Keynote Speakers

Learnware: Small models do big

Abstract

"Learnware = Model + Specification". Let's consider the following questions: First, do we believe that in the future (A) there will be a big model that is able to cope with all possible learning tasks, or (B) it is crucial to have many models to collaborate? Second, are these models to be developed by (A) one developer (or one company), or (B) lots of developers all over the world? Third, are training data used to train these models to be (A) freely shared, or (B) mostly not? If we choose (B) for the answers, it seems that we will encounter a mission impossible: how to identify helpful models from a growing huge pool of trained models developed by developers all over the world, and reuse or even reassemble the models to tackle new user's task, given that we could not touch developers' and users' training data? "Learnware" makes this possible. A key ingredient is the specification which enables a trained model to be adequately identified to reuse according to the requirement of new user who knows nothing about the model, while model developers' training data are preserved. Learnwares are accommodated in a learnware dock system, which enables small models do big, and enables models do things even beyond their original development purposes. This talk will briefly introduce some preliminary research advances in this direction.

Large Language Models: Capabilities and Limitations

Abstract

Large Language Models (LLMs) have made extraordinary advances in recent times. This has led many people to believe that LLMs provide a pathway to attaining Artificial General Intelligence. This talk will give a high level overview of Large Language Models and present my thoughts on their capabilities and limitations. I argue that LLMs are not intelligent and the LLMs alone do not provide a likely pathway to developing truly intelligent machines.

From Sight to Insight: Visual Memory for Smarter Assistants

Abstract

Imagine a personal assistant that, with user permission, persistently remembers moments from daily life-answering questions like "When and where did I see this lady?" or offering personalized suggestions like "You might enjoy The Little Prince-it relates to the statue you liked in Lyon." Realizing this vision requires overcoming major challenges: capturing visual memories under hardware constraints (e.g., memory, battery, thermal limits, bandwidth), extracting meaningful personalization signals from noisy, task-agnostic visual histories, and supporting real-time question answering and recommendations under tight latency requirements. In this talk, we present our early work toward this goal. Pensieve, our memory-based QA system, improves accuracy by 11% over state-of-the-art multimodal RAG baselines. VisualLens infers user interests from casual photos, outperforming leading recommendation systems by 5-10%. We also share initial results on efficient, event-triggered memory capture and compression. Our work points to a broad landscape of research opportunities in building richer, more context-aware personal assistants capable of learning from and reasoning over users' visual experiences.

TBC

Abstract

TBC

Agentic AI Tutors for Personalized Education at Scale

Abstract

Recent advances in large language and multimodal models have enabled agentic AI tutors to operate autonomously in real-world education systems. This keynote presents how adaptive and personalized educational AI agents are designed and deployed at scale using state-of-the-art techniques, including multi-agent LLM reasoning, multimodal error analysis, dialog-driven interaction, and longitudinal learner modeling. Drawing on large-scale applications from Squirrel Ai Learning, the talk demonstrates how these AI tutors perform fine-grained misconception diagnosis, personalized feedback generation, and adaptive learning path optimization through continuous interaction with real student data. Validated through large-scale real-world deployments, this work highlights key technical lessons on reliability, robustness, and trust in smart education ecosystems.