Designing and benchmarking agentic optimizers
First author · title withheld during double-blind review
Under review at ICLR 2027.
David Xia 夏锦源
I'm a senior at the University of Illinois Urbana-Champaign studying computer science, mathematics, and statistics. My current interests lie in AI for science, especially agents that learn to search and design using costly, imperfect evaluations. I'm applying to PhD programs for fall 2027, with broader interests in empirical reinforcement learning and language model agents.
I'm interested in how an agent chooses what to test and which evaluations to trust. My current focus is on multifidelity optimization and reinforcement learning for allocating a limited budget between experiments, model improvement, and planning. More broadly, I'm interested in empirical reinforcement learning and language model agents, including exploration, tool use, and learning strategies that transfer across tasks.
News
- Jul 2026 RSK linear operators and the Vershik–Kerov–Logan–Shepp curve was published in the Electronic Journal of Combinatorics.
- May 2026 Started a research internship at Northwestern University with Prof. Zhaoran Wang on agentic reinforcement learning.
- Apr 2026 Presented our LLM-journalism audit at the HEAL workshop at CHI 2026 in Barcelona.
- Feb 2026 Joined the iDEA-iSAIL Lab at UIUC with Prof. Hanghang Tong, working on theoretical RL.
- May 2025 EveGuard appeared at the IEEE Symposium on Security and Privacy 2025.
Publications
Under Review
Journal & Conference Publications
RSK linear operators and the Vershik–Kerov–Logan–Shepp curve
Authors in alphabetical order, as is standard in mathematics
EveGuard: Defeating Vibration-based Side-Channel Eavesdropping with Audio Adversarial Perturbations
Workshop Papers & Manuscripts
Frozen Policy Iteration for MDPs with Stochastic Transitions under Linear Qπ Realizability
Manuscript, 2026. [PDF]
Predictability in Major League Sports: Betting Odds versus Mathematical Models
Mathematics and Sports (MAS) Journal, 2026. Recommended for acceptance modulo minor revisions. [PDF]
Revitalizing Local Democracy: A Human-Centered Audit of LLMs in City Council Journalism
HEAL Workshop at CHI 2026 [PDF] [Workshop]
Curriculum Vitae
Education
University of Illinois Urbana-Champaign
Aug 2023 – May 2027 (expected)
Champaign, Illinois · GPA 3.99 / 4.00
- B.S. in Computer Science
- B.S. in Mathematics
- B.S. in Statistics
Research Experience
Northwestern University
May 2026 – present
Advised by Prof. Zhaoran Wang
- Developed and evaluated language model agents as black-box optimizers, studying how they propose solutions, interpret evaluation results, and revise their search strategies.
- First author of a paper under review at ICLR 2027. Project details withheld during double-blind review.
iDEA-iSAIL Lab
Feb 2026 – present
Advised by Prof. Hanghang Tong
- Studied sample-efficient exploration and function approximation in reinforcement learning, focusing on Markov decision processes with linearly representable policy values.
- Developed a manuscript extending Frozen Policy Iteration to stochastic transitions under linear Qπ realizability, with a sample-complexity guarantee under an additional coverage assumption.
- Currently investigating an algorithm for deterministic MDPs with improved sample-complexity guarantees, while extending the analysis to broader settings.
Illinois Geometry Lab
Jan 2024 – May 2026
Advised by Prof. AJ Hildebrand
- Led a study of sports predictability, comparing betting odds, Elo ratings, polls, and statistical models. Presented at Rose-Hulman in 2024 and the Joint Mathematics Meetings in 2025.
- Paper recommended for acceptance modulo minor revisions at Mathematics and Sports (MAS).
Illinois Combinatorics Lab for Undergraduate Experience (ICLUE)
Aug 2024 – May 2026
Advised by Prof. Alexander Yong
- Proved asymptotic probability bounds for Schensted insertion and RSK linear operators using the Vershik-Kerov-Logan-Shepp limit shape curve. Published in the Electronic Journal of Combinatorics in 2026.
Carnegie Mellon University, HCII
May 2025 – Feb 2026
Advised by Prof. Jeff Bigham
- Built an LLM pipeline for city council reporting, from transcript segmentation to headline generation and topic prioritization.
- Led crowdsourced evaluation of headline quality and topic prioritization. Presented the work at the HEAL workshop at CHI 2026.
University of California San Diego
Jun 2023 – Dec 2024
Advised by Prof. Xinyu Zhang
- Implemented EveGuard's domain-translation model for estimating eavesdropped audio, enabling end-to-end training of protective audio perturbations with limited data.
- The resulting system achieved over 97% protection against tested audio classifiers while preserving audio quality. Published at IEEE Security and Privacy 2025.
Teaching
Course Assistant, CS 374, UIUCSpring 2026, Fall 2026
Research Support
- 2025CMU HCII REU participant, supported by NSF award 2349558
- 2025ICLUE RTG participant, supported by NSF award 1937241
Relevant Coursework
Bold courses are graduate-level.
In progress (Fall 2026): Statistical Reinforcement Learning, Randomized Algorithms
CS: Machine Learning, Deep Learning, Natural Language Processing, Algorithms, Numerical Analysis, Numerical Methods
Math: Honors Real Analysis, Algebraic Combinatorics, Abstract Linear Algebra, Combinatorics, Graph Theory, Abstract Algebra, Complex Variables
Stats: Bayesian Analysis
Languages
English: Native
Mandarin Chinese: Fluent, HSK6