History-Dependent Predictive Readout in Adaptive Resistive Networks

History-Dependent Predictive Readout in Adaptive Resistive NetworksAuthor: Boris Chernov Affiliation: Flow Systems Lab DOI: 10.6084/m9.figshare.34983370AbstractThis preprint presents a reproducible computational investigation of history-dependent predictive readout in a synthetic adaptive resistive network. The model consists of six nodes and seven adaptive resistive edges whose states evolve under sequences of applied electrical stimuli.The study examines whether the magnitude of observed signal components corresponds to their usefulness for predicting subsequent network responses, and whether a limited measurement budget can be allocated more effectively using predictive performance rather than signal amplitude.Across several stimulation sequences, multiscale analysis reveals that strongly expressed signal components are not necessarily the most informative for predicting future responses. In particular, fine-scale components may carry substantial additional predictive information despite contributing only a small fraction of the observed signal magnitude.When only two measurements are permitted, selecting measurement phases according to predictive performance on independent validation data can outperform amplitude-based selection. However, this advantage is sequence-dependent: it is substantial for XXXXYYYY and XXXXXXXX, absent for XYXYXYXY, and may be comparable to simple fixed measurement schedules in some conditions.The computational validation includes variations in adaptation strength, initial resistance heterogeneity, and measurement noise. An independent execution reproduced the principal numerical outputs of the supplied research package.Scope and limitationsThe network is synthetic, and its adaptation rule has not been independently calibrated against experimental TiO₂ device kinetics. Future-response contrasts are calculated using counterfactual branches of the simulator. The results therefore establish computational properties of the specified model, not experimentally verified behavior of physical memristive materials.Research significanceThe findings distinguish signal expression from predictive utility and demonstrate that the usefulness of sparse electrical measurements depends on the stimulation sequence, prediction target, and observation strategy.The accompanying research package provides source code, numerical results, validation tables, figures, and reproducibility instructions to support independent examination and potential experimental investigation in physical adaptive resistive networks.Materials includedEnglish-language preprint (PDF and DOCX).Reproducible computational research package (ZIP).Source code, experimental results, and reviewer documentation.Licensing: CC BY 4.0 for the preprint and research materials; MIT for the software source code.Funding: No external funding.Competing interests: None declared.