Zhi-ning Liu (刘芷宁)

News

Education

University of Illinois Urbana-Champaign

Ph.D. in Computer Science · 2022 - 2026 (expected)

Department of Computer Science · Advisor: Prof. Hanghang Tong

Jilin University

M.Eng. in Computer Science · 2019 - 2022

School of Artificial Intelligence · Advisor: Prof. Yi Chang

B.Sc. in Computer Science · 2015 - 2019

Tang-Aoqing Honors Program · Computer Science

Experience

Amazon

Applied Scientist II · Palo Alto, CA · starting June 2026

Amazon Ads · LLM for recommendation

Applied Scientist Intern · Palo Alto, CA · May - Dec 2025

Amazon Ads · Vision language model reasoning reliability -> ICLR 2026, ICML 2026, ACL 2026

Applied Scientist Intern · Palo Alto, CA · May - Aug 2024

Amazon Rufus · RAG-based language model reasoning -> ACL 2025

Applied Scientist Intern · Seattle, WA · May - Aug 2023

Amazon Search · Multi-task learning for partially ordered entity ranking

Microsoft Research

Research Intern · Beijing · Aug 2018 - June 2019

Machine Learning Group · Extreme class-imbalanced learning -> ICDE 2020, NeurIPS 2020

Publications

Publications, preprints, and submissions, sorted by year. See the full list on Google Scholar.

• Do VLMs Have a Moral Backbone? A Study on the Fragile Morality of Vision-Language Models
  Zhining Liu*, Tianyi Wang*, Xiao Lin, Penghao Ouyang, Gaotang Li, Ze Yang, Hui Liu, Sumit Keswani, Vishwa Pardeshi, Huijun Zhao, Wei Fan, Hanghang Tong
  ACL Findings 2026
  Key Insight: VLM moral judgments are fragile when visual evidence and textual cues create subtle ethical tension.
  Paper
  VLMMultimodalEthicsTrustworthy-AISafety
• MORALISE: A Structured Benchmark for Moral Alignment in Visual Language Models
  Zhining Liu*, Xiao Lin*, Ze Yang*, Gaotang Li, Ruizhong Qiu, Shuke Wang, Hui Liu, Haotian Li, Sumit Keswani, Vishwa Pardeshi, Huijun Zhao, Wei Fan, Hanghang Tong
  ICML 2026
  Key Insight: Structured visual moral scenarios reveal alignment failures hidden by coarse benchmark scores.
  Paper Dataset
  VLMMultimodalEthicsTrustworthy-AIBenchmarking
• Agentic Reasoning for Large Language Models
  Tianxin Wei*, Ting-Wei Li*, Zhining Liu*, Xuying Ning, Ze Yang, Jiaru Zou, Zhichen Zeng, Ruizhong Qiu, Xiao Lin, et al.
  arXiv preprint, 2026
  Key Insight: A unified taxonomy connects planning, tool use, memory, reflection, and self-improvement in agentic reasoning.
  Paper Project HF Paper
  LLMReasoningSurvey
• Mixture of Sequence: Theme-aware Mixture-of-Experts for Long-Sequence Recommendation
  Xiao Lin, Zhicheng Tang, Weilin Cong, Mengyue Hang, Kai Wang, Yajuan Wang, Zhichen Zeng, Ting-Wei Li, Hyunsik Yoo, Zhining Liu, Xuying Ning, Ruizhong Qiu, Wen-Yen Chen, Shuo Chang, Rong Jin, Huayu Li, Hanghang Tong
  WebConf 2026 Oral
  Key Insight: Theme-aware experts let recommender models specialize over long, shifting user behavior sequences.
  Paper Code
  Recommender-SystemsEnsembleModel-Adaptation
• Seeing but Not Believing: Probing the Disconnect Between Visual Attention and Answer Correctness in VLMs
  Zhining Liu, Ziyi Chen, Hui Liu, Chen Luo, Xianfeng Tang, Suhang Wang, Jingying Zeng, Zhenwei Dai, Zhan Shi, Tianxin Wei, Hanqing Lu, Benoit Dumoulin, Hanghang Tong
  ICLR 2026
  Key Insight: Correct visual attention does not guarantee correct visual-language reasoning.
  Paper
  VLMMultimodalInterpretabilityTrustworthy-AI
• Continual Low-Rank Adapters for LLM-based Generative Recommender Systems
  Hyunsik Yoo, Ting-Wei Li, SeongKu Kang, Zhining Liu, Charlie Xu, Qilin Qi, Hanghang Tong
  ICLR 2026
  Key Insight: Continually evolving LoRA adapters preserve generative recommendation quality as user data shifts.
  Paper Code
  LLMRecommender-SystemsModel-Adaptation
• Language in the Flow of Time: Time-Series-Paired Texts Weaved into a Unified Temporal Narrative
  Zihao Li, Xiao Lin, Zhining Liu, Jiaru Zou, Ziwei Wu, Lecheng Zheng, Dongqi Fu, Yada Zhu, Hendrik Hamann, Hanghang Tong, Jingrui He
  ICLR 2026
  Key Insight: Time-series and paired text can be aligned into unified temporal narratives for multimodal reasoning.
  Paper Code
  Time-SeriesLLMMultimodal
• TRQA: Time Series Reasoning Question And Answering Benchmark
  Baoyu Jing, Sanhorn Chen, Lecheng Zheng, Boyu Liu, Zihao Li, Jiaru Zou, Tianxin Wei, Zhining Liu, Zhichen Zeng, Ruizhong Qiu, Xiao Lin, Yuchen Yan, Qi Yu, Dongqi Fu, Jingrui He, Hanghang Tong
  In submission, 2026
  Key Insight: Time-series QA should test compositional reasoning over temporal patterns, not just forecasting accuracy.
  Paper
  Time-SeriesReasoningBenchmarking
• PLANETALIGN: A Comprehensive Python Library for Benchmarking Network Alignment
  Qi Yu, Zhichen Zeng, Yuchen Yan, Zhining Liu, Baoyu Jing, Ruizhong Qiu, Ariful Azad, Hanghang Tong
  ICLR 2026
  Key Insight: A unified Python library makes network alignment algorithms easier to compare, reproduce, and extend.
  Paper Code
  GraphNetwork-AlignmentBenchmarking
• Flow Matching Meets Biology and Life Science: A Survey
  Zihao Li, Zhichen Zeng, Xiao Lin, Feihao Fang, Yanru Qu, Zhe Xu, Zhining Liu, Xuying Ning, Tianxin Wei, Ge Liu, Hanghang Tong, Jingrui He
  npj Artificial Intelligence 2026
  Key Insight: Flow matching provides a flexible generative lens for molecular, cellular, and biological modeling tasks.
  Paper arXiv
  BiologySurvey
• ReMix: Reinforcement Routing for Mixtures of LoRAs in LLM Finetuning
  Ruizhong Qiu, Hanqing Zeng, Yinglong Xia, Yiwen Meng, Ren Chen, Jiarui Feng, Dongqi Fu, Qifan Wang, Jiayi Liu, Jun Xiao, Xiangjun Fan, Benyu Zhang, Hong Li, Zhining Liu, Hyunsik Yoo, Hanghang Tong
  arXiv preprint, 2026
  Key Insight: Reinforcement routing learns how to select and combine LoRA experts during LLM finetuning.
  Paper
  LLMEnsembleReasoningModel-Adaptation
• Inference Scaling of LLM Ensembling: Bridging Token Spaces with Token Translation
  Tianxin Wei, Zhichen Zeng, Ruizhong Qiu, Zhining Liu, Xuying Ning, Xinrui He, Wenxuan Bao, Qi He, Xianfeng Tang, Hanghang Tong, Jingrui He
  In submission, 2026
  Key Insight: Token translation bridges heterogeneous vocabularies so LLM ensembles can scale at inference time.
  Paper
  LLMEnsembleReasoning
• WAPITI: A Watermark for Finetuned Open-Source LLMs
  Lingjie Chen, Ruizhong Qiu, Siyu Yuan, Zhining Liu, Tianxin Wei, Hyunsik Yoo, Zhichen Zeng, Deqing Yang
  In submission, 2026
  Key Insight: Finetuned open-source LLMs can retain detectable ownership signals without sacrificing utility.
  Paper
  LLMWatermarkingSafetyModel-Adaptation
• AdaFuse: Adaptive Ensemble Decoding for Large Language Models
  Chengming Cui, Tianxin Wei, Ziyi Chen, Ruizhong Qiu, Zhichen Zeng, Zhining Liu, Xuying Ning, Duo Zhou, Jingrui He
  ACL 2026 Main
  Key Insight: Adaptive ensemble decoding fuses multiple LLM outputs during generation for stronger test-time reasoning.
  Paper Code
  LLMEnsembleReasoningModel-Adaptation
• Mem-Gallery: Benchmarking Multimodal Long-Term Conversational Memory for MLLM Agents
  Yuanchen Bei, Tianxin Wei, Xuying Ning, Yanjun Zhao, Zhining Liu, Xiao Lin, Yada Zhu, Hendrik Hamann, Jingrui He, Hanghang Tong
  ACL 2026 Main
  Key Insight: Multimodal agents still struggle to store, retrieve, and use long-term conversational memory.
  Paper Code
  LLMVLMMultimodalBenchmarking
• CLIMB: Class-imbalanced Learning Benchmark on Tabular Data
  Zhining Liu, Zihao Li, Ze Yang, Tianxin Wei, Jian Kang, Yada Zhu, Hendrik Hamann, Jingrui He, Hanghang Tong
  NeurIPS 2025
  Key Insight: A standardized benchmark exposes when class-imbalanced tabular methods actually generalize.
  Paper Code Docs
  Class-ImbalanceTabular-DataBenchmarking
• Breaking Silos: Adaptive Model Fusion Unlocks Better Time Series Forecasting
  Zhining Liu, Ze Yang, Xiao Lin, Ruizhong Qiu, Tianxin Wei, Yada Zhu, Hendrik Hamann, Jingrui He, Hanghang Tong
  ICML 2025
  Key Insight: Sample-level adaptive fusion lets heterogeneous forecasters complement each other instead of competing in silos.
  Paper Code
  Time-SeriesEnsembleMeta-Learning
• SelfElicit: Your Language Model Secretly Knows Where the Relevant Evidence is
  Zhining Liu, Rana Ali Amjad, Ravinarayana Adkathimar, Tianxin Wei, Hanghang Tong
  ACL 2025 Main
  Key Insight: LLMs can self-elicit where relevant evidence lies, reducing reliance on external retrieval heuristics.
  Paper Code
  LLMReasoningRetrieval
• ClimateBench-M: A Multi-Modal Climate Data Benchmark with a Simple Generative Method
  Dongqi Fu, Yada Zhu, Zhining Liu, Lecheng Zheng, Xiao Lin, Zihao Li, Liri Fang, Katherine Tieu, Onkar Bhardwaj, Kommy Weldemariam, Hanghang Tong, Hendrik Hamann, Jingrui He
  CIKM 2025
  Key Insight: Climate modeling benefits from benchmarks that jointly test time-series, image, and generative signals.
  Paper Code Dataset-TS Dataset-IMG
  MultimodalClimateBenchmarking
• Not All Voices Are Rewarded Equally: Probing and Repairing Reward Models across Human Diversity
  Zihao Li, Feihao Fang, Xitong Zhang, Jiaru Zou, Zhining Liu, Wei Xiong, Ziwei Wu, Baoyu Jing, Jingrui He
  EMNLP Findings 2025
  Key Insight: Reward models can encode demographic preference gaps, and targeted repair can reduce those disparities.
  Paper Code
  LLMReward-ModelingFairnessTrustworthy-AI
• LLM-RecG: A Semantic Bias-Aware Framework for Zero-Shot Sequential Recommendation
  Yunzhe Li, Junting Wang, Hari Sundaram, Zhining Liu
  RecSys 2025
  Key Insight: Semantic group information helps LLM recommenders recognize and reduce bias in zero-shot settings.
  Paper ACM
  LLMRecommender-SystemsFairness
• Matcha: Mitigating Graph Structure Shifts with Test-Time Adaptation
  Wenxuan Bao, Zhichen Zeng, Zhining Liu, Hanghang Tong, Jingrui He
  ICLR 2025
  Key Insight: Test-time adaptation can mitigate graph structure shifts without retraining on the target graph.
  Paper arXiv Code
  GraphGNNModel-Adaptation
• THeGCN: Temporal Heterophilic Graph Convolutional Network
  Yuchen Yan, Yuzhong Chen, Huiyuan Chen, Xiaoting Li, Zhe Xu, Zhichen Zeng, Lihui Liu, Zhining Liu, Hanghang Tong
  In submission, 2026
  Key Insight: Temporal heterophily requires graph convolutions that model changing cross-class neighborhood patterns.
  Paper
  GraphGNNTemporal-Graph
• BackTime: Backdoor Attacks on Multivariate Time Series Forecasting
  Xiao Lin, Zhining Liu, Dongqi Fu, Ruizhong Qiu, Hanghang Tong
  NeurIPS 2024 Spotlight
  Key Insight: Time-series forecasters can be compromised by backdoor triggers embedded in multivariate temporal patterns.
  Paper Code
  Time-SeriesSafetyTrustworthy-AI
• AIM: Attributing, Interpreting, Mitigating Data-encoded Unfairness
  Zhining Liu, Ruizhong Qiu, Zhichen Zeng, Yada Zhu, Hendrik Hamann, Hanghang Tong
  KDD 2024
  Key Insight: Data-encoded unfairness can be attributed, interpreted, and mitigated before it becomes model behavior.
  Paper Code
  FairnessInterpretabilityTrustworthy-AI
• Class-Imbalanced Graph Learning without Class Rebalancing
  Zhining Liu, Zhichen Zeng, Ruizhong Qiu, Hyunsik Yoo, David Zhou, Zhe Xu, Yada Zhu, Kommy Weldemariam, Jingrui He, Hanghang Tong
  ICML 2024
  Key Insight: Bias-aware graph learning can handle class imbalance without naive resampling or class reweighting.
  Paper Code
  GraphGNNClass-Imbalance
• Group Fairness via Group Consensus
  Zhining Liu*, Eunice Chan*, Ruizhong Qiu, Yuheng Zhang, Ross Maciejewski, Hanghang Tong
  FAccT 2024
  Key Insight: Group consensus offers a practical fairness signal when protected groups disagree in complex ways.
  Paper PDF
  FairnessTrustworthy-AI
• Graph Mixup on Approximate Gromov-Wasserstein Geodesics
  Zhichen Zeng, Ruizhong Qiu, Zhe Xu, Zhining Liu, Yuchen Yan, Tianxin Wei, Lei Ying, Jingrui He, Hanghang Tong
  ICML 2024
  Key Insight: Gromov-Wasserstein geodesics create topology-aware graph mixup paths for stronger augmentation.
  Paper PMLR Code
  GraphGNNOptimal-Transport
• Ensuring User-side Fairness in Dynamic Recommender Systems
  Hyunsik Yoo, Zhichen Zeng, Jian Kang, Ruizhong Qiu, David Zhou, Zhining Liu, Fei Wang, Charlie Xu, Eunice Chan, Hanghang Tong
  WWW 2024
  Key Insight: Dynamic recommenders need fairness constraints that evolve with users, items, and exposure patterns.
  Paper arXiv Code
  FairnessRecommender-SystemsTrustworthy-AI
• Hierarchical Multi-Marginal Optimal Transport for Network Alignment
  Zhichen Zeng, Boxin Du, Si Zhang, Yinglong Xia, Zhining Liu, Hanghang Tong
  AAAI 2024
  Key Insight: Hierarchical multi-marginal optimal transport aligns multiple networks through shared structural signals.
  Paper arXiv
  GraphNetwork-AlignmentOptimal-Transport
• Taming Over-Smoothing Representation on Heterophilic Graphs
  Kai Guo, Xiaofeng Cao, Zhining Liu, Yi Chang
  Information Sciences, 2023
  Key Insight: Heterophilic graphs require representation smoothing to be controlled rather than blindly increased.
  Paper
  GraphGNNTrustworthy-AI
• Web-based Long-term Spine Treatment Outcome Forecasting
  Zhining Liu*, Hangting Ye*, Wei Cao, Amir Mohammad Amiri, Jiang Bian, Yi Chang, Jon D. Lurie, Jim Weinstein, Tie-Yan Liu
  KDD 2023
  Key Insight: Web-based modeling can support long-term spine treatment outcome forecasting from real clinical data.
  Paper
  HealthcareTime-Series
• UADB: Unsupervised Anomaly Detection Booster
  Zhining Liu*, Hangting Ye*, Xinyi Shen, Wei Cao, Shun Zheng, Xiaofan Gui, Huishuai Zhang, Yi Chang, Jiang Bian
  ICDE 2023
  Key Insight: A boosting wrapper can improve unsupervised anomaly detectors without relying on anomaly labels.
  Paper arXiv Code
  Anomaly-DetectionClass-ImbalanceTabular-Data
• A Survey of Explainable Graph Neural Networks for Cyber Malware Analysis
  Dana Warmsley, Alex Waagen, Jiejun Xu, Zhining Liu, Hanghang Tong
  IEEE BigData 2022
  Key Insight: Explainable GNN techniques can make cyber malware analysis more transparent and actionable.
  Paper Info
  GraphGNNInterpretabilitySurvey
• MESA: Boost Ensemble Imbalanced Learning with Meta-sampler
  Zhining Liu, Pengfei Wei, Jing Jiang, Wei Cao, Jiang Bian, Yi Chang
  NeurIPS 2020
  Key Insight: A meta-sampler can learn how to construct better ensemble training sets for imbalanced data.
  Paper Code Zhihu
  Class-ImbalanceEnsembleMeta-Learning
• Self-paced Ensemble for Highly Imbalanced Massive Data Classification
  Zhining Liu, Wei Cao, Zhifeng Gao, Jiang Bian, Hechang Chen, Yi Chang, Tie-Yan Liu
  ICDE 2020
  Key Insight: Self-paced sampling lets ensembles learn from massive imbalanced data from easy cases to harder ones.
  Paper Code Zhihu
  Class-ImbalanceEnsembleTabular-Data

Selected Awards & Honors

Fun Facts