[R] On the Structural Limitations of Weight-Based Neural Adaptation and the Role of Reversible Behavioral Learning

Hi everyone, I recently uploaded a working paper on the arXi and would love some feedback. The working paper examines a potential structural limitation in the ability of modern neural networks to learn. Most networks update in response to new experiences through changes in weights, which means that learned behaviors are tightly bound with the network's parameter space.