Truncated backpropagation through time induces a two-sided gradient turnpike property
Abstract
Truncated backpropagation through time reduces the computational burden of training recurrent models by restricting gradient computations to shorter temporal contexts. This induces two distinct boundary effects: missing past information perturbs the forward state evolution, while truncating the backward recursion omits future information. This motivates decoupling the interval over which gradient contributions are averaged for the parameter update from the surrounding temporal context over which they are computed. Under exponential decay of the state-transition Jacobians and mild regularity conditions, this leads to a two-sided gradient turnpike property: the error between truncated- and full-context gradient contributions decays exponentially with the distance from both context boundaries. We derive bounds for the averaged update gradient, yielding guarantees on gradient alignment and approximate full-context stationarity; under unbiased interval sampling and a Polyak-Lojasiewicz condition, these further imply a bound on the resulting optimality gap. Numerical experiments on the Wiener-Hammerstein benchmark illustrate the role of past and future context and the trade-off between predictive performance and computational cost.
Related
- Truncate or Not? A Scaling Study of Looped Transformers with Truncated Backpropagation
- Optimal nodal control and the exponential turnpike property for the flow in gas networks
- Turnpike in optimal control and beyond: a survey
- Exponential Turnpike in Optimal Boundary Control of Multi-Dimensional First-Order Hyperbolic Systems
- Turnpike Properties for Exploratory Stochastic Linear-Quadratic Optimal Control with Indefinite Cost Functionals
BibTeX
@misc{schiller2026truncated,
title = {{Truncated backpropagation through time induces a two-sided gradient turnpike property}},
author = {Julian D. Schiller and Matthias A. Müller},
howpublished = {arXiv},
year = {2026},
month = nov,
doi = {10.48550/arxiv.2611.07415},
eprint = {2611.07415},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2611.07415},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Truncated backpropagation through time induces a two-sided gradient turnpike property
AU - Julian D. Schiller
AU - Matthias A. Müller
PY - 2026
DA - 2026/11/05
PB - arXiv
DO - 10.48550/arxiv.2611.07415
UR - https://arxiv.org/abs/2611.07415
AB - Truncated backpropagation through time reduces the computational burden of training recurrent models by restricting gradient computations to shorter temporal contexts. This induces two distinct boundary effects: missing past information perturbs the forward state evolution, while truncating the backward recursion omits future information. This motivates decoupling the interval over which gradient contributions are averaged for the parameter update from the surrounding temporal context over which they are computed. Under exponential decay of the state-transition Jacobians and mild regularity conditions, this leads to a two-sided gradient turnpike property: the error between truncated- and full-context gradient contributions decays exponentially with the distance from both context boundaries. We derive bounds for the averaged update gradient, yielding guarantees on gradient alignment and approximate full-context stationarity; under unbiased interval sampling and a Polyak-Lojasiewicz condition, these further imply a bound on the resulting optimality gap. Numerical experiments on the Wiener-Hammerstein benchmark illustrate the role of past and future context and the trade-off between predictive performance and computational cost.
ER -