auditing-deep-learning-overfit

What problem does it solve?

This Skill helps you determine whether a deep-learning model is truly overfitting, or whether the symptoms are actually caused by distribution shift, label noise, underfitting, or a simple training plateau.

Core Features & Use Cases

• Trajectory-based diagnosis: Compares training and validation curves across epochs instead of guessing from a single checkpoint.
• Guardrailed triage: Checks for train-validation distribution mismatches, mislabeled validation examples, and weight-norm trends before recommending regularization.
• Actionable remediation: Prioritizes early stopping, augmentation, dropout, weight decay, capacity reduction, and more data in the right order.
• Best for: CNNs, RNNs, LSTMs, Transformers, and dense MLPs when validation performance degrades after initially improving.

Quick Start

Ask the skill to audit your deep-learning training run using the train and validation histories, optional weight norms, and any dataset summaries, then return the diagnosis and the recommended next steps in priority order.