Data and trained models for: Structure, disorder, and dynamics in task-trained recurrent neural circuits

Trained RNN weights, SGLD sampling outputs, DMFT results, and precomputed analysis caches for reproducing all figures in Clark*, Bordelon*, Zavatone-Veth*, and Pehlevan (2026). See the GitHub repository at https://github.com/davidclark1/RNN-Learning-Theory for code and instructions.

本数据集包含用于复现Clark*、Bordelon*、Zavatone-Veth*与Pehlevan（2026）一文所有图表的训练后循环神经网络（Recurrent Neural Network, RNN）权重、随机梯度朗之万动力学（Stochastic Gradient Langevin Dynamics, SGLD）采样结果、动态平均场理论（Dynamic Mean-Field Theory, DMFT）计算结果，以及预计算分析缓存。相关代码与操作指南可参见GitHub仓库：https://github.com/davidclark1/RNN-Learning-Theory