide University

XL

I am a second-year Ph.D. student in computer science at Adelaide University (Oct 2025–), supervised by Prof. Thuc Le, Prof. Jiuyong Li and Prof. Lin Liu in the 4LLAB. Previously, I received a master's degree from Southeast University, jointly supervised by Prof. Min-Ling Zhang and Prof. Weijia Zhang in the PALM Lab. I also received a master's degree in artificial intelligence from Monash University.

My research is on causality and machine learning. It centres on two complementary questions: how to uncover the latent causal relationships underlying real-world data, and how to use causal insights to improve prediction, even when the world is only partially observed.

My research interests include:

My thinking about philosophy has shaped my academic taste.

I always welcome research discussions and collaborations. Please feel free to reach out by email: xin.liu@adelaide.edu.au.

News

Sep 2026Our paper Choosing Before Acting is accepted at NeurIPS 2026. See you in Sydney in December! 🎉
Mar 2026FocusMIL received scores of 5/4/3 at CVPR 2026 and was recommended to CVPR Findings. We opted out and are submitting it elsewhere.
Dec 2025Our paper From Correlation to Causation: Max-Pooling-Based Multi-Instance Learning Leads to More Robust Whole Slide Image Classification (FocusMIL) is now available! Paper · Code
Oct 2025Started my Ph.D. at Adelaide University.
May 2025Presented HACSurv at AISTATS 2025 in Mai Khao, Thailand.
Jan 2025Our paper HACSurv is accepted at AISTATS 2025, with scores of 4/4/4/3.

Publications and preprints

  1. FocusMIL architecture: ResNet18 features, a variational encoder, and an instance classifier
    From Correlation to Causation: Max-Pooling-Based Multi-Instance Learning Leads to More Robust Whole Slide Image Classification
    Xin Liu, Weijia Zhang, Wei Tang, Thuc Duy Le, Jiuyong Li, Lin Liu, Min-Ling Zhang
    arXiv preprint arXiv:2408.09449, 2024 · revised Dec 2025

    We show that max-pooling predicts from one-sided positive evidence, which makes it inherently robust to bias factors in negative bags. We identify the failure modes of existing max-pooling methods and propose FocusMIL to fix them.

    arXivCode Dataset
  2. HACSurv architecture: three monotonic networks producing survival functions joined by copulas
    HACSurv: A Hierarchical Copula-Based Approach for Survival Analysis with Dependent Competing Risks
    Xin Liu, Weijia Zhang, Min-Ling Zhang
    Proceedings of the 28th International Conference on Artificial Intelligence and Statistics (AISTATS), PMLR 258:3079–3087, 2025

    We relax the assumption of conditional independence among competing risks. We use a hierarchical Archimedean copula to capture the asymmetric dependence structure induced by unobserved confounders.

    HTMLarXivCodeTalkPoster
  3. CITA overview: the rolling decision cycle guided by the comparative inference model
    Choosing Before Acting: Comparative Value Estimation for Long-Horizon Tool-Use Agents
    Yu Li, Zheng Zhang, Xin Liu, Shengtian Yang, Guangfeng Cai, Lei Feng
    Advances in Neural Information Processing Systems (NeurIPS), 2026

    We train a comparative inference model to estimate the long-horizon value of candidate tool calls before execution. The model guides tool selection and provides step-level feedback for reinforcement learning.

    arXivPDF

Awards and scholarships

2025 – 2029Australian Government Research Training Program (RTP) Scholarship, Adelaide University.
Jun 2025Outstanding Graduate, Southeast University.

Education

Beyond research

I am a philosophy enthusiast. Reading and thinking about philosophy has shaped my academic taste. These are the three philosophers who have influenced me the most:

High schoolWang Yangming 王阳明 (1472–1529) — critically inherited traditional Chinese philosophy.
UndergraduateLudwig Wittgenstein (1889–1951) — showed that many long-standing philosophical problems arise from the misuse of language.
Graduate schoolJudea Pearl (b. 1936) — advanced the development of causal science.

In my spare time, I enjoy cooking, working out, and singing.

Invited talks

Nov 2024Copula-based Survival Analysis — University of Alberta (online).

Service

Reviewer: ICLR 2027, IJCAI 2025.