Temporal Transition Sequence Generation for Sequential Decision Models: Solving the LSTM Cold-Start Problem Through Five Complementary Temporal Pattern Generators

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 temporal transition sequence generation methodology that addresses the cold-start problem in training LSTM networks for sequential decision systems. When an LSTM receives static state vectors repeated across its sequence length, the temporal pathway contributes no discriminative signal. Our approach generates 7,500 synthetic 32-step temporal sequences through five complementary generators: boundary crossing (smooth ramp across decision thresholds), steady drift (within-region stability), multi-phase arcs (degradation-recovery trajectories), spike/recovery (sudden perturbation with exponential decay), and production log replay (interpolated reconstruction from real events). The methodology respects production distributional priors while ensuring all eight decision classes receive representation. Temporal coherence is verified with maximum normalized step jumps below 0.15.

Fields of Study & Subjects

Neural Networks and Applications Time Series Analysis and Forecasting Stock Market Forecasting Methods

Publication details

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