Local Certificate Repair for Asynchronous Controller Updates in Recommender Systems
Abstract
We study how to adapt an existing distributed recommendation controller when local dynamics or objective weights change. Starting from a quadratic stability and cost certificate, we construct positive semidefinite corrections with finite spatial support and an explicit residual. A joint admission test allocates the available dissipation margin to these residuals and to interactions among changes and corrections, including pending updates. The test covers complete declared parameter ranges, allowing different operators to progress independently. For a fixed compatible batch, the corrections define a common storage that guarantees exponential stability and bounds the cost accumulated throughout deployment. When the admitted collection evolves, a reservation protocol preserves the certificate and a rate bound on storage increases ensures exponential convergence. Regional linear-quadratic regulator (LQR) design proposes new gain rows for the operators' own users, and a separate residual-based calculation can certify that the final gain lowers the expected cost. A recommendation-network example illustrates the tradeoff between repair depth, data access, and the benefits of deploying an update.
Related
- Efficient Clustering with Quality Guardrails for LLM-based Recommender Systems at Industry Scale
- Flatness-Based Controller Design for Backward-Flat Systems
- The distance-based formation controller design for multi-agent systems in port-Hamiltonian form
- Safe Controller Synthesis for Nonlinear Switched Systems via Probabilistic Abstractions
- Performance Evaluation of Metaheuristic Algorithms for Fractional-Order PID Controller Tuning in Non-Linear Systems
BibTeX
@misc{mariano2026local,
title = {{Local Certificate Repair for Asynchronous Controller Updates in Recommender Systems}},
author = {Simone Mariano and Paolo Frasca},
howpublished = {arXiv},
year = {2026},
month = nov,
doi = {10.48550/arxiv.2611.11462},
eprint = {2611.11462},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2611.11462},
}
RIS
TY - UNPB
TI - Local Certificate Repair for Asynchronous Controller Updates in Recommender Systems
AU - Simone Mariano
AU - Paolo Frasca
PY - 2026
DA - 2026/11/08
PB - arXiv
DO - 10.48550/arxiv.2611.11462
UR - https://arxiv.org/abs/2611.11462
AB - We study how to adapt an existing distributed recommendation controller when local dynamics or objective weights change. Starting from a quadratic stability and cost certificate, we construct positive semidefinite corrections with finite spatial support and an explicit residual. A joint admission test allocates the available dissipation margin to these residuals and to interactions among changes and corrections, including pending updates. The test covers complete declared parameter ranges, allowing different operators to progress independently. For a fixed compatible batch, the corrections define a common storage that guarantees exponential stability and bounds the cost accumulated throughout deployment. When the admitted collection evolves, a reservation protocol preserves the certificate and a rate bound on storage increases ensures exponential convergence. Regional linear-quadratic regulator (LQR) design proposes new gain rows for the operators' own users, and a separate residual-based calculation can certify that the final gain lowers the expected cost. A recommendation-network example illustrates the tradeoff between repair depth, data access, and the benefits of deploying an update.
ER -