Grant: Alliance Grants

Learning is a fundamental aspect of both human and machine intelligence. For humans, the learning process benefits from the ability to adapt to changing environments while retaining and applying old knowledge. This ability enables us to abstract, generalize, maintain, and apply acquired concepts across diverse contexts. Additionally, human learning requires the capability of explaining why, how, and what, facilitating the justification of reasoning processes, outcomes, and decision-making results. In contrast, achieving similar capabilities in machine learning and artificial intelligence presents significant challenges. Standard machine learning models often operate on closed-world datasets. This restricts their ability to adapt to dynamic environments and to make intelligent decisions on new concepts. They often require training from scratch with updated data, leading to catastrophic forgetting of previously acquired knowledge and a reduction in overall efficiency. Moreover, many machine learning models, especially those based on deep learning, struggle with explainability. Without a clear understanding of why and how results are produced, it becomes difficult for decision-makers to justify and accept the outcomes. This project seeks to address these challenges by initiating long-term collaborations between Canadian and international researchers. The goal is to enhance both the adaptability and explainability of machine learning methods. Specifically, continual learning will be explored to address adaptability, allowing models to update their knowledge base with new information while retaining previously learned knowledge. In addressing explainability, the project will integrate the theory of three-way decisions and its rapidly evolving models, which have demonstrated superior semantic capabilities over standard learning methods. By focusing on these areas, the project aims to make significant strides in advancing machine learning technologies.