Purpose
Make neural networks easier to understand by letting users build and compare small CNN descendants directly in the browser.
Human benefit
Learners see how architecture choices affect tensor shapes, training behavior, parameters, and feature maps.
Population design
| Role | What it does |
|---|---|
| Layer builder | Creates model architecture variants. |
| Tensor visualizer | Shows shape changes across layers. |
| Training progress viewer | Displays local metrics. |
| Feature-map explorer | Shows learned representations. |
| Architecture comparator | Preserves before/after evidence. |
Fitness vector
Measure useful output, confidence calibration, speed, memory, local privacy, lineage completeness, novelty, reusable value, and human benefit.
Release path
Start in draft, evaluate in a lab, run shadow comparisons, then promote useful specialists with a release packet. Keep the current champion as a rollback target and archive branches that are no longer active.
What to build first
Start with a toy image classifier and a two-variant lineage record that compares parameter count, speed, and validation score.
PROCEDURE build_browser_cnn_learning_lab(workload)
niche <- DEFINE_NICHE(workload)
parents <- SELECT_INITIAL_PARENTS(niche)
candidates <- CREATE_SPECIALIST_DESCENDANTS(parents)
evidence <- MEASURE_FITNESS_VECTOR(candidates)
packet <- BUILD_RELEASE_PACKET(SELECT_USEFUL_SPECIALIST(evidence))
RETURN RELEASE_WITH_EVIDENCE_OR_NO_OP(packet)
END PROCEDUREPositive future expansion
Add 3D layer views, feature-map cards, and exportable evidence packets for classroom use.
Source reports used for this guide
These reports are preserved verbatim in the site archive. The guide above is an editorial synthesis and may narrow, qualify, or reorganize claims from the source material.