Title: Solving Primary School Math Problem based on understanding  

Abstract: There has been a lot of research carried out on solving primary school math problems (PSMP). Most of this research is based on statistical, neural network, or LLM approaches, which could be less reliable. In this study, we investigate how to provide human-comprehensible knowledge to computers to solve PSMP. Such knowledge could also be used to guide students toward a deeper understanding of how to solve PSMP themselves.

For human beings, learning and memorization must go hand in hand. Therefore, the ability to solve a math problem depends on how effectively memory can be utilized. We propose the following problem-solving method for students, which is very similar to the way we train AI. To solve PSMP, a student should be able to translate a natural language sentence into a math formula (if there is any). Hence, it is important to distinguish (or classify) sentences that describe a scenario, which provides context but is not closely related to the calculation; the quantity of an object; the transfer of an object; the quantitative relationship between two objects, etc. Learning sentence types involves memorizing sentence patterns. Computers can remember abstract patterns, such as “Mary (person) went to school (place) this morning (time).” However, it is easier for humans to remember the actual meaningful sentences, because substitution (or association) seems to be an innate ability, and deriving abstract patterns from these sentences is transparent for humans.

For each sentence type, there is either a corresponding formula or none. In addition, a sentence could entail other meaningful sentences. For example, the sentence “Father bought two apples” could entail “Father has two more apples” and “Father spent money to buy two apples.” These entailed sentences could further produce additional formulas. This is one way common sense knowledge is encoded into various sentence types, since such knowledge usually does not appear literally.

When all sentence formulas are translated, we start constructing a formula network by connecting formula pairs that share at least one variable. Then, we gradually plug the numbers into the formulas through the network to solve for the unknowns. Such a strategy is performed by the computer. Later on, depending on the different ways to solve the problem, the computer will create solution trees to indicate the solution flow for students. For each solution path, a corresponding explanation can be generated. A lot of BERT and LLM tools are used in our sentence type classification and explanation generation.

Finally, instead of presenting the explanation directly to the students, our system will create a dialogue to assess whether the student truly understands how to solve the problem. For example, the system might test students on sentence types by letting them choose which example sentence is most similar or ask students what the most appropriate first step is to solve the problem. The purpose is to encourage students to think throughout the problem-solving process.

In summary, we aim to provide enough knowledge to train our AI system with the above capability. Hopefully, such a design will fulfill the goal of AI-assisted PSMP solvers.

 

BIO: Dr. Wen-Lian Hsu is a Chair Professor in the Department of Computer Science and Information Engineering at Asia University, Taiwan. He earned his Ph.D. in Operations Research from Cornell University and served as a tenured Associate Professor at Northwestern University before joining the Institute of Information Science at Academia Sinica, where he served as Director from 2012 to 2018.

Dr. Hsu’s early contributions focused on graph algorithms. In 1993, he developed a software system called GOING, which revolutionized Chinese input methods on computers in Taiwan. Building on similar semantic analysis techniques, he later advanced research in question answering and chatbot systems. Dr. Hsu is particularly renowned for applying natural language processing (NLP) techniques to Chinese language processing and biological literature mining. Under his leadership, his research team has achieved first-place rankings in numerous international
competitions, including protein named entity recognition (NER), biological relation extraction, recognition of inference in Chinese text, Chinese question answering, NER, and word segmentation.

Dr. Hsu’s recent research focuses on advancing the deep understanding of natural language. His projects include commonsense reasoning in primary school mathematics, relation extraction for protein-protein interactions, and the annotation of electronic medical records.

Dr. Hsu has received many prestigious awards in recognition of his contributions to academic research. These include the Outstanding Research Awards from the National Science Council (NSC) in 1991, 1994, and 1996, the inaugural K. T. Li Research Breakthrough Award in 1999, the NSC Appointed Distinguished Research Fellow Award in 2005, the IEEE Fellowship in 2006, and the Teco Award in 2008. Additionally, he served as the President of the Artificial Intelligence Society of Taiwan from 2001 to 2002 and as the President of the Computational Linguistics Society of Taiwan from 2011 to 2012.

 

 

Title: Impact of disruptive technologies and the vision for future of education  

Abstract: Recent advances in the technologies, particularly the emerging visualization technologies and generative AI, are changing the landscape of education. Many of these advancements have significant potential to be disruptive, bringing both excitement and concern. They can foster deeply engaged learning experiences by providing the affordances that were not possible before and provide means for continuously monitoring, assessing, and guiding students in their learning process to ensure they receive tailored support. These same advances also have potential for creating biased content, spreading false information, sharing private information by mistake, causing job loss, and raising questions about human dependency and accountability. This talk will look into the vision for future of education and investigate how to exploit these emerging technologies before they exploit us. We shall explore, instead of trying to stop the progress due to the various concerns, how we can make effective plans and rules to reduce those risks and take advantage of the opportunities such technologies bring to transform the current education, with constraints and safeguards in place to make learning process safe and reliable.

 

BIO: Dr. Kinshuk is the Dean of the College of Information and Full Professor of Learning Technologies at the University of North Texas. He also serves on the Board of Directors for iSchools – an international organization of around 130 universities focused on all aspects of research and teaching about information, and Board of Advisors for Dallas AI – the largest non-profit AI forum in North Texas with over 8000 members. Prior to that, he held the NSERC/CNRL/Xerox/McGraw Hill Research Chair for Adaptivity and Personalization in Informatics, funded by the Federal government of Canada, Provincial government of Alberta, and by national and international industries. He was also Full Professor in the School of Computing and Information Systems and Associate Dean of Faculty of Science and Technology, at Athabasca University, Canada. After completing first degree from India, he earned his Masters’ degree from Strathclyde University (Glasgow) and PhD from De Montfort University (Leicester), United Kingdom. His work has been dedicated to advancing research on the innovative paradigms, architectures and implementations of online and distance learning systems for individualized and adaptive learning in increasingly global environments.

 

 

Title: Teaching soft skills via videos: AI-based support for engagement  

Abstract: Soft skills play a crucial role in our lives, and are critical for students in their education and for professionals in their careers. However, soft skills are challenging to teach as they require substantial time, practice and feedback from instructions. Video watching is one of the methods for teaching soft skills, as they allow the student to explore multiple viewpoints and reflect on their experience. Although learning from videos has many advantages, such as providing flexible, self-controlled learning opportunities, and raising students’ motivation, watching videos can be a passive activity and result in shallow learning.

In this talk, I will present the approach we developed for teaching soft skills using the AVW-Space, a video watching platform. AVW-Space allows the teacher to select publicly available videos from YouTube and define a space for their students. Learning happens in two phases in the platform. In the first phase, students watch videos and write comments on them. The teacher can specify aspects for students to use when writing comments, which focus students’ attention to important concepts in videos or to encourage students to self-reflect. In the second phase, the teacher can select some comments to open anonymously to the whole class, to review and rate. The teacher can specify rating categories to reinforce important activities, such as self-reflection. In addition to writing/rating comments, AVW-Space uses AI-based support in order to track the learner’s behaviour and provide personalized nudges in order to improve engagement.

I will present the evolution of AVW-Space and various types of AI-based support we have added to it over the years. In early studies, when there was no support, half of the participants watched videos passively. To improve engagement via comment writing, extended the platform by adding a set of reminder nudges, which are given to students who are passively watching videos or not commenting on a variety of topics. Those nudges resulted in a significantly higher percentage of students writing comments. We then developed machine learning models which classify students’ comments immediately after they are written into low, medium or high-quality comments. Based on these classification, AVW-Space provides additional nudges to students based on the quality of comments they write. We also added visualizations of students’ activities, the comment quality and nudges, so that students can easily review their progress. The most recent studies show the effectiveness of theAI-based support: the vast majority of students are now active and writing high-quality comments, which result in increased learning.

 

BIO: Dr Antonija (Tanja) Mitrovic is a full professor and the Head of the Department of Computer Science and Software Engineering at the University of Canterbury, Christchurch, New Zealand. She is the leader of ICTG (Intelligent Computer Tutoring Group). Dr Mitrovic received her PhD in Computer Science from the University of Nis, Yugoslavia, in 1994. She is the past President of the Asia-Pacific Society for Computers in Education, and Past President of the International Society for Artificial Intelligence in Education. She is an associate editor of the following journals: International Journal on Artificial Intelligence in Education and Research and Practice in Technology Enhanced Learning (RPTEL). Tanja is a Distinguished member of ACM, Senio member of IEEE and AAAI. and a Fellow of the Asia-Pacific Society for Computers in education (ASPCE). She was awarded the Distinguished Researcher Award in 2011 by the Asia-Pacific Society for Computers in Education.

Dr Mitrovic’s primary research interests are AI in education. ICTG has developed a number of constraint-based intelligent tutoring systems in a variety of domains, which have been thoroughly evaluated in real classrooms, and proven to be highly effective. These systems provide adaptive support for acquiring both problem-solving skills and meta-cognitive skills (such as self-explanation and self-assessment). Although most of the ITSs developed by ICTG support students learning individually in areas such as database querying (SQL-Tutor), database design (EER-Tutor and ERM-Tutor), data normalization (NORMIT), there are also constraint-based tutors for object-oriented software design and collaborative skills, various engineering topics (thermodynamics, mechanics), training to interpret medical images and language learning. ICTG has also developed ASPIRE, a full authoring and deployment environment for constraint-based tutors. Recent research includes using personalized nudges to improve engagement and learning outcomes in Video-Based Learning (VBL). ICTG developed AVW-Space, an online portal for VBL that provides personalized nudges to students.