Hi, I’m Yuxin Yang.
I am a Ph.D. student at USC, advised by Prof. Viktor Prasanna. I work on large language models, agentic AI systems, generative recommendation, and graph machine learning.
Before USC, I received my B.Eng. in Automation from Tsinghua University. I have had the opportunity to work with Prof. Muhan Zhang, Prof. Yitao Liang, and Prof. Yilin Mo.
Experience, Education & Support
Capital One Ph.D. Fellowship
2024 – 2027
Research supported by the Capital One Fellowship in 2024–2025, 2025–2026, and 2026–2027.
ByteDance Summer Intern
–
LLM-native generative recommendation. Used semantic ID (SID) sequences from users’ historical item interactions as LLM inputs to directly predict the SID of the next item a user would interact with.
Tsinghua University B.Eng. in Automation
Sep 2019 – Jun 2023
Department of Automation
BIGAI Research Intern
Nov 2021 – Jun 2022
Beijing Institute for General Artificial Intelligence. Worked with Prof. Muhan Zhang on parameter-adaptive graph neural networks.
Selected Publications
* Equal contribution Google Scholar ↗
EMNLP 2026Main Conference
SPARC-RAG: Adaptive Sequential-Parallel Scaling with Context Management for Retrieval-Augmented Generation
A multi-agent RAG framework that coordinates sequential and parallel reasoning through shared context management.
ICML 2026
SWE-Milestone: Evaluating AI Agents on Continuous Software Evolution
Evaluating coding agents across connected milestones in a continuously evolving codebase.
WSDM 2026
SAGERec: Sampling and Gating for Enhanced Long-Tail Item Recommendations
Learning which historical items to sample, with complementary experts and tail-aware gating for long-tail recommendation.
Preprint2025
Training Diverse Graph Experts for Ensembles: A Systematic Empirical Study
A systematic study of how training choices create complementary graph experts for more effective ensembles.
EACL 2026
RECIPE-TKG: From Sparse History to Structured Reasoning for LLM-based Temporal Knowledge Graph Completion
Combining richer historical context, contrastive learning, and test-time filtering for temporal knowledge graph reasoning.
VLDB 2025
Towards Ideal Temporal Graph Neural Networks: Evaluations and Conclusions after 10,000 GPU Hours
Over 10,000 GPU hours of controlled experiments to understand temporal GNN modules and their interactions.
ICML 2022DyNN Workshop, Spotlight
PA-GNN: Parameter-Adaptive Graph Neural Networks
Generating node-specific aggregation parameters to capture local patterns in graphs.
Beyond Research
In my spare time, I enjoy music, piano, and sports. I built an interactive webpage on harmony chords while learning music theory. I’m also exploring cycling and running routes throughout Los Angeles — find me on Strava ↗.