Rachit Agarwal and Alexander Sasha Rush have each received Sloan Research Fellowships. Agarwal, an assistant professor in the Department of Computer Science, and Rush an associate professor of Computer Science at Cornell Tech are heralded as being among "128 of the brightest young researchers working in the U.S. and Canada." The funding program, sponsored by the Alfred P. Sloan Foundation, aims to "stimulate fundamental research by early-career scientists and scholars of outstanding promise."
As reported in the Cornell Chronicle by Tom Fleischman:
Agarwal’s recent research has focused on building a software ecosystem that enables applications to achieve the bare-metal high performance of modern hardware. Today’s software ecosystem is unable to achieve this goal due to its server-centric architecture: It organizes computation, memory and storage resources around server boundaries, and optimizes for performance at individual servers. As a result, for large-scale distributed applications, the performance can be far from global optimum.
He has been leading research on resource disaggregation, which offers distributed applications the illusion of having computer, memory and storage resources as large individual resource pools that can be “stitched” together over a network fabric. Blurring the server boundaries allows near-perfect resource utilization by matching available resources with application demands. Realizing this vision requires innovation across the entire stack, from hardware to operating systems to distributed systems.
Rush studies natural language processing (NLP), which has rapidly grown from a specialized area of AI to an applied technology with real-world impact. His research spans this progression, and as tools such as translation, question-answering and summarization become widely deployed, many important and understudied challenges have come to light based on the user experiences.
Rush’s research aims to develop deep generative models combining two areas of machine learning: deep learning, which excels at flexible representation learning from large datasets; and probabilistic modeling, which provides tools for principled declarative specification of missing information. The ultimate goal for the next generation of NLP technology is to allow users to seamlessly analyze unstructured documents, interact through new interfaces and communicate across languages.
In related coverage, read about Agarwal's distinguished paper award at USENIX and Rush's successful transition of the ICLR conference to a virtual format.