IINAE 2025: The 7^th Workshop on Innovation Initiatives in NLP/AI/Education

Reasoning ability is crucial for AI systems to handle complex, multi-step tasks in a robust and interpretable way. I will first review prompting-based strategies such as Chain-of-Thought, Tree-of-Thoughts, and Graph-of-Thoughts for inducing explicit reasoning structure in large language models. I then present the Hierarchical Reasoning Model (HRM), which employs multi-timescale computation to coordinate high-level planning and low-level execution within a unified architecture. Finally, I discuss Tiny Recursive Models (TRMs), which achieve compositional reasoning through repeated application of a small shared module, yielding strong performance with minimal parameters. Together, these methods point toward structured, modular approaches as a promising direction for advancing reasoning in AI.

Ye Kyaw Thu is a Visiting Professor at the National Electronics and Computer Technology Center (NECTEC), Thailand, where he has been since January 2019, and is the Founder of the Language Understanding Lab in Myanmar. He holds a Doctor of Science (2011) and a Master of Science (2006) from Waseda University, Japan, and a Bachelor of Science in physics from Dagon University, Myanmar (2000). He also holds diplomas in Computer Studies (UK). His research focuses on artificial intelligence (AI), natural language processing (NLP), and human-computer interaction (HCI). He supervises students at various institutions, including Assumption University (AU), Kasetsart University (KU), King Mongkut's Institute of Technology Ladkrabang (KMITL), Sirindhorn International Institute of Technology (SIIT), Thammasat University (TU), and the Japan Advanced Institute of Science and Technology (JAIST).

Homepage: https://sites.google.com/site/yekyawthunlp
GitHub: https://github.com/ye-kyaw-thu

In the age of generative AI and large language models, we have more data than ever, yet transforming this information into real knowledge remains a challenge. This talk explores the next frontier of AI: building knowledge-aware systems that combine the learning power of neural networks with the structured reasoning of Knowledge Graphs (KGs). By organizing information from diverse sources, enabling reasoning across interconnected data, and designing models that are interpretable, reliable, and transferable, Knowledge Graphs help us move from raw data to meaningful understanding. Through examples in natural language processing, financial intelligence, and educational technology, we will see how knowledge-centric design enhances interpretability, reliability, and cross-domain reasoning. The talk outlines a path from data to intelligent knowledge, showing how KGs can help us build AI that truly understands.

Natthawut Kertkeidkachorn is currently an Associate Professor in the Graduate School of Advanced Science and Technology at the Japan Advanced Institute of Science and Technology (JAIST). He received his B.Eng. and M.Eng. degrees in Computer Engineering from Chulalongkorn University, Bangkok, Thailand, in 2011 and 2013, respectively, and his Ph.D. in Informatics from the Graduate University for Advanced Studies (SOKENDAI), Japan, in 2017. From 2017 to 2021, he was a Researcher at the Artificial Intelligence Research Center, National Institute of Advanced Industrial Science and Technology (AIST), Japan, before moving to JAIST in 2021. His research interests include knowledge graphs, machine learning, and natural language processing.