Ricardo Baeza-Yates is a part-time professor at KTH Royal Institute of Technology in Sweden, and also holds part-time positions at Universitat Pompeu Fabra in Barcelona and at the University of Chile in Santiago. He is additionally affiliated with Chalmers University of Technology, the University of Gothenburg, and the University of Waterloo, and is a visiting professor at Northeastern University’s Silicon Valley campus. He was previously Director of Research at the Institute for Experiential AI at Northeastern University (2021-2025), Vice President of Research at Yahoo Labs (2006-2016), and CTO of NTENT (2016-2020). He is co-author of the best-selling textbook Modern Information Retrieval (Addison-Wesley, 1999; 2nd ed., 2011), winner of the ASIST 2012 Book of the Year Award. Ricardo Baeza-Yates is one of the leading global experts in Responsible AI, with significant contributions to web search, data mining, information retrieval, and algorithmic bias. He is an ACM Fellow and IEEE Fellow, and a member of several international academies, including Academia Europaea and the Chilean Academy of Engineering. He earned his Ph.D. in Computer Science from the University of Waterloo in 1989.

Recent advancements in Large Language Models (LLMs) have made them a popular information-seeking tool among end users. However, the statistical training methods for LLMs have raised concerns about their representation of under-represented topics, potentially leading to biases that could influence real-world decisions and opportunities. These biases could have significant economic, social, and cultural impacts as LLMs become more prevalent, whether through direct interactions—such as when users engage with chatbots or automated assistants—or through their integration into assessment, representation, and dissemination of knowledge. One important case is geographical bias, that surfaces in recommendations for relocation, tourism, or new business. It also appears in cultural stereotypes and even representation of maps.

Lina Yao is a Professor and ARC Future Fellow at UNSW (https://research.unsw.edu.au/people/professor-lina-yao). She strives to develop generalizable and explainable data-efficient data mining, machine learning, and deep learning algorithms—as well as designing systems and interfaces—to enable novel ways of human-AI interactions and cooperations, including an improved understanding of challenges such as robustness, trust, explainability, and resilience that improve human-autonomy partnership. Particularly, here research interest includes Few-Shot Learning, Zero-Shot Learning, Deep Reinforcement Learning, Meta-Learning, Neural Process, Self-supervised Learning, Graph Neural Networks and their applications in a broad range of applications in Recommender Systems, Computer Vision, Brain Computer Interface, Intelligent Transportation System, and Internet of Things. Here research is motivated by, and contributes to, various applications in Healthcare Informatics, Cyber Security, Transportation, Defence, Industry 4.0 and FinTech.

LLM-based agents are transforming recommender systems with capabilities like autonomous exploration and self-evolution through memory augmentation, but their reliability remains underexplored. In this talk, I will introduce three perspectives stemming from our recent effort across the agentic recommendation pipeline. First, we frame implicit feedback as a counterfactual estimation problem with missing treatments, addressing positive-unlabeled and missing-not-at-random biases via causal sample stratification. Second, we introduce a black-box attack that exploits memory vulnerabilities in LLM-powered recommender agents, achieving two- to five-fold higher promotion rates while bypassing current defenses. Third, we present an approach that stabilizes offline policy learning using PDE-inspired Jacobian regularization and adaptive timestep sampling for flow-matching policies. Collectively, these efforts chart the vulnerability landscape—spanning feedback bias, memory security, and behavioral stability—and lay the groundwork for trustworthy agent-based recommendation.