Salvato in:
| Autore principale: | |
|---|---|
| Natura: | Recurso digital |
| Lingua: | inglese |
| Pubblicazione: |
Zenodo
2026
|
| Soggetti: | |
| Accesso online: | https://doi.org/10.5281/zenodo.20058190 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
Sommario:
- <p>## 1. The Empirical Case for Negative-Space-Aware Pattern Recognition</p> <p>### 1.1 Why this section comes first</p> <p>This paper specifies a Pattern Recognition Engine (PRE) whose architectural commitments — context-light Coalescence–Entropy Boundary detection, autonomic operation, seamless void-positive substrate handling, family-aware null routing, recognition-as-trajectory — are not aesthetic preferences. They are the engineering response to a measured failure mode in standard interpolation and pattern-recognition methods.</p> <p>The failure mode is *destructive interpolation*: the systematic destruction of negative-space structure by methods that treat absence as something to be smoothed over rather than as load-bearing geometric content. The destruction is measurable, severe, reproducible, and worsens monotonically with the sophistication of the interpolation method used.</p> <p>Section 11 v2 of the empirical testbed (Destructive Interpolation Paper v2) demonstrates this failure mode across five interpolation/recognition methods on a controlled substrate with typed nulls and known structural features. The results establish, empirically, that:</p> <p>(a) Standard interpolation methods (Nearest, Linear, IDW p=4, Gaussian RBF) fail to preserve negative-space integrity at all tested removal levels (10%, 30%, 50%);</p> <p>(b) The most sophisticated standard method (Gaussian RBF) is the worst offender, falling well below the negative-space integrity floor (N ≥ 0.7) and exceeding the warm-water risk ceiling (W ≤ 0.4) at all removal levels;</p> <p>(c) NSE_V2 — the Negative-Space-Aware engine prototype that operationalizes a predecessor subset of the principles formalized in this paper — preserves negative-space integrity at N = 1.0 and warm-water risk at W ≤ 0.10 across all tested removal levels;</p> <p>(d) NSE_V2 is the only method that promotes true features without promoting distractors;</p> <p>(e) NSE_V2 alone preserves structural features (cliffs and ridges) below the warm-water ceiling.</p> <p>These results justify the architectural commitments that follow. They are not validation-after-the-fact. They are the *reason* for the design.</p> <p>### 1.2 The empirical setup</p> <p>The Section 11 v2 experiment evaluates five methods against a controlled substrate containing:</p> <p>- Positive scalar content (the standard signal that conventional methods are designed to handle)<br>- Pit interiors (regions of typed-null content where the architecture's negative-space integrity must be preserved)<br>- Structural features (cliffs and ridges — sharp transitions between distinct regions that must not be smoothed)<br>- True features and distractors (signal content that must be promoted vs. noise that must not)</p> <p>Removal levels of 10%, 30%, and 50% simulate the substrate degradation that any realistic pattern recognition engine must handle. The tuple-aware variant (used in v2) ensures that the test is sensitive to the typed structure of the substrate rather than only to scalar values.</p> <p>Two metrics anchor the evaluation:</p> <p>**N (Negative-Space Integrity):** the fraction of typed-null content that survives the recognition pass with its family membership and ε/Z structure intact. The integrity floor is N ≥ 0.7; methods falling below this floor are destroying void structure in ways the architecture cannot tolerate.</p> <p>**W (Warm-Water Risk):** the degree to which the method smooths over distinctions that should remain sharp. The ceiling is W ≤ 0.4; methods exceeding this ceiling are generating warm-water contamination — the smoothing-as-corrosion failure mode named in PRE v3.1 and operationalized here.</p> <p>▿ The N floor (0.7) and W ceiling (0.4) values are inherited from PRE v3.1's empirical calibration. Phase B testing should validate whether these thresholds remain appropriate on the post-ROSA, post-ADR-003 substrate, or whether tighter bounds are now achievable. (Open: empirical calibration of threshold values.)</p>