Circadian Predictive Coding Benchmark Dashboard

Interactive benchmark views with a compact overview plus detailed drill-down charts across backprop, predictive coding, and circadian predictive coding.

Historical snapshot: These CIFAR-100 charts are not corrected-protocol results. The reviewed vision runner used test labels for early stopping and sleep rollback, its baselines were unmatched, and source CSV files for some charts are absent. Read the provenance notes before interpreting rankings.

Source artifacts are generated by scripts/generate_readme_figures.py from benchmark summary CSV files.

Focus

Circadian Predictive Coding

Reference Baselines

Backprop + Predictive Coding

Current Dataset Profile

CIFAR-100 Multi-seed Summary

Hardest-Case Dynamics (Training + Inference)

Interactive replay: slider + play controls, phase-B inference map, and circadian internals graph (node/edge weights, chemical/plasticity state).

Hardest-case training and inference dynamics GIF

GIF fallback preview for README-compatible rendering.

Compact Overview (Interactive)

Detailed Accuracy (Interactive)

Detailed Training Throughput (Interactive)

Detailed Inference Latency P95 (Interactive)