Wenbin Li, Yingxia Shao, Yawen Li, Hongguang Zhang, Rui Yan
To deal with the dynamic changes of the real world, incorporating new knowledge into existing models to adapt to new problems is a fundamental challenge of computer vision. Humans and animals need to incrementally absorb new experiences in order to survive in new environments, and evolution endows the entire lifetime system with stronger adaptability.With the advancement of foundation models, there is a growing interest in enhancing their capabilities through continual learning. This technology allows models to continually update their generalization abilities based on real-time data, thereby enhancing their robustness and functionality. Consequently, continual learning has emerged as a pivotal paradigm in machine learning, leveraging the continuous refinement of foundation models through fine-tuning.
Schedule (Dec 01, 2023): https://dai-cl.github.io/
Ying Wen, Yaodong Yang, Weinan Zhang, Muning Wen
The burgeoning field of Large Language Models (LLMs) and Multi-modal Models, underscored by their prowess in assimilating vast swathes of vision and language data, have carved a niche in executing a myriad of downstream tasks across domains like dialogue systems, autonomous navigation, healthcare, and robotics. However, as these models transcend into real-world applications, they grapple with novel challenges encompassing external feedback assimilation, task modality adaptation, long-term reasoning and planning, and action grounding — realms traditionally navigated by sequential decision-making paradigms like reinforcement learning, planning, and optimal control. Unlike the narrowly focused task-specific models of yore, the transcendent nature of LLMs, equipped with a broad spectrum of prior knowledge, promises a fertile ground for augmenting sample efficiency and generalization.
Schedule (Nov 30, 2023): https://sites.google.com/view/dai23-llmdm/home
Junge Zhang, Zhaofeng He, Mingyi Zhang, Liuyu Xiang, Xiang Cheng
Reinforcement learning (RL) has been a driving force in the development of artificial intelligence, enabling machines to learn from experience and make optimal decisions in various environments. However, traditional reinforcement learning techniques often require a vast number of interactions with the environment, making them inefficient and time-consuming for real-world applications. Sample-efficient reinforcement learning aims to address this challenge by reducing the number of interactions required to achieve successful outcomes. This cutting-edge approach has the potential to revolutionize the field of reinforcement learning, making it a game-changer for applications such as robotics, autonomous vehicles, and smart grids.
Schedule (Nov 30, 2023): https://sites.google.com/view/dai23-sample-efficient-rl/home
Qingshan Li, Yishuai Lin, Lu Wang
Applied Artificial Intelligence is a topic of interest from a practical standpoint. The goal of this workshop is to bring together researchers and practitioners who are interested in utilizing AI theories and methods in AI practices (general AI methods, multi-agent systems, distributed learning, computational game theory, etc.) or intelligent systems. It is important to note that this workshop is focused on the area of applied artificial intelligence and intelligent systems, which refer to how AI can be applied to real-world scenarios. This topic is therefore of great importance and has received a great deal of attention
Schedule (Dec 01, 2023): https://xidiancs.github.io/dai/workshops.html
Fangwei Zhong, Siyuan Qi
This workshop explores the dynamic and challenging realm of Multi-agent Systems (MAS) operating within complex environments characterized by dynamism, uncertainty, heterogeneity, adversarial nature, large scale, interdependencies, and high dimensionality. Participants are encouraged to present case studies and real-world applications, shedding light on how MAS can revolutionize domains such as embodied agents, autonomous vehicles, robotics, and beyond.
Schedule (Dec 01, 2023): https://sites.google.com/view/dai-2023-masce/home
J. Senthilnath, Nagarajan Raghavan
Traditionally, lab-to-market in the case of Drug or Materials development typically occurs over 15 to 25 years from research to development. Also, the exploration of search space for synthesis and characterization requires millions of samples. Hence, there is a need to accelerate scientific discovery from lab-to-market. Recently, Generative Artificial Intelligence (Gen-AI) approaches have progressed as valuable AI tools to accelerate the process by overcoming the traditional ways to synthesize and characterize materials in a laborious fashion.
Schedule (Dec 01, 2023): https://autonomous-scientific-discovery.github.io/gai/