This research introduces a novel continuous energy landscape framework utilizing Graph Neural Networks (GNNs) to overcome the information loss and computational challenges associated with traditional binarized brain state representations. This approach directly learns a continuous precision matrix from functional magnetic resonance imaging (fMRI) signals, preserving the full range of signal values during energy landscape computation.

Validation using both synthetic and real-world fMRI datasets demonstrated that the proposed method achieved higher likelihood and more accurate recovery of basin geometry, state occupancy, and transition dynamics compared to conventional binary models. Specifically, the fMRI data showed a 0.27 increase in AUC for predicting working memory and executive function, and a 0.35 improvement in explained variance (R2) for predicting reaction time. These findings highlight the advantages of utilizing the full signal values in capturing neuronal dynamics, offering significant implications for the diagnosis and monitoring of neurological disorders.

👉 Read the full article on arXiv

This research addresses the fundamental challenge of efficient learning for both biological and artificial agents. For effective learning, an agent must regulate its learning speed, balancing the benefits of rapid improvement against the costs of effort, instability, or resource usage. The study introduces a normative framework that formalizes this problem as an optimal control process, maximizing cumulative performance while incurring a cost of learning.

From this objective, a closed-form solution for the optimal learning rate is derived, taking the form of a closed-loop controller dependent only on the agent's current and expected future performance. This solution generalizes across tasks and architectures and numerically reproduces optimized schedules in simulations. Furthermore, it is shown that a simple episodic memory mechanism can approximate the required performance expectations by recalling similar past learning experiences, providing a biologically plausible route. These findings offer a normative and biologically plausible account of learning speed control, linking self-regulated learning, effort allocation, and episodic memory estimation within a unified and tractable mathematical framework.

👉 Read the full article on arXiv