Eight scholars from the Cornell Ann S. Bowers College of Computing and Information Science with expertise in generative artificial intelligence (AI) are this year’s recipients of grants from the strategic partnership between the college and LinkedIn.
Launched in 2022 with a multimillion-dollar grant from LinkedIn, the five-year strategic partnership between the world’s largest professional network and Cornell Bowers advances and stewards cutting-edge research in AI. The partnership establishes a research connection between LinkedIn’s scientists and engineers and Cornell Bowers’ leading AI researchers. Awards to doctoral students include academic year funding and discretionary funds.
Faculty award winners
Yoav Artzi, associate professor of computer science at Cornell Tech, will explore an adaptive method that compresses documents for AI systems by keeping only information that is new to the AI, making the whole process more efficient. Artzi’s project is called “Adaptive Compression for Retrieval Augmented Generation.”
Volodymyr Kuleshov, the Joan Eliasoph, M.D. Assistant Professor of Computer Science at Cornell Tech, the Jacobs Technion-Cornell Institute, and Cornell Bowers, seeks to further develop diffusion language models, which are potentially faster and more efficient than traditional LLMs. His project is called “Diffusion Language Models that Support Fast and Efficient Parallel Generation.”
Jennifer Sun, assistant professor of computer science, aims to develop a new framework for large language models (LLMs) that will allow knowledge bases behind LLMs to be updated continuously with verified, accurate information. Updating information, verifying accuracy or correcting errors within today’s LLMs is expensive, and, in turn, models quickly become outdated, researchers said. Kilian Weinberger, professor of computer science, is a co-principal investigator on the project, which is called “Continuously Evolving Verifiable Memory for Large Models.”
Dana Yang, assistant professor of statistics and data science, will explore a generative AI framework to empower users to adjust the novelty of an AI’s output and generate outputs that are both accurate and original. Yang’s project is called “Innovation Guaranteed: A Framework to Induce Creativity in Generative AI.”
Doctoral student award winners
Dhruv Agarwal, a doctoral student in the field of computer science advised by Aditya Vashistha, will develop strategies to reduce cultural bias in LLMs that, though deployed globally, tend to skew toward Western cultural norms. In his previous work, Agarwal found that AI writing suggestions nudge non-Western users toward adopting Western writing styles. Agarwal’s project is called “Reinforcement Learning for Making LLMs Globally Relevant.”
Eric Enouen, a doctoral student in the field of computer science advised by Sainyam Galhotra, aims to make generative AI models more transparent, adaptable, and user-centered through the Concept Bottleneck framework, a machine learning approach that helps users to better understand and intervene on a model’s reasoning process. Enouen’s project is called “Transparent and Steerable Generative Models with Self-Refining Concept Bottlenecks.”
Byungsoo Oh, a doctoral student in the field of computer science advised by Rachee Singh, will design new systems techniques to improve the efficiency of Mixture-of-Experts language models by reducing GPU memory and communication overheads that currently limit their scalability. Oh’s project is called “Resource-Efficient Mixture-of-Experts LLMs via Pipeline-Parallel Prefetching.”
Linxi Zhao, a doctoral student in the field of computer science advised by Sun and Weinberger, will explore Large Memory Language Models (LMLMs) that store factual knowledge in external databases rather than within the model itself, making it easier to update information and protect privacy while maintaining factuality. Zhao’s project is called “Towards Editable, Reliable and Privacy-Preserving LLMs via Knowledge Offloading.”
Louis DiPietro is a writer for the Cornell Ann S. Bowers College of Computing and Information Science.