Progressive Neural Networks 2D: Zero-Forgetting Column Growth

This is the 2D schematic companion to the 3D progressive-network simulator. The same structural-isolation mechanism drives both: every new task grows a brand-new, permanently frozen column of layers, and lateral adapter connections let the new column read — never write — every earlier column's features. Here the growing column chain is drawn as a pannable node-and-edge diagram you drag and zoom directly, paired with a live line chart tracking retained accuracy over training steps for the progressive network against a naive shared-weight baseline whose old-task accuracy erodes with every new task. Two extra sliders — transfer strength and forgetting rate — expose constants the original hides, so you can push positive transfer and catastrophic forgetting to their extremes and watch the isolation-versus-interference trade-off play out in real time.