CICS Doctoral Candidates Alireza Salemi and Cuong Than Named 2025 Google PhD Fellows
Content
Two Manning College of Information and Computer Sciences (CICS) doctoral candidates, Alireza Salemi and Cuong Than, have been named recipients of the 2025 Google PhD Fellowship for their research in natural language processing and in algorithms and optimization, respectively.
The Google PhD fellowship is a prestigious annual award that provides exceptional graduate students with financial support, mentorship from a Google Research scientist, and opportunities to engage with broader research communities. Since its launch in 2009, the fellowship has supported over 950 students across 44 countries. This year, Google.org has committed over $10 million to 255 PhD students from 35 countries, including Salemi and Than.
Alireza Salemi: Toward Proactive Collaborative Multi-Agent Systems
Salemi’s fellowship research proposal, “Collaborative Learning in Agentic AI,” explores how feedback written in everyday language—known as Natural Language Feedback (NLF)—can build artificial intelligence (AI) teams that collaborate more like humans: dividing work, explaining mistakes, and improving together over time.
Modern large language models (LLMs) like ChatGPT or Google Gemini can power multi-agent frameworks in which AI workers (also known as agents) leverage the LLMs to perform a specialized role—for example, one might search for information while another summarizes it. These frameworks, called reactive systems, typically feature one central agent delegating subtasks to others, providing feedback through a single number known as a “scalar reward”—like getting a grade from a professor without any comments, which tells you how well you did but not how to improve.
Salemi’s research replaces that single-number feedback with NLF, where one agent (like a coach) identifies errors and suggests refinements using ordinary language. By using NLF, agents are no longer limited to simply scoring how well another completed a task—they can proactively explain what went wrong and how to fix it, improving learning over the long term.
Salemi envisions NLF as a pathway toward a “new paradigm” of more human-like proactive collaboration in multi-agent systems, where agents bid for subtasks and select collaborating agents based on factors such as cost, quality, and speed—much like how humans choose teammates.
Cuong Than: Optimizing Network Algorithms
Than’s fellowship research proposal, “Algorithmic Foundation of Compact Networks,” investigates how smaller sub-networks—known as spanners—can increase the efficiency, resiliency, and scalability of large-scale network systems like Google Maps. Spanners act as simplified versions of networks that preserve the approximate shortest distance between any two nodes (like road intersections). Think of them as a way for Google Maps to store a lighter “summary” of a road system while keeping routes accurate.
Than focuses on developing instance-optimal spanners that tailor these simplified networks to a specific layout—like a dense urban area versus a sparsely populated rural one—improving route computation speed and memory use.
He also studies fault-tolerant spanners that are designed to work well when roads or intersections are closed, including color fault-tolerant versions that assign colors to different types of roads—like highways versus bridges—ensuring the network remains reliable even if entire categories are disrupted.
Finally, Than proposes dynamic programming techniques to build shallow-light trees, a lightweight spanner variant that preserves distances from a central “hub” like an airport and enables quick route computation. Similarly, Than aims to discover how tree covers—a collection of smaller, regional “backbone” maps optimized for certain routes—can efficiently approximate network distances and reduce storage needs, which have applications in routing and path-reporting.
Salemi is a fourth-year doctoral candidate and currently works at the Center for Intelligent Information Retrieval (CIIR) under the supervision of Associate Professor Hamed Zamani. He earned his master’s degree in computer science from CICS in 2024 and his bachelor’s degree in computer engineering from the University of Tehran in 2022.
Than is a fifth-year doctoral candidate working under the supervision of Associate Professor Hung Le. He earned his master’s degree in computer science from the University of Nebraska-Lincoln in 2021 and his bachelor’s degree in computer science from the Hanoi University of Science and Technology in 2019.