🪜 2D Discriminative Fine-Tuning: Layer-Wise Learning-Rate Ladder, Slanted Triangular Schedule and Forgetting
Side-by-side decision maps for uniform versus layer-wise learning rates on a shifted target task, with per-layer drift bars, target loss and source-accuracy curves, verified by finite-difference gradients, an exact linear-regression contraction law and STLR boundary values.
⚙ Under the hood
Side-by-side decision maps for uniform versus layer-wise learning rates on a shifted target task, with per-layer drift bars, target loss and source-accuracy curves, verified by finite-difference gradients, an exact linear-regression contraction law and STLR boundary values.
Fine-TuningTransfer LearningLearning Rate ScheduleCatastrophic ForgettingBackpropagationULMFiTNeural NetworkMachine Learning
2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install