A group of scholars from the Cornell Ann S. Bowers College of Computing and Information Science are receiving grants from the strategic partnership between the college and LinkedIn, the world’s largest professional network, to drive innovation in areas like generative artificial intelligence (AI), AI architecture, and large language models (LLM).
This is the fifth and final year of the partnership, which was launched in 2022 with a multimillion-dollar grant from LinkedIn that established a research connection between Cornell Bowers’ leading AI scholars and LinkedIn’s scientists and engineers.
Fostering exploration and learnings across academia and business, the partnership was the first of its kind for both Cornell Bowers and LinkedIn, supporting a total of 40 pioneers from the college – 20 faculty members and 20 doctoral students, along with other university co-researchers – who furthered discoveries in areas like agentic AI, algorithmic fairness, privacy, computing efficiency, and more. The partnership supported research breakthroughs that led to safer clinical trials, more efficient prediction algorithms, and controllable, updatable AI systems, among other discoveries.
This year’s Bowers-LinkedIn grant recipients are:
Faculty
Rachee Singh, assistant professor of computer science, will develop a caching architecture to improve efficiency in AI agents and reduce latency. Singh’s project is called, “Stateful Tool-Value Caching for LLM agents,” and Emaad Manzoor, assistant professor of marketing at Cornell SC Johnson College of Business, is a co-principal investigator.
John Thickstun, assistant professor of computer science, will study text diffusion, an alternative approach to AI text generation that can generate text more quickly than standard methods but is currently harder to train. His project, “Shifting the Scaling Law for Text Diffusion,” aims to build improved text diffusion models that are easier to train.
Immanuel Trummer, associate professor of computer science, will explore the design of database systems that lower the costs incurred when AI agents, based on LLMs, perform iterative data analysis. The project is called “Making Agentic Data Analysis Token-Efficient.”
Angelina Wang, assistant professor of information science at Cornell Tech, will evaluate the performance of AI models used in online and offline settings. Most models are put to the test in isolated, offline environments, and Wang’s prior work has shown models behave differently in the wild. Wang’s project, “Behavioral Stability of LLM Agents in the Wild,” aims to better understand these output discrepancies between offline and online contexts.
Doctoral Students
Ali Behrouz, a Cornell Tech-based doctoral student in the field of computer science advised by Ramin Zabih, will develop faster and more efficient “transformers” – the architecture behind most modern AI systems – to analyze large amounts of data while using less power and computing memory. Behrouz’s project is called “Designing Next Generation of Deep Learning Architectures with Learning How to Memorize at Test Time.”
Cristiana Firullo, a doctoral student in the field of information science advised by Cristobal Cheyre, will study how AI-generated summaries reshape consumer search behavior, decision quality, and attention allocation across heterogeneous digital markets. Firullo’s project is called “AI-Generated Summaries and Consumer Search Across Digital Markets.”
Karuna Grewal, a doctoral student in the field of computer science advised by Justin Hsu, will develop SafeAge, a runtime monitor that brings formally grounded, real-time safety guardrails to agentic AI systems. Grewal’s project is called “Formally Securing Agentic AI Systems Using Runtime Verification.”
Haruka Kiyohara, a doctoral student in the field of computer science advised by Thorsten Joachims and Sarah Dean, will develop a framework for a smarter, LLM-based recommendation system that tweaks its recommendations as individual users change their preferences. The project is called “Personalized Generative Recommendation and RAG at Scale.”
Louis DiPietro is a writer for the Cornell Ann S. Bowers College of Computing and Information Science.