Ruixiang Tang

Ruixiang (Ryan) Tang

Department of Computer Science

Rutgers-New Brunswick

Address: 110 Frelinghuysen Rd, Hill Center 416, Piscataway, NJ 08854

Email: ruixiang.tang@rutgers.edu

Portrait of Ruixiang (Ryan) Tang

About Our Lab

I am an Assistant Professor in the Department of Computer Science at Rutgers University–New Brunswick, where I lead the Trustworthy and Reliable AI Lab (TRAIL). I received my Ph.D. in Computer Science from Rice University and my B.S. from Tsinghua University.

TRAIL works on Trustworthy AI, with the goal of infusing trust throughout the AI lifecycle. We pursue this goal by formalizing trustworthiness objectives (e.g., robustness, interpretability, and safety) and developing principled algorithms with measurable guarantees and transparent failure modes. We also collaborate closely with biologists and health informaticians to translate trustworthy learning methods to high-impact problems in biology and medicine.

Large Foundation Models introduce new and complex trustworthiness challenges. Our long-term objective is to develop learning algorithms that improve reasoning reliability under distribution shift and adversarial manipulation, supported by theoretical frameworks that connect representation, optimization, and uncertainty to trust outcomes. By advancing the underlying reasoning mechanisms, we aim to make these models more reliable, transparent, and resistant to misuse.

I am always glad to work with outstanding students. If you would like to collaborate on research, please send me an email at ruixiang.tang@rutgers.edu. For prospective PhD students, I read every application after the Rutgers application deadline and follow up directly with candidates whose interests align with ours.

Research Overview

Operational Interpretability

Turning mechanistic insight into practical tools that monitor, audit, and steer model behavior once deployed.

Agent Safety & Control

Keeping autonomous and multiagent systems reliable through continuous auditing, early failure prediction, and corrective intervention.

Perception Reliability

Making multimodal perception systems robust to uncertainty, distribution shift, and adversarial inputs.

AI for Biomedicine

Translating trustworthy learning into high-impact problems across biology and medicine with clinical and biological collaborators.

News

  • [2026/11]studentCongratulations to Boxuan Zhang on being accepted to the AAAI-27 Doctoral Consortium!
  • [2026/11]talkGlad to join the PanKbase retreat at the University of Michigan, where we presented our temporally evolving knowledge graph of the type 1 diabetes literature. Thanks to the organizers for a great meeting!
  • [2026/11]awardHonored to be selected for the IEEE BigData 2026 New Faculty Highlights program. Many thanks to the organizers!
  • [2026/11]talkI gave a talk at SRI International introducing our group's recent work. Many thanks to Dr. Rakesh Kumar for hosting.
  • [2026/11]serviceI am happy to serve as a Tutorial Co-Chair for the ACM Conference on Bioinformatics, Computational Biology, and Health Informatics (ACM-BCB 2027), to be held August 10–13, 2027 at the University of Massachusetts Boston.
  • [2026/10]talkI gave a guest lecture in COMP 5630/6630 Machine Learning at Auburn University on "When AI Fails: From Mechanistic Understanding to Proactive Auditing." [Slides Link] Many thanks to Prof. Jiaqi Wang for hosting.
  • [2026/10]awardWe received an Amazon Research Award. Thanks, Amazon!
  • [2026/09]paperTwo papers accepted to NeurIPS 2026. We introduced online auditing for early failure prediction in multi-agent systems, and studied token value inequality for efficient reasoning.
  • [2026/09]serviceI am joining the ML Alignment & Theory Scholars (MATS) program as a research mentor, working with fellows on AI control and monitoring and on interpretability.
  • [2026/08]serviceI'm happy to serve as guest editor for the Pattern Recognition Special Issue on “World Modeling for Dynamic Video Understanding.” Submissions are warmly welcome.
  • [2026/08]paperOne paper accepted to EMNLP 2026. We studied proactive safety auditing for multi-turn LLM agents, estimating risk from partial trajectories before unsafe behavior surfaces in the outcome.
  • [2026/07]paperTwo papers accepted to COLM 2026. We studied reward hacking in large language models and grounded consistency in video multimodal large language models.
  • [2026/06]paperThree papers accepted to ECCV 2026. We explored personalization, counting mechanisms, and post-hallucination reasoning in large vision-language models.
  • [2026/06]talkI am pleased to be invited to give a talk at IBM on agent safety and early failure detection [Slides Link].
  • [2026/05]grantWe are honored to receive an AI Pilot Award from NIDDK through the PanKbase program. We thank the program leadership, and look forward to contributing to this important diabetes research community resource.
  • [2026/05]awardOur ICME paper "Shifting Uncertainty to Critical Moments: Towards Reliable Uncertainty Quantification for VLA Models" has been selected as a Best Paper Award candidate.
  • [2026/04]paperTwo papers accepted to ICML 2026. We explored the actionable mechanistic interpretation of massive values in LLMs and proposed generalizable implicit In-Context Learning via attention routing.
  • [2026/04]studentThis summer, Boxuan and Yihao will be research interns at AMD, Zeru at NVIDIA, and Yufei at Genentech. Congratulations to all!
  • [2026/04]paperOne paper accepted by ACL 2026, we develop a case-augmented alignment approach that improves LLM safety and reduces over-refusal compared to rule-based methods.
  • [2026/03]grantWe received AMD University Program: AI & HPC Cluster Allocation Award to support our research on AI agents and memory systems, thanks to AMD for their support!
  • [2026/02]paperOne paper has been accepted at ICME 2026. We developed a framework to accurately predict the uncertainty of VLA models.
  • [2026/02]paperOne paper has been accepted to CVPR 2026 Findings, we evaluate and enhance LLMs’ counterfactual reasoning ability for video understanding tasks.
  • [2026/01]grantOur interdisciplinary project "AI–Geoscience Collaborations for a Changing Arctic" has been funded by the Rutgers SAS Research in Academic Themes. We look forward to advancing foundational results at the AI–Earth systems interface.
  • [2026/01]paperOne paper accepted by ICLR 2026, we design a memory efficient LLM fine tuning method via instance-aware token ditching.
  • [2026/01]paperOne paper accepted by EACL 2026, we provide FaithLM, an intervention-based framework that evaluates and improves LLM explanation faithfulness.
  • [2025/12]serviceI'm happy to serve as organizer for the ICME 2026 Special Session on “Trustworthy and Secure Vision-Language-Action Models: Foundations, Methods, and Evaluation.” Submissions are warmly welcome.
  • [2025/11]paperThree papers accepted to AAAI 2026 (two orals, one poster). Our work advances multimodal reasoning, improves reasoning efficiency, and explores counterfactual reasoning in LLMs.
  • [2025/10]studentCongratulations to Zeru Shi for being selected as a Rutgers Climate and Energy Institute (RCEI) Fellowship!
  • [2025/10]paperOne paper accepted by AACL 2025. We provide a dataset, taxonomy, and baselines for analysis of rational persuasion and manipulation.
  • [2025/08]paperThree papers (one main, two findings) accepted by EMNLP 2025. We explored multimodal in-context learning, LoRA security, and an LLM agent for text style transfer.
  • [2025/07]paperOne paper accepted by Bioinformatics Advances. We explored a LLM-based framework that improves rare disease gene prioritization through phenotype-driven classification.
  • [2025/07]paperOne paper accepted by COLM 2025: we explored replacing multimodal examples with learnable vectors to improve in-context learning in Large Vision-Language Models.
  • [2025/06]paperTwo papers accepted by ICCV 2025: we explored the robustness of Vision-Language-Action model-based robotics, and addressed hallucination in Large Vision-Language Models.
  • [2025/05]serviceInvited to serve as an Associate Editor for Smart Health.
  • [2025/05]studentCongratulations to Leroy Souz for being named a Paul Robeson Centennial Scholar and for reaching the semifinals of the Rice Business Plan Competition.
  • [2025/05]paperTwo papers accepted by ACL 2025 Main, we explored using explainability to detect backdoors in LLMs and disentangle reasoning from memory in LLMs.
  • [2025/05]paperOne paper accepted by ICML 2025, we show that Massive Q/K values are key to contextual understanding in LLMs.
  • [2025/04]paperOur paper "Harnessing the Power of LLMs in Practice: A Survey on ChatGPT and Beyond" is now the all-time most-read article of ACM Transactions on Knowledge Discovery from Data (TKDD)!
  • [2025/02]grantReceived $2,000 API credit from OpenAI to support our research on detecting and mitigating hallucination in VLMs.
  • [2025/02]paperOne paper accepted by SIGKDD Explorations, we proposed to mitigate shortcuts in natural language understanding tasks.
  • [2025/01]paperOne paper accepted by ICLR 2025, we proposed MRT to efficiently edit multimodal representations.
  • [2025/01]talkOur tutorial about "Understanding Large-Scale Machine Learning Robustness under Paradigm Shift" is accepted by SDM 2025. See you in Virginia!
  • [2024/09]paperThree papers have been accepted for EMNLP 2024, focusing on issues related to LLM intellectual property protection, bias mitigation, and building trustworthy agents.
  • [2024/07]paperOur paper, Mitigating Relational Bias on Knowledge Graphs, has been accepted by TKDD.
  • [2024/04]awardOur paper, The Science of Detecting LLM-Generated Text, has been accepted for a cover paper in the April 2024 issue of Communications of the ACM (CACM)!
  • [2024/04]awardOur research about AI for watermarking has been awarded the annual Hershel M. Rich Invention Award, congratulations to the team!
  • [2024/03]paperOne paper has been accepted by the NAACL 2024. We proposed a key prompt protection mechanism to safeguard large language models.
  • [2024/02]paperOne paper has been accepted by the Journal of Biomedical Informatics. We proposed a soft prompt calibration method to mitigate the variance of large language model output in the radiology report summarization task.
  • [2024/01]paperOur paper, Harnessing the Power of LLMs in Practice: A Survey on ChatGPT and Beyond, has been accepted by TKDD.
  • [2023/10]awardOur paper about using LLMs for patient-trial matching was selected as the AMIA 2023 Best Student Paper and KDDM Student Innovation Award.
  • [2023/10]awardOur paper about building knowledge refinement and retrieval systems for interdisciplinarity in biomedical research has been selected as CIKM 2023 Best Demo Paper Honorable Mention.
  • [2023/10]paperOne paper has been accepted by EMNLP 2023. We introduce a membership inference attack aimed at Large Language Models to analyze associated privacy risks.
  • [2023/09]paperTwo papers have been accepted by NeurIPS 2023. We proposed a honeypot mechanism to defend against backdoor attacks in language models.
  • [2023/08]paperOne paper has been accepted by ECML-PKDD 2023. We proposed a serial key protection mechanism for safeguarding DNN models.
  • [2023/08]paperTwo papers have been accepted by CIKM 2023. We proposed a transferable watermark for defending against model extraction attacks.
  • [2023/07]paperThree Papers have been accepted by AMIA 2023 Annual Symposium. We investigated methods for harnessing the capabilities of LLMs in several healthcare tasks, such as Named Entity Recognition, Relation Extraction, and Patient-Trial Matching.
  • [2023/05]paperOne paper has been accepted by ACL 2023. We reveal that LLMs are "lazy learners" that tend to exploit shortcuts in prompts for downstream tasks.
  • [2023/04]paperOne paper has been accepted by SIGKDD Exploration 2023. We proposed a clean-label backdoor-based watermarking framework for safeguarding training datasets.
  • [2023/03]paperOne paper has been accepted by ICHI 2023. We proposed a deep ensemble framework for improving phenotype prediction from multi-modal data.

Services

Editorial Board:

Program Committee Member / Reviewer:

Journal Reviewer:

Recognition: