The Speakers

Invited Speakers:

Speaker: Assoc. Prof Cam-Tu Nguyen

Title:  Making Conversational AI Reliable: Insights from the
Retrieval-Augmented Generation Approach.

Abstract:

Recent conversational AI systems, such as OpenAI’s GPT-4o and Anthropic’s Claude Sonet, have become increasingly powerful, thanks to advancements in Large Language Models (LLMs). However, despite their impressive language capabilities, LLMs often struggle with hallucination—generating inaccurate or misleading information—which raises significant reliability concerns.

In this context, Retrieval-augmented generation (RAG) has emerged as a promising solution to mitigate hallucinations in LLMs. RAG leverages external knowledge sources to ground language model responses and provide citations for cross-checking, improving both accuracy and trustworthiness. However, current RAG systems face challenges due to noise in both the retrieval and generation processes. Specifically, annotation noise hampers the effective training of dense retrieval systems, whereas retrieval noise—the imperfections in the retrieval process—can mislead the language model, distracting it from generating correct responses.

This talk provides an overview of recent advancements in LLMs and RAG, discusses the current challenges associated with RAG systems, and presents several solutions to address these issues.

Bio:

Cam-Tu Nguyen earned her PhD from Tohoku University, Japan, and is currently an Associate Professor at Nanjing University, China. She has approximately 15 years of experience in Artificial Intelligence and its applications, with a strong publication record in leading international journals such as TKDE, TKDD, TALLIP, TWEB, and TON, as well as in major AI and computer science conferences like IJCAI, AAAI, EMNLP, CIKM, ACM MM, USENIX ATC, and InfoCom. Her research over the past five years has focused on conversational AI, particularly in tasks such as Retrieval-Augmented Generation (RAG), LLM-based agents, and multi-modal conversational systems. She has received numerous awards and recognitions, including the Nanjing University Outstanding Researcher Award (2015), the IBM Best Paper Award at the ArgMining Workshop (2018), and the Alibaba Outstanding Collaborator Award (2024). She also actively contributes to the AI community as a program committee member for leading conferences, including IJCAI, ACL, EMNLP, and PAKDD.

Speaker: Prof. Truyen Tran

Title:  Robust AI as digital species

Abstract:

The development of robust AI systems increasingly parallels the evolution of digital species, with cooperation and social understanding emerging as critical capabilities. This talk explores how AI agents can be engineered to exhibit sophisticated social behaviors through various mechanisms including theory of mind, future consequence consideration, and social behavior priors. We examine how diverse training approaches—from self-play with behaviorally distinct agents to decentralized learning under partial observability—can foster adaptable AI systems capable of zero-shot coordination with novel partners. Drawing parallels with biological social systems, we demonstrate how intrinsic social motivation and the ability to model other agents' intentions enable effective cooperation, even in challenging scenarios like resource management dilemmas. The talk presents evidence that equipping AI agents with social priors and the capability to consider future consequences leads to more sustainable and cooperative behaviors, mirroring the evolutionary advantages of social cognition in natural species. Our findings suggest that robust AI systems may need to evolve not just as individual agents, but as inherently social entities capable of understanding and adapting to others' behaviors. This perspective offers insights into developing AI systems that can navigate complex social dynamics while maintaining long-term cooperative strategies, much like successful biological species. 

Bio:

Dr. Truyen Tran is a Full Professor at Deakin University, Australia, where he serves as Head of AI, Health and Science, Applied Artificial Intelligence Institute (A2I2). In his role, he leads a world-class team developing robust human-compatible Generalist AI (AI Future). This technology is then leveraged to accelerate science and engineering (AI4Science) and improve health outcomes (AI4Health). AI Future envisions AI agents as a new digital species integrated into our society, equipped with advanced reasoning and planning capabilities while maintaining robust alignment with human values. His AI4Science research program develops AI Scientists spanning STEM fields, while the AI4Health program focuses on clinical prediction, medical image analysis, and Generative AI for healthcare. Dr. Tran has received multiple international awards for his significant research contributions. He obtained his BSc degree from the University of Melbourne in 2001 and a PhD in Computer Science from Curtin University in 2008.