About Me

I am a Ph.D. student in Computer Science at the University of Maryland, College Park, advised by Prof. Ang Li.

My research focuses on trustworthy AI, with a particular interest in how AI systems handle uncertainty, learn from data, and make decisions based on incomplete or imperfect information.

One question that drives my research is: How can we tell whether an AI system’s decisions are genuinely supported by the information it has learned or observed?

I have explored this question from different perspectives. In machine unlearning, I study whether models actually remove the influence of training data or simply stop producing certain outputs. In genomic privacy, I investigate how sensitive information can be exposed through learned models. More recently, I have been working on clinical AI agents, focusing on how they interpret uncertain patient responses, decide what information to gather next, and verify that their conclusions are supported by clinical evidence.

Beyond model reliability, I am also interested in making large-scale AI systems more efficient. My experience includes optimizing LLM inference through CUDA/HIP kernels, speculative prefill, and KV-cache optimization.

Ultimately, I want to develop AI systems that can reason reliably under uncertainty, recognize when they need more information, and provide decisions that we can meaningfully verify.

Research Interests

• Trustworthy AI & Machine Unlearning: Understanding how training data influences model behavior, distinguishing genuine knowledge removal from behavioral suppression, and developing methods to evaluate these effects.
• Uncertainty-Aware Reasoning & AI Agents: Representing uncertain evidence, updating beliefs, and deciding what information to seek during multi-step reasoning.
• Clinical AI & Evidence-Grounded Verification: Building diagnostic agents that handle incomplete patient information and verifying clinical findings against medical evidence.
• Privacy-Preserving Machine Learning: Investigating privacy leakage and information control in genomic foundation models and distributed learning.
• Efficient LLM Inference: Improving inference performance through GPU kernel optimization, speculative prefill, and memory-efficient computation.

Research & Publications

Under Review

Exposure Leaves a Trace: A Provenance Audit for Machine Unlearning
Daeun Jung and Ang Li.
International Conference on Learning Representations (ICLR 2027) — Under Review.

Privacy Risks of Federated Genomic Foundation Models for Multi-institutional Digital Medicine
Daeun Jung, Ziyao Wang, Jiaheng Guo, Song Wang, Tianlong Chen, and Ang Li.
npj Digital Medicine — Under Review, 2026.

FedMOA: Federated GRPO for Personalized Reasoning LLMs under Heterogeneous Rewards
Ziyao Wang*, Daeun Jung*, Yexiao He, Guoheng Sun, Zheyu Shen, Myungjin Lee, and Ang Li.
ACL Rolling Review (ARR), October 2026 Cycle — Under Review.
* Equal contribution.

International Conferences

• Feature Expansion of Single Dimensional Time Series Data for Machine Learning Classification — First author.
• Traffic Data Classification using Machine Learning Algorithms in SDN Networks — Co-first author.
• An Iterative Algorithm of Key Feature Selection for Multi-class Classification — First author.

Domestic Conferences

• Impact of Input Data Randomness on Training Performance of Autoencoder — Second author, Best Paper Award.
• On Improving Network Data Anomaly Detection Performance based on Meta Characteristics — Second author.
• Machine Learning-based Algorithm for Small-scale Multi-featured Data Classification — First author.
• Study on Impact of Class Combinations on Performance of Multi-class Classification — First author.

Domestic Journal Articles

• Survey on Machine Learning Algorithms for SDN/NFV Automation — Second author.

Experience

University of Maryland, College Park

Graduate Research Assistant | Aug. 2022 – Present

• Investigate causal data influence and representation-level auditing in large language models, particularly the distinction between knowledge removal and behavioral suppression.
• Study privacy risks in genomic foundation models and personalized reasoning under heterogeneous learning objectives.
• Develop uncertainty-aware clinical reasoning methods and evidence-grounded medical AI systems.

Teaching Assistant

Aug. 2022 – May 2023

• Advanced Data Structures (CMSC 420)
• Discrete Structures (CMSC 250)
• Led discussion sessions and supported students in understanding algorithms, data structures, and mathematical reasoning.

Runara.ai

Research Intern | May 2026 – Jul. 2026

• Optimized large language model inference using vLLM, CUDA, and HIP across NVIDIA H100 and AMD MI300X GPUs.
• Developed and benchmarked fused transformer kernels, achieving up to 2.17× single-operation speedup and 1.24× end-to-end speedup.
• Evaluated speculative prefill for GPT-OSS-120B, achieving 3.70–3.88× reduction in target-model time to first token in benchmarked settings.
• Investigated KV-cache optimization and quantization for efficient inference of large reasoning models.

Carnegie Mellon University

Visiting Scholar | Jan. 2020 – Jul. 2020

• Full scholarship recipient for the Intensive AI Program supported by the Ministry of Science and ICT.
• Developed NLP-based conversational systems and worked on large-scale multimedia generation using GANs and AWS.
• Led system architecture design and explored recommendation algorithms, including collaborative and content-based filtering.

Honors & Awards

• Best Paper Award, KICS Summer Conference, 2021.
• Research Assistant Scholarship, Ewha Womans University, 2020.
• Dean’s List, Ewha Womans University, 2015, 2017, 2018.