Date & Time: June 13-14, 2024
Location: 171 Xiangyuan Road, Gongshu District, Hangzhou
Morning of June 13, 10:00-12:00
Moderator: Jieting Luo
10:00-12:00
Report by Christoph Schommer: AI for Education and Training!
Abstract:
It is my conviction that Artificial Intelligence (AI) should serve the common good and thus be used sensibly and responsibly. This is especially true in education, as AI can improve learner engagement and learning outcomes, for example by personalising learning experiences, improving administrative efficiency and providing intelligent tutoring systems. In this way, AI-driven solutions can adapt to the individual needs of pupils, students and staff and provide customised content and feedback. The integration of AI in education and training also supports teachers by assisting with exams, grading and administrative tasks. However, these opportunities also raise questions, for example about fairness, data protection and the human factor. In my opinion, it is important to find a balance between innovation, ethical considerations and inclusivity when analysing with AI.
Lunch at 501 Seafood Restaurant
Afternoon of June 13
13:30-13:45
Ceremony of Zhejiang-Europe Center for Advanced Intelligence and Ethics
13:50-16:00
Research and discussion in Room 301
Morning of June 14, 9:30-12:00
Moderator: Liuwen Yu
9:30-10:00
Report by Davide Liga: LLMs: Stochastic Parrots or Oracles
Abstract:
Large Language Models (LLMs) have revolutionised the field of AI, enabling single neural
architectures to capture and process vast amounts of human knowledge. These powerful models
are increasingly being used as tools for problem-solving, content generation, and knowledge
acquisition. As people begin to rely on LLMs for work and learning, it raises a pivotal
question: Can we consider LLMs as a reliable and comprehensive sources of knowledge?
The potential for LLMs to serve as virtual teachers is immense, but it is accompanied
by significant challenges. While LLMs have demonstrated impressive performance across a wide
range of tasks, they still grapple with limitations such as biases, hallucinations, and lack
of deep understanding. Transforming LLMs into reliable sources of knowledge and trustworthy
virtual instructors will require addressing these limitations head-on.
In this talk, we will discuss about LLMs, exploring the open questions surrounding the
ability of LLMs "to know" and discussing some key challenges that need to be overcome to
realise their full potential as reliable knowledge sources (and, therefore, as effective
tools for teaching or acquiring new knowledge in general). By understanding the intricacies
of how LLMs process and generate information, we can work towards developing techniques to
align them with our goals and create a future where LLMs serve as invaluable partners in our
quest for knowledge and understanding.
10:00-10:30
Caffe Break
10:30-11:00
Report by Xiangwei Lu: Debiased Cognitive Representation Learning for Knowledge Tracing
Abstract:
Knowledge tracing (KT) is a fundamental task in intelligent education aimed at tracking students’ knowledge status and predicting their performance on new questions. The primary challenge in KT is accurately inferring a high-quality representation of students’ knowledge state that effectively captures their understanding of questions. However, existing methods are typically developed under the assumption that students’ behaviors directly reflect their knowledge state, which may not hold true especially in online learning scenarios. Abnormal behaviors exhibited by students, such as guessing and plagiarism, can introduce biases into the data, making it difficult to accurately assess students’ true knowledge state. To address this limitation, we propose a novel DebiAsed Cognitive rEpresentation (DACE) modeling approach. This approach introduces a novel adversarial training strategy based on information bottleneck theory to obtain a debiased knowledge state representation that retains only the most reliable information for accurately predicting students’ performance on new questions. Moreover, we incorporate a sequential contrastive learning module and educational data augmentation strategies to further enhance the robustness and generalizability of the learned knowledge state representation. We conduct extensive experiments on three public KT datasets to demonstrate the superiority of our model over strong baselines, particularly when confronted with biased data.
11:00-11:30
Report by Salima Lamsiyah: Education in the Age of Generative AI: Leveraging Large Language Models and Reinforcement Learning for Educational Question Generation
Abstract:
In this talk, we will delve into the transformative impact of Generative AI, with a particular emphasis on Large Language Models (LLMs), on education. The rapid advancements in AI technology are revolutionizing our approach to teaching and learning, introducing new tools and methodologies that significantly enhance educational outcomes. A key focus will be the application of reinforcement learning to fine-tune LLMs for generating educational questions, a topic central to our recently accepted paper at the AIED 2024 Conference. Additionally, we will explore future research directions at the intersection of education and AI, specifically within the realm of natural language processing.
Lunch at Dongda Fang Restaurant
Afternoon of June 14, 14:00-17:00
Moderator: Davide Liga
14:00-14:40
Visiting Exhibition Hall and Physical Space in China Europe Center
14:40-15:10
Caffe Break
15:10-17:00
Prof. Beishui Liao (Zhejiang University)Prof. Christoph Schommer (University of Luxembourg)