11/7/2024 Rudy San Miguel
Written by Rudy San Miguel
More than one hundred faculty, students and Amazon partners attended the second annual Fall Research Symposium on September 26, 2024, held by The Amazon-Illinois Center on AI for Interactive Conversational Experiences (AICE) at the University of Illinois Urbana-Champaign Grainger School of Engineering Siebel School of Computing and Data Science.
According to AICE, “The symposium showcased AI conversational systems work by AICE researchers and students. The event brought together thought leaders from academia and industry to foster collaborations and build community for research, education and infrastructure to accelerate AI for interactive conversational experiences.”
AICE Director and Professor at the Siebel School of Computing and Data Science Heng Ji opened the symposium by highlighting AICE’s first year successes: twenty-plus publications including several co-authored with Amazon scientists, a best abstract award, Best Paper nomination at ICRA 2024, and numerous Illinois Engineers completed successful internships at Amazon in the summer of 2024.
“AICE is thriving. We’ve expanded from conversational AI to everything AI, including Large Language Models, Vision-Language, Search and Retrieval, Efficient Generative AI and Responsible Generative AI.” Ji announced AICE has received 42 project proposals and 18 PhD fellowship proposals, adding that almost every AI faculty member on campus submitted proposals.
Ruhi Sarikaya, AICE Liaison and Amazon Alexa’s vice president for applied science, spoke at the event, providing an overview of AI at Amazon and opportunities for future research. “I am delighted to see AICE expand as we embark on our second year working together,” Sarikaya said. “The new research projects are ambitious, leveraging the unique viewpoints and expertise of Illinois to advance generative AI innovation. A hallmark of the AICE collaboration is the interaction between Illinois and Amazon scientists, and this fall’s symposium provided a perfect venue for an exchange of ideas.”
Dilek Hakkani-Tür, CS professor, and Manling Li, assistant professor in Computer Science at Northwestern University, provided keynote speeches.
Hakkani-Tür posited utilizing accountability models in LLM-based dialogue systems for task completion. By introducing friction turns, in cases of model uncertainty and errors, this results in higher accuracy in task completion and alleviates the over-reliance in LLM-based AI through Accountability Models.
Li Spoke on planning and reasoning with physical world knowledge to extend Large Language Models to Large Agent Models. She suggested the fundamental difference between the two models “lies in their decision-making capabilities. This fundamental difference necessitates a shift in how we approach the development of large agent models.”
Two panels were also held during the packed day. The first focused on AI for Science and was moderated by AICE Business Development Principal Rajiv Dhawan. The panel included AICE Director Ji; AICE Advisory Board Member Pradeep Natarajan; Department of Chemistry at the University of Illinois professor Martin Burke; Nuclear, Plasma, and Radiological Engineering at the University of Illinois professor Tomasz Kozlowski; and CS professors Jiawei Han, Hao Peng, and Tong Zhang.
The second panel centered on Amazon science and graduate student opportunities and featured a panel of Grainger engineers and former science interns at Amazon. The panel was moderated by Amazon Academic Partnerships Principal Kathleen Allen and included Chi Han, Jeonghwan Kim, Raunak Sengupta, and Evelyn Ma.
Grainger engineers also provided a snapshot of their 2024-25 AICE research projects:
“Building AI Engineers via Integrating LLMs and Domain Knowledge,” Bin Hu, professor, Electrical and Computer Engineering, Illinois.
“Multi-Agent LLM-Based Reasoning, Computation and Data Discovery for Decision Services,” Vikram Adve, professor, Siebel School of Computing and Data Science.
“Mastering Extensive Environmental Interactions: Building Agentic Foundation Models with Reinforcement Learning,” Yuxiong Wang, professor, Siebel School of Computing and Data Science.
“Long-form Video Understanding via Active Token Compression,” Yuxiong Wang, professor, Siebel School of Computing and Data Science.
“Physiology of Large Language Models,” Chi Han, AICE PhD Fellow, Siebel School of Computing and Data Science.
“Advancing Long Context Reasoning in Open LLMs,” Hao Peng, professor, Siebel School of Computing and Data Science.
“Synergizing LLM and World Model for Zero-Shot 3D Human-Object Interaction Generation,” Liangyan Gui, professor, Siebel School of Computing and Data Science.
At the close of the symposium, a networking social was held where Grainger engineers presented a poster session of their work.
Grainger Engineering Affiliations
Heng Ji is an Illinois Grainger Engineering professor of computer science affiliated with the electrical and computer engineering department and coordinated science laboratory. Heng Ji is the AICE Director.
Dilek Hakkani-Tür is an Illinois Grainger Engineering professor of computer science.
Tomasz Kozlowski is an Illinois Grainger Engineering professor of nuclear, plasma & radiological engineering.
Jiawei Han is an Illinois Grainger Engineering professor of computer science. Jiawei Han holds the Michael Aiken Chair.
Hao Peng is an Illinois Grainger Engineering professor of computer science.
Tong Zhang is an Illinois Grainger Engineering professor of computer science.
Bin Hu is an Illinois Grainger Engineering professor of electrical and computer engineering and is affiliated with the coordinated science laboratory.
Vikram Adve is an Illinois Grainger Engineering professor of computer science and is affiliated with the coordinated science laboratory and information trust institute. Vikram Adve is the Donald B. Gillies Professor in Computer Science.
Yuxiong Wang is an Illinois Grainger professor of computer science.
Liangyan Gui is an Illinois Grainger Engineering professor of computer science.