Ting-Yun (Charlotte) Chang
Hi, I'm Ting-Yun Chang 張婷雲. I work on the Copilot Tuning team at Microsoft. I completed my PhD in 2026 at USC CS, where I was coadvised by Robin Jia and Jesse Thomason. Overall, I am interested in discovering scientific insights and using them to improve large language models post-training. I've thought a lot about how to make scientific findings actionable, and I believe that efficiency follows naturally: when we have a deep understanding of the model, we can figure out where to be frugal w/o hurting model accuracy. I have been working on targeted weight updates, model quantization, and KV cache compression, with an emphasis on identifying the root causes of performance degradation arising from these approximations.
Previously, I did my bachelor's and master's degrees in Taiwan, both in Computer Science. I was advised by Yun-Nung (Vivian) Chen at National Taiwan University and Chi-Jen Lu at Academia Sinica.
Hi, I'm Ting-Yun Chang 張婷雲. I work on the Copilot Tuning team at Microsoft. I completed my PhD in 2026 at USC CS, where I was coadvised by Robin Jia and Jesse Thomason. Overall, I am interested in discovering scientific insights and using them to improve large language models post-training. I've thought a lot about how to make scientific findings actionable, and I believe that efficiency follows naturally: when we have a deep understanding of the model, we can figure out where to be frugal w/o hurting model accuracy. I have been working on targeted weight updates, model quantization, and KV cache compression, with an emphasis on identifying the root causes of performance degradation arising from these approximations.
Previously, I did my bachelor's and master's degrees in Taiwan, both in Computer Science. I was advised by Yun-Nung (Vivian) Chen at National Taiwan University and Chi-Jen Lu at Academia Sinica.
