Brain-inspired learning and adaptation in large models

Recent advances in large-scale AI models, including large language and vision-language-action models, have significantly expanded the capabilities of artificial systems in perception, reasoning, and decision-making. However, these models remain limited in their ability adapt continuously, learn from limited data, and operate reliably in dynamic, real-world environments. In contrast, biological systems exhibit remarkable adaptability through mechanisms such as neural plasticity, efficient coding, and modular organization. This Research Topic focuses on the intersection of neuroscience principles and large-scale AI, exploring how brain-inspired mechanisms can address these limitations and support the development of more flexible and resilient intelligent systems.This Research Topic aims to advance understanding of how brain-inspired principles can drive learning and adaptation in large-scale models, and how these models can, in turn, be leveraged to study adaptive intelligence. A central challenge lies in moving beyond static training paradigms toward systems capable of continual learning and flexible generalization, especially in open-world settings. We seek contributions that bridge this gap, focusing on novel architectures, training algorithms, and evaluation frameworks inspired by biological learning. The goal is to bring together research that enhances the capabilities of large models by improving their adaptability and efficiency, particularly in embodied settings like robotics, while also using these models as platforms to generate and test hypotheses about brain function and behavior, creating a synergistic loop between AI and neuroscience.We invite original research, reviews, and perspectives, alongside additional relevant article types, that address brain-inspired learning and adaptation in or for large models. Areas of interest include, but are not limited to:- Brain-inspired mechanisms (e.g., neural plasticity, efficient coding, modular organization, predictive coding) for improving large model architectures, training paradigms, and continual/online learning algorithms.- Approaches to making large models more adaptive, efficient, and robust in dynamic, non-stationary, or resource-constrained environments.- Large models as computational platforms to simulate, model, and test hypotheses about brain function, neural dynamics, and adaptive behavior.- Large models as the "cognitive core" for embodied agents, including vision-language-action models in robotics and autonomous systems, enabling real-world interaction and learning.- Theoretical and empirical studies on the convergence and divergence between learning in biological systems and in artificial large-scale models. Topic editors: Xianmin Wang, Xiaowei ZHANG, Jianfei Sun Source: https://www.frontiersin.org/research-topics/80426/brain-inspired-learning-and-adaptation-in-large-models