rnn-task-degradation-analysis

What problem does it solve?

This framework analyzes how weight initialization, the diversity of solutions, and degradation influence RNN performance on temporal tasks, enabling systematic study of robustness and dynamical regimes.

Core Features & Use Cases

• Analyze how different initializations lead to distinct dynamical solutions that achieve the same task.
• Evaluate performance degradation when network size decreases, time intervals increase, or connections are damaged.
• Provide a Python-based workflow for training, analysis, and visualization of diversity and degradation metrics.

Quick Start

Train several RNNs with different initializations and observe how solution diversity relates to performance under varying size, damage, and interval conditions.