時間：114年11月13日(星期四) 15:10-17:00

地點：大仁樓301

主持人：彭彥璁老師

演講者: Google  Sr. Software Engineer 嚴梓鴻 博士

演講題目: In the AI era, many students and researchers focus primarily on making or using AI models at a high level. They often miss the essential computer science fundamentals that make AI fast and efficient. This talk argues that the biggest breakthroughs in AI now depend on core system engineering. To truly innovate, scale, and speed up AI technology, you need a strong, complete understanding of Computer Architecture, Compilers, Operating Systems (OS), and Algorithms.

We will take a journey through the computer science stack, from specialized hardware to advanced software, showing how basic ideas are directly linked to AI success. This includes: Computer Architecture (understanding memory, data types like BF16, and specialized hardware like TPUs/NPUs for maximum speed); the Compiler (which acts as a smart translator to arrange tasks and data for fast code execution); and the Operating System (OS) (which manages resources, handles different hardware, and runs the entire AI model executable at a large scale).

The main goal of this talk is to show graduate students why investing time in foundational computer science is essential. We want to inspire you to move beyond simply using existing AI tools. Deep knowledge of these fundamentals—Architecture, Compilers, OS, and the principles of Algorithms (like ground truth definition and verification)—is the key to future industry leadership. Mastering these basics will give you the power to design faster, next-generation systems and solve difficult problems that current high-level tools cannot address.

Speaker Bio:

Champ Yen is a Software Engineer at Google, where he previously served as a DSP Architect. He holds a Bachelor's degree in Computer Science and Information Engineering from National Cheng Kung University and a Master's from National Chiao Tung University.

Champ brings 18 years of expertise in embedded systems, drivers, and firmware development from the IC industry, with previous roles at Novatek, MediaTek, OnePlus, Qualcomm, and Cadence. He has significant experience in performance optimization, leveraging technologies like SIMD, GPGPU, DSP, and domain-specific programming.

In his recent work, Champ has focused on developing and optimizing software for multimedia and on-device machine learning applications

演講綱要: