Yimeng (Damon) Zhang

Applied Scientist · Amazon · LLM Agents · Tool Use · Reliability

Santa Clara, CA

01 — About

  1. B.Eng. Electrical & Electronic Engineering University of Sheffield, UK 2018
  2. M.Sc. Electrical Engineering Columbia University, NY 2020
  3. Ph.D. Computer Science Michigan State University, MI 2025 (advised by Prof. Sijia Liu)
  4. Applied Scientist Multimodal AI Agents Amazon, Santa Clara, CA 2025 — now

02 — Research

Agents fail in boring ways. They call the wrong API, malform an argument, assert a fact they have no business being sure about — and then carry on as if nothing happened. I work on the data and rewards that make AI agents reliable enough to act on real systems, and on measuring how much of what they say can be trusted. Not by making the models bigger.

Agents that use tools

Agents fail at tool calls for unglamorous reasons: malformed arguments, the wrong API, no recovery. I generate verifiable tool-call trajectories from real APIs at scale, and train on failure rather than only on success.

Firefly · NeurIPS 2026 SENTINEL Trajectory2Task · ACL 2026

Agents that simulate people

If an agent reproduces how a real shopper browses, hesitates and buys, a product can be evaluated without running a study on humans. I train LLM and VLM agents to simulate behaviour, with rewards grounded in observed sessions.

Shop-R1 · ICLR 2026 Customer-R1 See, Think, Act

Measuring what models hide

Token-level confidence is noise and sequence-level confidence cannot point at the error. I estimate uncertainty over semantically coherent spans, and study what edits to a model leave behind in its outputs.

SpanUQ · NeurIPS 2026 Unlearning Traces · ICLR 2026 Reverse Engineering · ICLR 2022

Unlearning & concept erasure

Making a generative model forget a concept is easy to claim and hard to verify. I build erasure methods that survive adversarial prompting, and the attacks and benchmarks that show when they do not.

AdvUnlearn · NeurIPS 2024 UnlearnDiffAtk · ECCV 2024 UnlearnCanvas · NeurIPS 2024 SOUL · EMNLP 2024 Dual Power

Earlier work — zeroth-order optimisation and memory-efficient training (DeepZero, ZO-LLM benchmark, Black-Box Defense), dataset pruning and coreset selection (Selectivity Drives Productivity), and diffusion personalisation (ID-Patch · CVPR 2025).

03 — Service

04 — Publications

Research Summary Google Scholar * equal contribution

2026

  1. SpanUQ: Span-Level Uncertainty Quantification for Large Language Model GenerationNeurIPS 2026

    Y. Zhang, Y. Zhuang, Z. Wang, Y. Lu, P. Chen, A. Gupta, Z. Su, M. Tan, Z. Zhang, Q. Liu, M. Ramanathan, R. Maragoud, E. Vul, J. Huang, D. Wang

    PaperCodeModelDataset
  2. SENTINEL: Failure-Driven Reinforcement Learning for Training Tool-Using Language Model AgentsarXiv

    Z. Wang, Y. Lu, Y. Zhang, Q. Liu, C. Luo, J. Gesi, H. Lu, Y. Sang, M. Li, J. Huang, D. Wang

    Paper
  3. Firefly: Illuminating Large-Scale Verified Tool-Call Data Generation from Real APIsNeurIPS 2026

    Y. Lu, Z. Wang, Y. Lu, Y. Sang, J. Gesi, X. Tang, Y. Zhang, Z. Dai, H. Liu, H. Lu, C. Luo, Q. He, B. Dumoulin, J. Huang, D. Wang

    Paper
  4. Trajectory2Task: Training Robust Tool-Calling Agents with Synthesized Yet Verifiable Data for Complex User IntentsACL 2026

    Z. Wang, Y. Lu, Y. Zhang, P. Chen, Z. Dong, J. Huang, J. Gesi, X. Tang, C. Luo, Q. Liu, Y. Sang, H. Lu, M. Li, J. Lai, D. Wang

    Paper
  5. Shop-R1: Rewarding LLMs to Simulate Human Behavior in Online Shopping via Reinforcement LearningICLR 2026

    Y. Zhang, T. Wang, J. Gesi, Z. Wang, Y. Lu, J. Lin, S. Zhan, V. Gao, R. Jiao, J. Liu, K. Qian, Y. Tang, R. Xue, H. Zhang, Q. Cui, Y. Guo, D. Wang

    PaperDataset
  6. Unlearning Isn’t Invisible: Detecting Unlearning Traces in LLMs from Model OutputsICLR 2026

    Y. Chen, S. Pal, Y. Zhang, Q. Qu, S. Liu

2025

  1. Customer-R1: Personalized Simulation of Human Behaviors via RL-based LLM Agent in Online ShoppingarXiv

    Z. Wang, Y. Lu, Y. Zhang, J. Huang, D. Wang

    Paper
  2. See, Think, Act: Online Shopper Behavior Simulation with VLM AgentsarXiv

    Y. Zhang, J. Gesi, R. Xue, T. Wang, Z. Wang, Y. Lu, S. Zhan, H. Zeng, Q. Cui, Y. Guo, J. Huang, M. Shah, D. Wang

  3. The Dual Power of Interpretable Token Embeddings: Jailbreaking Attacks and Defenses for Diffusion Model UnlearningarXiv

    S. Chen, Y. Zhang, S. Liu, Q. Qu

  4. ID-Patch: Robust ID Association for Group Photo PersonalizationCVPR 2025

    Y. Zhang, T. Zhi, J. Liu, S. Sang, L. Jiang, Q. Yan, S. Liu, L. Luo

    PaperCodeHF ModelSlidePosterDemo

2024

  1. Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion ModelsNeurIPS 2024

    Y. Zhang, X. Chen, J. Jia, Y. Zhang, C. Fan, J. Liu, M. Hong, K. Ding, S. Liu

    CodeHF ModelDemoPosterSlide
  2. To Generate or Not? Safety-Driven Unlearned Diffusion Models Are Still Easy To Generate Unsafe Images … For NowECCV 2024

    Y. Zhang*, J. Jia*, X. Chen, A. Chen, Y. Zhang, J. Liu, K. Ding, S. Liu

    CodeDemoPosterUnlearned DM BenchmarkSlide
  3. UnlearnCanvas: A Stylized Image Dataset to Benchmark Machine Unlearning for Diffusion ModelsNeurIPS 2024

    Y. Zhang, C. Fan, Y. Zhang, Y. Yao, J. Jia, J. Liu, X. Liu, S. Liu

    PaperPosterCodeBenchmarkDataset
  4. SOUL: Unlocking the Power of Second-Order Optimization for LLM UnlearningEMNLP 2024

    J. Jia, Y. Zhang, Y. Zhang, J. Liu, B. Runwal, J. Diffenderfer, B. Kailkhura, S. Liu

    CodePosterSlide
  5. Revisiting Zeroth-Order Optimization for Memory-Efficient LLM Fine-Tuning: A BenchmarkICML 2024

    Y. Zhang*, P. Li*, J. Hong*, J. Li, Y. Zhang, W. Zheng, P.-Y. Chen, J. D. Lee, W. Yin, M. Hong, Z. Wang, S. Liu, T. Chen

    CodePoster
  6. DeepZero: Scaling up Zeroth-Order Optimization for Deep Model TrainingICLR 2024

    A. Chen*, Y. Zhang*, J. Jia, J. Diffenderfer, J. Liu, K. Parasyris, Y. Zhang, Z. Zhang, B. Kailkhura, S. Liu

    CodeSlidePoster
  7. The power of few: accelerating and enhancing data reweighting with coreset selectionICASSP 2024

    M. Jafari, Y. Zhang, Y. Zhang, S. Liu

2023

  1. Selectivity Drives Productivity: Efficient Dataset Pruning for Enhanced Transfer LearningNeurIPS 2023

    Y. Zhang*, Y. Zhang*, A. Chen*, J. Jia, J. Liu, G. Liu, M. Hong, S. Chang, S. Liu

    PaperCodePosterSlide
  2. Text-Visual Prompting for Efficient 2D Temporal Video GroundingCVPR 2023

    Y. Zhang, X. Chen, J. Jia, S. Liu, K. Ding

    CodeHF ModelPosterSlide
  3. Data-Model-Circuit Tri-Design for Ultra-Light Video Intelligence on Edge DevicesASP-DAC 2023

    Y. Zhang*, A.K. Kamath*, Q. Wu*, Z. Fan*, W. Chen, Z. Wang, S. Chang, S. Liu, C. Hao

    Slide

2022

  1. How to Robustify Black-Box ML Models? A Zeroth-Order Optimization PerspectiveICLR 2022Spotlight · top 5%

    Y. Zhang, Y. Yao, J. Jia, J. Yi, M. Hong, S. Chang, S. Liu

    CodePosterSlide
  2. Reverse Engineering of Imperceptible Adversarial Image PerturbationsICLR 2022

    Y. Gong, Y. Yao, Y. Li, Y. Zhang, X. Liu, X. Lin, S. Liu

    Code

2020

  1. Video Synthesis via Transform-Based Tensor Neural NetworkACM MM 2020

    Y. Zhang, X. Y. Liu, B. Wu, A. Walid

  2. Tensor FISTA-Net for real-time snapshot compressive imagingAAAI 2020

    X. Han, B. Wu, Z. Shou, X. Y. Liu, Y. Zhang, L. Kong