Persistence, Exposure, and Credit Routing in Linear Temporal Predictive Coding
Abstract
Temporal Predictive Coding (tPC) networks offer a layer-local alternative to backpropagation. For linear tPC, we show, by applying tools from cellular sheaf theory, that temporal credit assignment is a tug-of-war between spatial displacement and temporal errors, whose ratios are modulated by persistence and exposure operators P and Q. Spatial inference therefore gives us the effective recurrent transition F=PA, whilst future credit is filtered by P∗. This creates a persistence--error tradeoff in spatial modes: under non-amplifying invariant recurrence, modes that preserve influence over long delays simultaneously fail to expose that influence strongly as temporal error. Recurrence can circumvent this via routing between persistent and exposed modes.
Related
- DTD-VAE: Disentangled Temporal Dependencies VAE for Credit Risk Prediction
- Explainable machine learning framework with Granger-based temporal predictive validation for monitoring fiscal risk in national health insurance expenditure
- How Much Graph Is Too Much? Graph Construction and Exposure in Spatio-Temporal Forecasting
- Credit Spreads' Term Structure: Stochastic Modeling with CIR++ Intensity
- A Multimetric Evaluation of Ensemble and Neural Models for Imbalanced Credit Scoring on the German Credit Dataset
BibTeX
@misc{tanase2026persistence,
title = {{Persistence, Exposure, and Credit Routing in Linear Temporal Predictive Coding}},
author = {Akira Tanase and Julian Gould and Jeffrey Seely},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.36412},
eprint = {2610.36412},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.36412},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Persistence, Exposure, and Credit Routing in Linear Temporal Predictive Coding
AU - Akira Tanase
AU - Julian Gould
AU - Jeffrey Seely
PY - 2026
DA - 2026/10/26
PB - arXiv
DO - 10.48550/arxiv.2610.36412
UR - https://arxiv.org/abs/2610.36412
AB - Temporal Predictive Coding (tPC) networks offer a layer-local alternative to backpropagation. For linear tPC, we show, by applying tools from cellular sheaf theory, that temporal credit assignment is a tug-of-war between spatial displacement and temporal errors, whose ratios are modulated by persistence and exposure operators P and Q. Spatial inference therefore gives us the effective recurrent transition F=PA, whilst future credit is filtered by P^∗. This creates a persistence--error tradeoff in spatial modes: under non-amplifying invariant recurrence, modes that preserve influence over long delays simultaneously fail to expose that influence strongly as temporal error. Recurrence can circumvent this via routing between persistent and exposed modes.
ER -