Pattern Multiplication: From 124 Production Events to 88,000 Training Patterns via Multi- Strategy Synthesis: A Transferable Methodology for Training Neural Decision Systems from Sparse Real-World Data

Read full text Summarise with VIRAT
Authors: Rafal Chalupka
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

We present a multi-strategy pattern multiplication methodology that transforms 124 sparse production events into 88,000 verified training patterns, achieving a 710x expansion while maintaining 100% alignment with a deterministic reference oracle. Seven complementary generation strategies are combined: parametric sweep, threshold surface sampling, adversarial boundary probing, Latin hypercube sampling, jitter clouds, temporal chains, and cross-area rotation. A kernel re-verification protocol corrects label drift caused by jitter-induced boundary crossing. The resulting dataset covers 36,595 unique state-space regions across 8 decision classes. An LSTM model trained on this dataset achieves F1=1.000 on held-out validation and 10,000/10,000 on independent generalization stress tests, compared to F1=0.559 when trained only on the original 124 events.

Fields of Study & Subjects

Neural Networks and Applications Machine Learning and Data Classification Imbalanced Data Classification Techniques

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
25 March 2026
Language
EN