Aryan Deshwal’s Work Featured at AAAI Conference

Department of Computer Science & Engineering Assistant Professor Aryan Deshwal was recently featured at the 2025 Association for the Advancement of Artificial Intelligence (AAAI) Conference. Deshwal was one of 40 faculty members from around the world chosen for the AAAI New Faculty speaker series at the conference. Additionally, he was a lead organizer of a popular conference workshop, AI to Accelerate Science Discovery and Engineering Design.

“I am very excited about this research direction and it is nice to see the rising interest of the larger AI community around this topic,” said Deshwal. “I am grateful for this opportunity and it was very rewarding to showcase my research to the community.”

Deshwal’s research is on AI driven experimental design and sequential decision making for accelerating scientific discovery and engineering design. His novel adaptive experiment design algorithms help determine which experiments will be most effective while carefully managing limited resources by formally reasoning about the potential information each experiment provides. These problems occur throughout scientific discovery and engineering design (materials design, molecule/protein design, additive manufacturing, AutoML, hardware design) and can ultimately help with a number of challenges facing society.

“In experimental design, the challenge is to find the best design in a large design space while minimizing the resources that are used,” said Deshwal. “For example, in material design, there are many candidate choices for materials, and each one you evaluate costs a large amount of resources (time/money), so how do you find the best material while minimizing the resource?”

In line with his personal research, the AI to Accelerate Science and Engineering workshop focuses on bringing AI opportunities to other scientific disciplines. Deshwal has helped organize the workshop at the past four conferences and has seen its popularity significantly grow with attendees and presenters doubling between 2024 and 2025. Deshwal was a co-organizer with Regents Professor Vipin Kumar, who hosted an NSF workshop over the summer focusing on similar topics.

In addition to the workshop and new faculty speaker series, Deshwal presented a research paper focused on reinforcement learning. In collaboration with Washington State University and Oregon State University, the project was inspired by an agriculture application, and explored how to create safe and reward maximizing decision policies from offline datasets to handle the challenging setting where cost constraints can change after deployment.

“Reinforcement learning is an effective technique used to learn sequential decision making policies under uncertainty but real-world applications often include additional constraints related to cost or safety. This paper approached decision making while considering these constraints by utilizing previous logged datasets of decisions.

“This is an exciting research direction, because there are a lot of real-world applications. These methods can be broadly applicable, so I am actively looking for collaborators in other fields that can benefit from this type of technology, especially scientists and experimentalists.”

Learn more about Deshwal’s work on his personal website.