A binary classifier that corrects its own decision boundary from feedback, in pure deterministic hardware.
This project asks how far adaptive behavior can go without software: can a purely deterministic gate-level circuit alter its own decision boundary in response to feedback, with no processor or instruction set involved? It evolves in three stages, from strict Boolean pattern matching, through pop-count similarity against a variable threshold, to a feedback-driven system that corrects itself.
Version 1.1 implements two distinct approaches to threshold correction, each with a formal convergence proof for thresholds M in the range 1 to 15: Max-Initialized Decremental Search (MIDS), which starts at the maximum threshold and searches downward, and State-Aware Threshold Update (SATU), which computes the correct threshold directly from the current and desired output.
| Property | MIDS | SATU |
|---|---|---|
| Correction speed | O(N) | O(1) |
| Strategy | Iterative threshold traversal | Direct threshold computation |
| Threshold memory | Required | Not required |
| Synchronization | Multiple requirements | Single synchronizer |
| Hardware complexity | Higher | Lower |