Welcome to Defu Cao’s Homepage
Research Identity
I build foundation models and agentic LLM systems for structured, scientific, and decision-critical data.
Agentic Post-Training · LLM Safety · Scientific Foundation Models · Numerical LLM Interfaces
Defu Cao /dəfu: tsaʊ/ is a Ph.D. candidate in the Thomas Lord Department of Computer Science at the University of Southern California, working with Prof. Yan Liu at the USC Melady Lab. He is also working closely with Prof. Yue Zhao and Prof. Angela Zhou from USC Marshall Data Sciences and Operations. Prior, he earned his Master’s Degree at the School of EECS Peking University.
He has been fortunate to work as a visiting scholar with Prof. Yisong Yue at the California Institute of Technology (2025), focusing on multimodal foundation models and LLM infrastructure, and Prof. Kun Zhang at MBZUAI (2023), focusing on causal inference and causal discovery.
Cao’s research focuses on foundation models for structured and scientific data, agentic post-training systems, and LLM safety. Time series and scientific signals are a major application domain, but the broader goal is to build models that can represent complex numerical structure, reason with evidence, interact with tools and humans, and remain reliable under distribution shift, physical constraints, privacy risks, and high-stakes deployment.
His current research agenda connects several layers:
| Direction | Representative Work | Core Question |
|---|---|---|
| Post-training and agentic systems | TSOrchestra, HILA, TS-Reasoner | How can LLMs become judges, planners, tool users, collaborators, and continually improving agents? |
| LLM safety and agent security | “Someone Hid It!”, Topology Matters | How do retrieval systems, RAG, agent memory, and multi-agent topologies fail under attack or privacy pressure? |
| Foundation models for structured/scientific data | TEMPO, ClimateLLM, TimeDiT, PINFDiT | How can foundation models capture numerical structure, uncertainty, multi-resolution dynamics, and physics constraints? |
| Numerical interfaces for LLMs | Speaking Numbers to LLMs, GPT4MTS | How should LLMs ingest continuous values, multimodal signals, and domain-specific numerical evidence? |
| Benchmarks and high-stakes evaluation | TSAIA, TemporalBench, Physiological Waveform Reasoning | How do we evaluate reasoning quality, traceability, and utility beyond benchmark accuracy? |
He has published his research in top conference proceedings including NeurIPS, ICML, ICLR, CVPR, ICRA, and NAACL.
Research Highlights
| Highlight | Signal |
|---|---|
| TSOrchestra | Rank #1 on the Salesforce GIFT-Eval leaderboard; LLM-as-judge orchestration for foundation-model ensembles. |
| Agentic systems | HILA, TS-Reasoner, human-in-the-loop multi-agent learning, tool-augmented reasoning, and adaptive collaboration. |
| LLM safety | Query-agnostic attacks on LLM-based retrieval and memory leakage in multi-agent systems. |
| Foundation models | TEMPO, ClimateLLM, TimeDiT, and PINFDiT for structured, scientific, and physics-constrained data. |
| Impact | 3,500+ citations across 30+ publications; invited talks at Caltech, Rice, UBC, UCSD, DataDog AI Research Lab, and Peking University; invited workshop presentation at Google; Best RA Award 2024 in the USC CS Department. |
Recent News (2025-2026)
- 05/2026: I am honored to have been awarded the USC Dissertation Completion Fellowship!
- 04/2026: One paper on LLM retrieval security, query-agnostic black-box attacks, accepted by ICML!
- 04/2026: One position paper on verifiable Physiological Waveform Reasoning with foundation models and agentic LLMs accepted by ICML!
- 04/2026: One paper on numerical embeddings for LLM time-series forecasting, multi-wavelet number embeddings, accepted by IJCAI!
- 03/2026: TS-Reasoner, our domain-oriented time series inference agent, was accepted by TMLR!
- 01/2026: One paper on human-in-the-loop multi-agent LLM systems accepted by ICLR!
- 01/2026: One paper on physics-informed diffusion transformers for time series foundation models accepted by ICLR!
- 01/2026: One paper on Offline RL accepted by AISTATS!
- 01/2026: Invited talk “Frontiers of Physics-Informed Time Series Foundation Models.” Host: Peking University, School of AI for Science.
- 01/2026: Invited talk “TSFoundation: From Foundation Models to Agents Orchestration.” Host: DataDog AI Research Lab.
- 12/2025: 🚀 TSOrchestra, our agentic-guided foundation model forecasting framework from USC Melady Lab, reached Rank #1 on the Salesforce GIFT-Eval leaderboard (MASE & CRPS)!
- 10/2025: Invited talk on Advancing Time Series Analysis with Unified Foundation Models at Rice University - CS Department!
- 10/2025: Invited talk on Advancing Time Series Analysis with Unified Foundation Models at Prairie View A&M University - Center of Excellence in Research and Education for Big Military Data Intelligence!
- 09/2025: Invited talk on Time Series Foundation Models at Caltech - Academia-Industry X!
- 07/2025: Invited talk on Time Series Foundation Models at UCSD & Abel.AI!
- 03/2025: Invited talk on New Frontiers in Time Series Foundation Models at University of British Columbia!
Past Achievements (click to expand)
- 10/2024: Invited talk on Time Series Foundation Model in QuantLLM Community!
- 10/2024: Invited talk on Time Series Foundation Model in Lehigh University!
- 09/2024: Invited talk on Time Series Foundation Model in Emory University!
- 09/2024: One paper on Simulator Calibration with Google Research accepted by NeurIPS!
- 06/2024: Started summer internship at Cubist as a Quantitative Researcher!
- 04/2024: Time Series Foundation Model - TEMPO (ICLR 2024) published on Github and Hugging Face!
- 04/2024: Selected for the 2024 Best Research Assistant Award in USC CS Department! (1 per department)
- 03/2024: Awarded a Graduate School Endowed Fellowship for the academic year!
01/2024: One paper accepted by WWW and two papers accepted by ICLR!
- 12/2023: One paper on multi-modal time series foundation model accepted by AAAI Mentored Undergraduate Research Program!
- 10/2023: One paper on Financial Time-Series Forecasting accepted by ICAIF as Oral paper!
- 06/2023: Joined MBZUAI as a visiting scholar, working with Prof. Kun Zhang!
- 06/2023: Invited talk at Google Sustainable Urban Mobility: Simulation and Optimization Workshop! Video \& Slides
- 02/2023: One paper on Representation Learning accepted by CVPR!
- 01/2023: One paper on Neural Operator accepted by ICLR!
- 12/2022: One paper on Time-Series Forecasting accepted by SDM!
- 11/2022: One paper on Causal Inference accepted by AAAI!
- 10/2022: One paper on Out-of-distribution benchmark accepted by NeurIPS DistShift!
- 09/2022: One paper on Causal Inference accepted by NeurIPS!
- 07/2022: One paper on Causal Inference accepted by ICML Continuous time workshop!
- 05/2022: One paper accepted by IEEE ICCCAS!
- 04/2022: One paper accepted by NAACL main conference!
- 03/2022: Published the pre-print survey paper of ‘Physics-Informed Machine Learning’!
- 02/2022: Accepted the Research Scientist Intern offer at Adobe Research!
- 01/2022: One paper accepted by PAKDD!
- 11/2021: Organized ICAIF 2021 workshop - Time Series in Finance!
Internships & Research Experience
- Summer 2025: Visiting Scholar at California Institute of Technology, working with Prof. Yisong Yue
- 2024-2025: Contributor to Humanity Unleashed’s open-source cooperative AI policymaking platform; helped lead the pretraining track for the foundation-model component.
- Summer 2024: Quantitative Researcher at Cubist Systematic Strategies
- Summer 2023: Research Assistant at MBZUAI, working with Prof. Kun Zhang and Prof. Biwei Huang on causality
- Summer 2022: Research Intern at Adobe Research, mentored by Dr. Zhaowen Wang
- Previous: Research Intern at Microsoft Research Lab Asia (MSRA) (twice), working with Dr. Yujing Wang and Alibaba Damo Academy’s Data Analytics and Intelligence Lab, advised by Dr. Jingren Zhou
- First Internship: Baidu, 7 years ago
Awards
- 2025, 1st Place on GIFT-EVAL Leaderboard, Salesforce
- 2024, Best Poster Award, USC
- 2024, Best Research Assistant Award, USC (1 per department)
- 2024, Endowed Fellowship, USC
- 2021, Annenberg Fellowship, USC
- 2021, MSRA Stars of Tomorrow Internship Program Award, Microsoft
- 2020, National Scholarship for Graduate Students, China
- 2020, Excellent Student Exchange Scholarship, Peking University
- 2020, Huawei Fellowship
- 2019, Intel Fellowship
- 2019, HeHE Fellowship
- Travel Fellowship, ICLR/NeurIPS/ICML/AAAI/SDM