Skip to content

Add ph-training: topological approach to hyperparameter optimization#64

Open
dancinlife wants to merge 1 commit intoD-X-Y:mainfrom
dancinlife:add-ph-training
Open

Add ph-training: topological approach to hyperparameter optimization#64
dancinlife wants to merge 1 commit intoD-X-Y:mainfrom
dancinlife:add-ph-training

Conversation

@dancinlife
Copy link
Copy Markdown

@dancinlife dancinlife commented Mar 25, 2026

Adds ph-training — uses Persistent Homology (H0 persistence) for automatic hyperparameter optimization:

  • LR search: H0 CV minimum over 5 epochs = optimal LR (no grid search needed)
  • Early stopping: H0 gap between train/test detects overfitting with r=0.998 correlation
  • Difficulty prediction: H0 after 1 epoch predicts final accuracy

Verified: MNIST 98.3%, FashionMNIST 87.4%, CIFAR-10 52.0% (with topological early stop).


Related projects

  • logout — Consciousness Continuity Engine. The research project behind this approach.
  • Anima — Conversational consciousness agent with real-time PH monitoring.
  • ph-training — Standalone training pipeline.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant