About

Simon Shaolei Du is an assistant professor in the Paul G. Allen School of Computer Science & Engineering at the University of Washington. His research interests are broadly in machine learning, including deep learning, representation learning, reinforcement learning, and data selection. Prior to his faculty position, Du was a postdoctoral researcher at the Institute for Advanced Study of Princeton, hosted by Sanjeev Arora. He completed his Ph.D. in Machine Learning at Carnegie Mellon University, where he was co-advised by Aarti Singh and Barnabás Póczos. His academic background also includes studies in EECS and EMS at UC Berkeley, and he has spent time at the Simons Institute as well as research labs of Meta, Google, and Microsoft.

Research topics

• Computer Science
• Machine Learning
• Artificial Intelligence
• Mathematics
• Applied mathematics
• Geometry
• Mathematical optimization
• Mathematical analysis
• Theoretical computer science

Selected publications

• Argus: Evidence Assembly for Scalable Deep Research Agents

  ArXiv.org · 2026-05-15

  articleOpen access

  Deep research agents have achieved remarkable progress on complex information seeking tasks. Even long ReAct style rollouts explore only a single trajectory, while recent state of the art systems scale inference time compute via parallel search and aggregation. Yet deep research answers are composed of complementary pieces of evidence, which parallel rollouts often duplicate rather than complete, yielding diminishing returns while pushing the aggregation context toward the model's limit. We propose Argus, an agentic system in which a Searcher and a Navigator cooperate to treat deep research as assembling a jigsaw from complementary evidence pieces, rather than brute forcing the whole answer in parallel. The Searcher collects evidence traces for a given sub-query through ReAct-style interaction. The Navigator maintains a shared evidence graph, verifying which pieces are still missing, dispatching Searchers to gather them, and reasoning over the completed graph to produce a source-traced final answer. We train the Navigator with reinforcement learning to verify, dispatch, and synthesize, while independently training the Searcher to remain a standard ReAct agent. The resulting Navigator supports rollouts with a single Searcher or many in parallel without retraining. With both Searcher and Navigator built on a 35B-A3B MoE backbone, Argus gains 5.5 points with a single Searcher and 12.7 points with 8 parallel Searchers, averaged over eight benchmarks. With 64 Searchers it reaches 86.2 on BrowseComp, surpassing every proprietary agent we benchmark, while the Navigator's reasoning context stays under 21.5K tokens.

  PublisherOA PDF

• Cold-Start Personalization via Training-Free Priors from Structured World Models

  ArXiv.org · 2026-02-16

  articleOpen access

  Cold-start personalization requires inferring user preferences through interaction when no user-specific historical data is available. The core challenge is a routing problem: each task admits dozens of preference dimensions, yet individual users care about only a few, and which ones matter depends on who is asking. With a limited question budget, asking without structure will miss the dimensions that matter. Reinforcement learning is the natural formulation, but in multi-turn settings its terminal reward fails to exploit the factored, per-criterion structure of preference data, and in practice learned policies collapse to static question sequences that ignore user responses. We propose decomposing cold-start elicitation into offline structure learning and online Bayesian inference. Pep (Preference Elicitation with Priors) learns a structured world model of preference correlations offline from complete profiles, then performs training-free Bayesian inference online to select informative questions and predict complete preference profiles, including dimensions never asked about. The framework is modular across downstream solvers and requires only simple belief models. Across medical, mathematical, social, and commonsense reasoning, Pep achieves 80.8% alignment between generated responses and users' stated preferences versus 68.5% for RL, with 3-5x fewer interactions. When two users give different answers to the same question, Pep changes its follow-up 39-62% of the time versus 0-28% for RL. It does so with ~10K parameters versus 8B for RL, showing that the bottleneck in cold-start elicitation is the capability to exploit the factored structure of preference data.

  PublisherOA PDF

• FeDMRA: Federated Incremental Learning with Dynamic Memory Replay Allocation

  arXiv (Cornell University) · 2026-03-30

  preprintOpen accessSenior author

  In federated healthcare systems, Federated Class-Incremental Learning (FCIL) has emerged as a key paradigm, enabling continuous adaptive model learning among distributed clients while safeguarding data privacy. However, in practical applications, data across agent nodes within the distributed framework often exhibits non-independent and identically distributed (non-IID) characteristics, rendering traditional continual learning methods inapplicable. To address these challenges, this paper covers more comprehensive incremental task scenarios and proposes a dynamic memory allocation strategy for exemplar storage based on the data replay mechanism. This strategy fully taps into the inherent potential of data heterogeneity, while taking into account the performance fairness of all participating clients, thereby establishing a balanced and adaptive solution to mitigate catastrophic forgetting. Unlike the fixed allocation of client exemplar memory, the proposed scheme emphasizes the rational allocation of limited storage resources among clients to improve model performance. Furthermore, extensive experiments are conducted on three medical image datasets, and the results demonstrate significant performance improvements compared to existing baseline models.

  PublisherDOI

• Cold-Start Personalization via Training-Free Priors from Structured World Models

  Open MIND · 2026-02-16

  preprint

  Cold-start personalization requires inferring user preferences through interaction when no user-specific historical data is available. The core challenge is a routing problem: each task admits dozens of preference dimensions, yet individual users care about only a few, and which ones matter depends on who is asking. With a limited question budget, asking without structure will miss the dimensions that matter. Reinforcement learning is the natural formulation, but in multi-turn settings its terminal reward fails to exploit the factored, per-criterion structure of preference data, and in practice learned policies collapse to static question sequences that ignore user responses. We propose decomposing cold-start elicitation into offline structure learning and online Bayesian inference. Pep (Preference Elicitation with Priors) learns a structured world model of preference correlations offline from complete profiles, then performs training-free Bayesian inference online to select informative questions and predict complete preference profiles, including dimensions never asked about. The framework is modular across downstream solvers and requires only simple belief models. Across medical, mathematical, social, and commonsense reasoning, Pep achieves 80.8% alignment between generated responses and users' stated preferences versus 68.5% for RL, with 3-5x fewer interactions. When two users give different answers to the same question, Pep changes its follow-up 39-62% of the time versus 0-28% for RL. It does so with ~10K parameters versus 8B for RL, showing that the bottleneck in cold-start elicitation is the capability to exploit the factored structure of preference data.

  DOI

• Argus: Evidence Assembly for Scalable Deep Research Agents

  arXiv (Cornell University) · 2026-05-15

  preprintOpen access

  Deep research agents have achieved remarkable progress on complex information seeking tasks. Even long ReAct style rollouts explore only a single trajectory, while recent state of the art systems scale inference time compute via parallel search and aggregation. Yet deep research answers are composed of complementary pieces of evidence, which parallel rollouts often duplicate rather than complete, yielding diminishing returns while pushing the aggregation context toward the model's limit. We propose Argus, an agentic system in which a Searcher and a Navigator cooperate to treat deep research as assembling a jigsaw from complementary evidence pieces, rather than brute forcing the whole answer in parallel. The Searcher collects evidence traces for a given sub-query through ReAct-style interaction. The Navigator maintains a shared evidence graph, verifying which pieces are still missing, dispatching Searchers to gather them, and reasoning over the completed graph to produce a source-traced final answer. We train the Navigator with reinforcement learning to verify, dispatch, and synthesize, while independently training the Searcher to remain a standard ReAct agent. The resulting Navigator supports rollouts with a single Searcher or many in parallel without retraining. With both Searcher and Navigator built on a 35B-A3B MoE backbone, Argus gains 5.5 points with a single Searcher and 12.7 points with 8 parallel Searchers, averaged over eight benchmarks. With 64 Searchers it reaches 86.2 on BrowseComp, surpassing every proprietary agent we benchmark, while the Navigator's reasoning context stays under 21.5K tokens.

  PublisherDOI

• CARDIAC-FM: A Multimodal Foundation Model for Cardiovascular Risk Prediction Using ECG and Cardiac MRI

  medRxiv · 2026-03-18

  articleOpen access

  Atrial fibrillation and heart failure impose substantial health burdens worldwide, yet existing prediction models lack sufficient accuracy and generalizability. We developed CARDIAC-FM, a multimodal foundation model that learns joint representations of 12-lead electrocardiogram (ECG) and cardiac magnetic resonance imaging (MRI) through contrastive learning. We trained CARDIAC-FM on 57,609 paired ECG-cardiac MRI samples from UK Biobank and evaluated it in two external cohorts: the Cardiovascular Health Study (CHS) and the Multi-Ethnic Study of Atherosclerosis (MESA). CARDIAC-FM consistently outperformed unimodal models across all cohorts, and jointly incorporating ECG features with established clinical risk scores yielded additive gains in discrimination, indicating that ECG and traditional risk factors capture complementary dimensions of cardiovascular risk. The learned representations improved prediction across a range of cardiovascular outcomes with minimal task-specific fine-tuning, reflecting real-world settings where many diseases have limited positive samples and lack dedicated risk models. Although trained on paired ECG and MRI data, CARDIAC-FM generates predictions using ECG alone or ECG combined with established risk scores, enabling broad clinical deployment without MRI. These findings demonstrate the promise of multimodal pre-training for generalizable cardiovascular risk prediction.

  PublisherDOI

• FeDMRA: Federated Incremental Learning with Dynamic Memory Replay Allocation

  arXiv (Cornell University) · 2026-03-30

  articleOpen accessSenior author

  In federated healthcare systems, Federated Class-Incremental Learning (FCIL) has emerged as a key paradigm, enabling continuous adaptive model learning among distributed clients while safeguarding data privacy. However, in practical applications, data across agent nodes within the distributed framework often exhibits non-independent and identically distributed (non-IID) characteristics, rendering traditional continual learning methods inapplicable. To address these challenges, this paper covers more comprehensive incremental task scenarios and proposes a dynamic memory allocation strategy for exemplar storage based on the data replay mechanism. This strategy fully taps into the inherent potential of data heterogeneity, while taking into account the performance fairness of all participating clients, thereby establishing a balanced and adaptive solution to mitigate catastrophic forgetting. Unlike the fixed allocation of client exemplar memory, the proposed scheme emphasizes the rational allocation of limited storage resources among clients to improve model performance. Furthermore, extensive experiments are conducted on three medical image datasets, and the results demonstrate significant performance improvements compared to existing baseline models.

  PublisherOA PDF

• Settling the Sample Complexity of Online Reinforcement Learning

  Journal of the ACM · 2025-05-02 · 2 citations

  articleOpen accessSenior author

  A central issue lying at the heart of online reinforcement learning (RL) is data efficiency. While a number of recent works achieved asymptotically minimal regret in online RL, the optimality of these results is only guaranteed in a “large-sample” regime, imposing enormous burn-in cost in order for their algorithms to operate optimally. How to achieve minimax-optimal regret without incurring any burn-in cost has been an open problem in RL theory. We settle this problem for finite-horizon inhomogeneous Markov decision processes. Specifically, we prove that a modified version of MVP (Monotonic Value Propagation), an optimistic model-based algorithm proposed by Zhang et al. [82], achieves a regret on the order of (modulo log factors) \begin{equation*} \min \big \lbrace \sqrt {SAH^3K}, \,HK \big \rbrace, \end{equation*} where S is the number of states, A is the number of actions, H is the horizon length, and K is the total number of episodes. This regret matches the minimax lower bound for the entire range of sample size K ≥ 1, essentially eliminating any burn-in requirement. It also translates to a PAC sample complexity (i.e., the number of episodes needed to yield ε-accuracy) of \(\frac{SAH^3}{\varepsilon ^2} \) up to log factor, which is minimax-optimal for the full ε-range. Further, we extend our theory to unveil the influences of problem-dependent quantities like the optimal value/cost and certain variances. The key technical innovation lies in a novel analysis paradigm (based on a new concept called “profiles”) to decouple complicated statistical dependency across the sample trajectories — a long-standing challenge facing the analysis of online RL in the sample-starved regime.

  PublisherDOI

• Optimal Multi-Distribution Learning

  Journal of the ACM · 2025-08-25

  articleOpen access

  Multi-distribution learning (MDL), which seeks to learn a shared model that minimizes the worst-case risk across k distinct data distributions, has emerged as a unified framework in response to the evolving demand for robustness, fairness, multi-group collaboration, and so on. Achieving data-efficient MDL necessitates adaptive sampling, also called on-demand sampling, throughout the learning process. However, there exist substantial gaps between the state-of-the-art upper and lower bounds on the optimal sample complexity. Focusing on a hypothesis class of Vapnik–Chervonenkis (VC) dimension d , we propose a novel algorithm that yields an ɛ-optimal randomized hypothesis with a sample complexity on the order of \(\frac{d+k}{\varepsilon ^2}\) (modulo some logarithmic factor), matching the best-known lower bound. Our algorithmic ideas and theory are further extended to accommodate Rademacher classes. The proposed algorithms are oracle-efficient, which access the hypothesis class solely through an empirical risk minimization oracle. Additionally, we establish the necessity of improper learning, revealing a large sample size barrier when only deterministic, proper hypotheses are permitted. These findings resolve three open problems presented in COLT 2023 (i.e., Awasthi et al. [ 4 , Problems 1, 3, and 4]).

  PublisherDOI

• Unregularized Linear Convergence in Zero-Sum Game from Preference Feedback

  arXiv (Cornell University) · 2025-12-31

  preprintOpen accessSenior author

  Aligning large language models (LLMs) with human preferences has proven effective for enhancing model capabilities, yet standard preference modeling using the Bradley-Terry model assumes transitivity, overlooking the inherent complexity of human population preferences. Nash learning from human feedback (NLHF) addresses this by framing non-transitive preferences as a two-player zero-sum game, where alignment reduces to finding the Nash equilibrium (NE). However, existing algorithms typically rely on regularization, incurring unavoidable bias when computing the duality gap in the original game. In this work, we provide the first convergence guarantee for Optimistic Multiplicative Weights Update ($\mathtt{OMWU}$) in NLHF, showing that it achieves last-iterate linear convergence after a burn-in phase whenever an NE with full support exists, with an instance-dependent linear convergence rate to the original NE, measured by duality gaps. Compared to prior results in Wei et al. (2020), we do not require the assumption of NE uniqueness. Our analysis identifies a novel marginal convergence behavior, where the probability of rarely played actions grows exponentially from exponentially small values, enabling exponentially better dependence on instance-dependent constants than prior results. Experiments corroborate the theoretical strengths of $\mathtt{OMWU}$ in both tabular and neural policy classes, demonstrating its potential for LLM applications.

  PublisherDOI

Frequent coauthors

• Jason D. Lee

  38 shared

• Ruosong Wang

  Beijing Institute of Technology

  33 shared

• Barnabás Póczos

  17 shared

• Yining Wang

  The University of Texas at Dallas

  17 shared

• Lin F. Yang

  16 shared

• Kevin Jamieson

  16 shared

• Qiwen Cui

  13 shared

• Sham M. Kakade

  12 shared

Education

• Ph.D., Machine Learning

  Carnegie Mellon University

• B.S., EECS and EMS

  UC Berkeley