News

Isaac Bilsel – Barbara Luchsinger Memorial Scholar
2nd Year Scholar, PhD Candidate

Computer Engineering
George Washington University

Research: 
Modern large language models use a Mixture-of-Experts (MoE) approach to scale to massive sizes by activating only parts of the model at a time. While this reduces computation, it creates complex communication between different computing units, leading to performance bottlenecks. This work investigates hardware-software co-design strategies to support these MoE communication patterns. 

How Will Your Research Benefit Society?
As machine learning models develop to process more and more data, traditional general purpose hardware is failing to keep up, causing excess energy consumption with devastating environmental and economic impacts. The current rate at which LLM model size and usage is scaling is unsustainable without drastic change in the underlying hardware. The development of AI accelerators can reduce global energy consumption, and allow for these models to be implemented on various smart devices to improve productivity.

How Will an ARCS Award Benefit Your Research?
The ARCS award would be primarily used for attending conferences to gain insight on modern trends and interact with experts in computer architecture. These conferences are essential to gaining a sound understanding into the state of current research. Additionally, because this field is rapidly changing, conference papers are generally valued more than publications in journals or books. 

Career Objectives: 
In the future, I hope to work in a research and design oriented role in ASIC hardware development. My aim is to develop low energy solutions for processing machine learning workloads in order to reduce the growing environmental impact of modern data-driven applications.