Chia-Yuan's Homepage

About me

I’m Chia-Yuan (Scott) Chang, an Applied Scientist at Amazon Rufus. I am broadly interested in how to efficiently post-train foundation models, particularly for RLVR and agentic coding RL for large-scale sparse Mixture-of-Experts (MoE) models.

My current work focuses on large-scale LLM post-training, especially reinforcement learning and supervised fine-tuning for sparse MoE models. At Amazon, I built the end-to-end coding RLVR recipe for our 100B–800B MoE models, spanning reward design and agentic coding with verifier in sandbox environments. I’ve also worked extensively on SFT training and data pipelines, as well as mid-training and long-context extension.

Before and beyond current work, my research interests include RAG, Long-context extension, and Generative Model Applications.

I received my Ph.D. in Computer Science from Texas A&M University, advised by Prof. Na Zou and Prof. Xia “Ben” Hu.

Email: cychang at tamu dot edu

News

• Aug. 2026: Our blog on how to prevent expert collapse in ultra-sparse MoE (8/768) training is out: Mitigate Silent Expert Death in Ultra-Sparse MoE , which is a small step towards more sparse MoE models!

• Jun. 2025: Start my full-time job as an applied scientist at Amazon Rufus team at Palo Alto, CA

• May. 2025: One paper accepted by ACL 2025 Main: MAIN-RAG: Multi-Agent Filtering Retrieval-Augmented Generation

• May. 2025: One paper accepted by KDD 2025: CODA: Temporal Domain Generalization via Concept Drift Simulator

• Nov. 2024: Start my internship as an applied scientist intern at Amazon Rufus team at Palo Alto, CA

• Jul. 2024: One paper accepted by TKDD

• Jun. 2024: One paper is selected as a Spotlight Paper by ICML 2024: LLM Maybe LongLM: Self-Extend LLM Context Window Without Tuning [GitHub]

• May 2024: Start my internship as a research intern at Visa Research at Foster City, CA

• May 2024: Two papers accepted by ICML 2024

• Mar. 2024: One paper accepted by NAACL 2024

• Oct. 2023: One paper accepted by NeurIPS 2023 AI for Science Workshop

• Oct. 2023: One paper selected as CIKM 2023 Best Demo Paper Honorable Mention

• Oct. 2023: One paper selected as AMIA 2023 Best Student Paper Finalist

• Aug. 2023: One paper accepted by CIKM 2023

• Aug. 2023: Give an invited tutorial on Machine Learning in Finance Workshop at KDD 2023
  [Website][Slides][Code]

• Jul. 2023: Two papers accepted by AMIA 2023

• Jun. 2023: Give an invited tutorial on Fairness in Machine Learning for Healthcare at QPRC 2023

• Jun. 2023: Receive NSF Travel Award for Quality and Productivity Research Conference (QPRC 2023)

• Feb. 2021: One paper accepted by AAAI 2021

• Oct. 2020: One paper accepted by TREC 2020 (2^nd Place Award) [Website]