Advances & challenges in foundation agents: Section 2.1 – Learning

This article is Section 2.1 of a series of articles featuring Liu and colleagues’ book Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems.

Learning is the mechanism by which an intelligent agent improves its performance over time through experience. It allows the agent to adjust its internal parameters—such as its models of the world, policy for action, or strategy for reasoning—based on observations and outcomes.

From supervised learning to reinforcement learning, from variational inference to meta-learning, diverse paradigms have been proposed to formalize how an agent can acquire knowledge, skills, or behavior through interaction with data or environments. Despite their apparent differences, these learning approaches share a common goal: optimizing an objective that aligns the agent’s internal process with its external goals.

Section 2.1 presents a unified perspective on learning, bridging multiple paradigms under a general formulation, and setting the stage for how learning principles can be integrated into agentic reasoning and decision-making:

2.1.1 – A unified formulation of learning
2.1.2 – Learning across mental state components
2.1.3 – Learning space
2.1.4 – Learning objective.

Next part: Section 2.1.1 – A unified formulation of learning.

Article source: Liu, B., Li, X., Zhang, J., Wang, J., He, T., Hong, S., … & Wu, C. (2025). Advances and challenges in foundation agents: From brain-inspired intelligence to evolutionary, collaborative, and safe systems. arXiv preprint arXiv:2504.01990. CC BY-NC-SA 4.0.

Header image: AI is Everywhere by Ariyana Ahmad & The Bigger Picture / Better Images of AI, CC BY 4.0.