Yue Zhao
Avatar of Yue Zhao
Assistant Professor
Thomas Lord Department of Computer Science
School of Advanced Computing

University of Southern California

Los Angeles, CA, USA
Email:
Google Scholar | Short Bio | GitHub | LinkedIn | 知乎 (微调) | 小红书 | 微信 | 300k+ followers in total

Student Committee Request (Update as of 8/29/2025): I have already served on more than 10 student committees this semester, and combined with upcoming travel commitments, my schedule is fully occupied. Therefore, I am unable to take on any new student committees (both internal at USC and external).

Lab Openings. We are warmly welcoming new members to the FORTIS Lab!

Ph.D. Students (1 Ph.D. student for Fall 2026):
  • Due to the large number of interested candidates, future Ph.D. students should be comparable to our current junior Ph.D. students -- see FORTIS Lab.
  • For Fall 2026, I am only considering Ph.D. applicants interested in cryptographic systems, distributed systems, ML systems.
Research Collaborators/Interns (Any Time, All Year Round):
  • We welcome both undergraduate and graduate interns from USC and other institutions.
  • We will provide GPUs/API keys for the project.
  • Preferred candidates are located in North America for time zone compatibility.
  • I do not hire in-person summer interns -- I am enjoying summer and working remotely :)
Application Process: To apply for either opportunities, complete the Application Form, email me after submitting the form, and review the FORTIS Lab website for more information before reaching out.

Collaboration with Me. I am open to external opportunities for invited talks, research collaborations, and employment (only on the part-time/advising/visiting basis). Let us have a chat by email. I frequently visit major cities, e.g., Seattle, NYC, Chicago, Boston, Atlanta, and Bay Area to meet people, give talks, and host social events.

Research Interests: My research aims to build trustworthy, robust, and scalable AI that advances science and benefits society. I focus on rigorous algorithmic foundations, open-source system development, and high-impact applications in both human-centric and scientific domains.

  1. Robust & Trustworthy AI: Detecting the Unexpected.
    I design core algorithms to detect anomalies, out-of-distribution (OOD) data, and outliers across diverse modalities (including graph-structured data). These methods reinforce AI systems against rare or unseen scenarios, enhancing reliability, security, and interpretability.
    Keywords: Anomaly Detection, OOD Detection, Trustworthy AI, Graph Anomaly Detection
  2. AI for Science & Society: Foundation Models in Action.
    By pairing robust detection with large language models (LLMs) and generative AI (GenAI), I tackle interdisciplinary challenges—from scientific discovery to political forecasting and computational social science. This approach bridges algorithmic research with real-world decision-making and public policy.
    Keywords: AI for Science, Generative AI, LLMs, Political Forecasting, Computational Social Science
  3. Scalable, Automated & Open-source ML Systems.
    To ensure widespread adoption, I build reproducible and efficient tools—most notably PyOD (27M+ downloads) for anomaly detection, along with PyGOD, ADBench, and other libraries with 20K+ GitHub stars (top 800 worldwide). My work emphasizes automated model selection, distributed inference, and user-friendly designs, democratizing advanced ML across academia and industry.
    Keywords: ML Systems, Automated ML, Open-source AI, Distributed Computing

Biography.

✈ News and Travel

[Aug 2025] We have two new papers accepted to EMNLP Findings 2025: one on causal methods for hallucination mitigation (Treble Counterfactual VLMs) and another introducing a benchmark for NLP anomaly detection (NLP-ADBench). See our Treble Preprint and NLP-ADBench Preprint!

[Aug 2025] We have a new paper, M3OOD: Automatic Selection of Multimodal OOD Detectors, introducing a meta-learning framework for robust multimodal OOD detection. See our Preprint!

[Aug 2025] We have a new paper, Mitigating Hallucinations in Large Language Models via Causal Reasoning, which proposes causal DAG construction and reasoning to reduce LLM hallucinations. See our Preprint!

[Aug 2025] We have a new paper on improving typhoon track forecasting with LLM-augmented transformers (TyphoFormer) accepted to ACM SIGSPATIAL 2025; see our Preprint!

[Jul 2025] Our new paper "JailDAM" has been accepted to COLM 2025! It proposes an adaptive memory approach for jailbreak detection in vision-language models. See the Preprint!

[Jul 2025] Collaborated on two new preprints on extreme weather modeling! One on predicting hurricane-induced economic losses (Learning from the Storm), and another on improving typhoon forecasting using LLMs (TyphoFormer).

[Jun 2025] We have a new paper accepted to ICCV 2025 on secure on-device video OOD detection without backpropagation; see our Preprint!

[Jun 2025] We have a new paper accepted to ECML PKDD 2025 on leveraging LLMs for few-shot graph OOD detection; see our Preprint!

[Jun 2025] We have a new paper, “SocialMaze,” introducing a benchmark to evaluate social reasoning in LLMs across games, interactions, and online platforms. See our Preprint!

[May 2025] We have a new paper on benchmarking personalized conversational reasoning for LLMs (PersonaConvBench). See our Preprint!

[May 2025] We have a new paper introducing AD-AGENT, a multi-agent LLM framework for anomaly detection. See our Preprint!

[May 2025] Our paper "AD-LLM: Benchmarking Large Language Models for Anomaly Detection" has been accepted to ACL 2025 Findings! Congrats to Tiankai Yang and see the Preprint.

[May 2025] Our survey paper "From Selection to Generation: A Survey of LLM-based Active Learning" has been accepted to ACL 2025 main conference! See the preprint.

[May 2025] Our tutorial "A Survey on Model Extraction Attacks and Defenses for Large Language Models" was accepted to KDD 2025 as a Lecture-Style Tutorial! Congrats to Kaixiang Zhao, Lincan Li, Kaize Ding, Neil Gong, and Yushun Dong!

[May 2025] We have a new paper on zero-shot graph OOD detection using foundation models (GLIP-OOD); see our Preprint!

[May 2025] We have a new paper introducing GOE-LLM, a framework using LLMs to generate synthetic OOD nodes for graph OOD detection without requiring real OOD data. See our Preprint!

🏅 Awards and Grants

As Principal Investigator (August 2023 onwards)
Prior to Principal Investigator Role (Before August 2023)