Grant: Discovery Grants Program - Individual

In the field of computer vision, remarkable progress has been made in recognizing and interpreting visual data. However, the current state of computer vision faces significant limitations that pose challenges when applied to real-world scenarios. These limitations include rigid models designed for specific tasks, making them unsuitable for dynamic and evolving environments. The assumptions of a closed-world, passive learning, and a lack of adaptability hinder the effectiveness of vision systems in complex, real-world applications. Motivated by these challenges, the long-term objective of my research program is to develop algorithms to build computer vision systems that not only see the world, but also possess the adaptability to learn and evolve continually over a more extended time horizon. This long-term vision underscores the importance of lifelong visual learning as a pivotal research direction. In order to achieve this long-term objective, I propose the following short-term objectives during the next cycle of the Discovery Grant. Each of these objectives represents a step towards achieving the long-term vision of lifelong visual learning: (1) continual model adaptation with evolving domains; (2) never-ending knowledge expansion in visual learning; (3) integrating active exploration with visual learning. These short-term objectives represent critical steps toward realizing our long-term vision of empowering computer vision systems with the adaptability and intelligence needed to excel in ever-changing real-world applications over an extended time horizon with the adaptability and intelligence needed to excel in ever-changing real-world applications over an extended time horizon. This project will involve at least 3 PhD students, 3 MSc students and several undergraduates. These HQP will receive research training in areas such as computer vision, machine learning (especially deep learning), artificial intelligence. These skills are in high demand in Canada and worldwide. Students trained from this research program will easily find employment in academia or industry. The proposed research program is expected to have profound impacts in terms of both scientific advancement and real-world innovation. By tackling the fundamental challenges of lifelong visual learning, we will contribute methodologies and insights that deepen our scientific understanding of how machines can continuously adapt and learn within complex visual environments. The proposed research will involve collaboration between diverse fields, such as machine learning, computer vision, robotics, and cognitive science. This cross-pollination of ideas will not only enrich the research program but also foster a holistic understanding of complex phenomena, influencing scientific research beyond the realm of computer vision. The proposed research will also have far-reaching impacts in practical applications, such as robotics, autonomous systems, AR/VR, etc.