Neural Network Training as an Optimal Control Problem : — An Augmented Lagrangian Approach —

Type: Article

Publication Date: 2021-12-14

Citations: 5

DOI: https://doi.org/10.1109/cdc45484.2021.9682842

Abstract

Training of neural networks amounts to nonconvex optimization problems that are typically solved by using backpropagation and (variants of) stochastic gradient descent. In this work, we propose an alternative approach by viewing the training task as a nonlinear optimal control problem. Under this lens, backpropagation amounts to the sequential approach (single shooting) to optimal control, where the states variables have been eliminated. It is well known that single shooting may lead to ill-conditioning, and for this reason the simultaneous approach (multiple shooting) is typically preferred. Motivated by this hypothesis, an augmented Lagrangian algorithm is developed that only requires an approximate solution to the Lagrangian subproblems up to a user-defined accuracy. By applying this framework to the training of neural networks, it is shown that the inner Lagrangian subproblems are amenable to be solved using Gauss-Newton iterations. To fully exploit the structure of neural networks, the resulting linear least-squares problems are addressed by employing an approach based on forward dynamic programming. Finally, the effectiveness of our method is showcased on regression datasets.

Locations

  • arXiv (Cornell University) - View - PDF
  • Kyushu University Institutional Repository (QIR) (Kyushu University) - View - PDF
  • Lirias (KU Leuven) - View - PDF
  • 2021 60th IEEE Conference on Decision and Control (CDC) - View
  • DataCite API - View

Similar Works

Action Title Year Authors
+ Neural Network Training as an Optimal Control Problem: An Augmented Lagrangian Approach 2021 Brecht Evens
Puya Latafat
Andreas Themelis
Johan A. K. Suykens
Panagiotis Patrinos
+ PDF Chat Controlled descent training 2024 Viktor Andersson
Balázs Varga
Vincent Szolnoky
Andreas Syrén
Rebecka Jörnsten
Balázs Kulcsár
+ Controlled Descent Training 2023 Viktor Andersson
Balázs Varga
Vincent Szolnoky
Andreas Syrén
Rebecka Jörnsten
Balázs Kulcsár
+ Artificial Neural Networks Nonlinear Least Squares Learning 2006 水谷英二
Eiji Mizutani
+ Optimization-Informed Neural Networks 2022 Dawen Wu
Abdel Lisser
+ Efficient Global Optimization of Two-Layer ReLU Networks: Quadratic-Time Algorithms and Adversarial Training 2023 Yatong Bai
Tanmay Gautam
Somayeh Sojoudi
+ Efficient Global Optimization of Two-layer ReLU Networks: Quadratic-time Algorithms and Adversarial Training 2022 Yatong Bai
Tanmay Gautam
Somayeh Sojoudi
+ Deep Learning Theory Review: An Optimal Control and Dynamical Systems Perspective 2019 Guan-Horng Liu
Evangelos A. Theodorou
+ PDF Chat Enhancing Deep Learning with Optimized Gradient Descent: Bridging Numerical Methods and Neural Network Training 2024 Yuhan Ma
Dan Sun
Erdi Gao
Ningjing Sang
Iris Li
Guanming Huang
+ PDF Chat A PINN approach for the online identification and control of unknown PDEs 2024 Alessandro Alla
Giulia Bertaglia
Elisa Calzola
+ PDF Chat Near-optimal control of dynamical systems with neural ordinary differential equations 2022 Lucas Böttcher
Thomas Asikis
+ PDF Chat Penalty Adversarial Network (PAN): A neural network-based method to solve PDE-constrained optimal control problems 2024 Shilin Ma
Yukun Yue
+ Optimization Methods for Supervised Machine Learning: From Linear Models to Deep Learning 2017 Frank E. Curtis
Katya Scheinberg
+ Stochastic Training of Neural Networks via Successive Convex Approximations 2017 Simone Scardapane
Paolo Di Lorenzo
+ Stochastic Training of Neural Networks via Successive Convex Approximations 2017 Simone Scardapane
Paolo Di Lorenzo
+ PDF Chat Stochastic Training of Neural Networks via Successive Convex Approximations 2018 Simone Scardapane
Paolo Di Lorenzo
+ PDF Chat Self-Supervised Learning of Iterative Solvers for Constrained Optimization 2024 Lukas Lüken
Sergio Lucia
+ DDPNOpt: Differential Dynamic Programming Neural Optimizer 2020 Guan-Horng Liu
Tianrong Chen
Evangelos A. Theodorou
+ DDPNOpt: Differential Dynamic Programming Neural Optimizer 2020 Guan-Horng Liu
Tianrong Chen
Evangelos A. Theodorou
+ Augmented Lagrangian Methods as Layered Control Architectures 2023 Anusha Srikanthan
Vijay Kumar
Nikolai Matni