A mean-field optimal control formulation of deep learning
From MaRDI portal
Abstract: Recent work linking deep neural networks and dynamical systems opened up new avenues to analyze deep learning. In particular, it is observed that new insights can be obtained by recasting deep learning as an optimal control problem on difference or differential equations. However, the mathematical aspects of such a formulation have not been systematically explored. This paper introduces the mathematical formulation of the population risk minimization problem in deep learning as a mean-field optimal control problem. Mirroring the development of classical optimal control, we state and prove optimality conditions of both the Hamilton-Jacobi-Bellman type and the Pontryagin type. These mean-field results reflect the probabilistic nature of the learning problem. In addition, by appealing to the mean-field Pontryagin's maximum principle, we establish some quantitative relationships between population and empirical learning problems. This serves to establish a mathematical foundation for investigating the algorithmic and theoretical connections between optimal control and deep learning.
Recommendations
- Deep learning as optimal control problems: models and numerical methods
- Dynamical Systems andOptimal Control Approach to Deep Learning
- Solving stochastic optimal control problem via stochastic maximum principle with deep learning method
- Deep neural networks algorithms for stochastic control problems on finite horizon: convergence analysis
- scientific article; zbMATH DE number 1146238
- A projected primal-dual gradient optimal control method for deep reinforcement learning
- Deep learning for control: the state of the art and prospects
- Deep neural networks algorithms for stochastic control problems on finite horizon: numerical applications
- Neural Approximations for Optimal Control and Decision
- Control on the manifolds of mappings with a view to the deep learning
Cites work
- A general stochastic maximum principle for SDEs of mean-field type
- A maximum principle for SDEs of mean-field type
- A proposal on machine learning via dynamical systems
- Approximation Methods for Nonlinear Problems with Application to Two-Point Boundary Value Problems
- Bellman equation and viscosity solutions for mean-field stochastic control problem
- Calculus of variations and optimal control theory. A concise introduction
- Deep learning
- Dynamic programming for mean-field type control
- Dynamic Programming for Optimal Control of Stochastic McKean--Vlasov Dynamics
- Existence of a solution to an equation arising from the theory of mean field games
- Forward-backward stochastic differential equations and controlled McKean-Vlasov dynamics
- Hamilton-Jacobi equations in infinite dimensions. I: Uniqueness of viscosity solutions
- Hamilton-Jacobi equations in infinite dimensions. II: Existence of viscosity solutions
- Hamilton-Jacobi equations in infinite dimensions. III
- scientific article; zbMATH DE number 4211245 (Why is no real title available?)
- scientific article; zbMATH DE number 1181255 (Why is no real title available?)
- scientific article; zbMATH DE number 1965513 (Why is no real title available?)
- Introduction to the mathematical theory of control
- Large population stochastic dynamic games: closed-loop McKean-Vlasov systems and the Nash certainty equivalence principle
- Learning deep architectures for AI
- Maximum principle based algorithms for deep learning
- Mean field games
- Mean field games and applications
- Mean field games and mean field type control theory
- Mean-field optimal control
- Mean-field Pontryagin maximum principle
- Neural network with unbounded activation functions is universal approximator
- Optimization of functions on certain subsets of Banach spaces
- Remarks on Inequalities for Large Deviation Probabilities
- Sparse stabilization and control of alignment models
- Stable architectures for deep neural networks
- The elements of statistical learning. Data mining, inference, and prediction
- The method of characteristics for Hamilton-Jacobi equations and applications to dynamical optimization
- The theory of differential equations. Classical and qualitative
- User’s guide to viscosity solutions of second order partial differential equations
- Viscosity Solutions of Hamilton-Jacobi Equations
Cited in
(73)- A projected primal-dual gradient optimal control method for deep reinforcement learning
- Forward stability of ResNet and its variants
- Variational networks: an optimal control approach to early stopping variational methods for image restoration
- Selection dynamics for deep neural networks
- Spectral methods for nonlinear functionals and functional differential equations
- A mean field games approach to cluster analysis
- Rank-adaptive tensor methods for high-dimensional nonlinear PDEs
- Semiconcavity and sensitivity analysis in mean-field optimal control and applications
- A measure theoretical approach to the mean-field maximum principle for training NeurODEs
- Do ideas have shape? Idea registration as the continuous limit of artificial neural networks
- Dynamic tensor approximation of high-dimensional nonlinear PDEs
- Random features for high-dimensional nonlocal mean-field games
- A mean field game inverse problem
- A review on deep learning in medical image reconstruction
- A mean field games model for finite mixtures of Bernoulli and categorical distributions
- Deep learning as optimal control problems: models and numerical methods
- Deep relaxation: partial differential equations for optimizing deep neural networks
- Linear-quadratic stochastic delayed control and deep learning resolution
- A backward SDE method for uncertainty quantification in deep learning
- Control on the manifolds of mappings with a view to the deep learning
- Deep limits of residual neural networks
- Neural network architectures using min-plus algebra for solving certain high-dimensional optimal control problems and Hamilton-Jacobi PDEs
- Optimal control by deep learning techniques and its applications on epidemic models
- Computational mean-field games on manifolds
- Substantiation of the backpropagation technique via the Hamilton-Pontryagin formalism for training nonconvex nonsmooth neural networks
- The Random Feature Model for Input-Output Maps between Banach Spaces
- Maximum principle based algorithms for deep learning
- Neural ODEs as the deep limit of ResNets with constant weights
- Structure-preserving deep learning
- Continuous-domain assignment flows
- Algorithms for solving high dimensional PDEs: from nonlinear Monte Carlo to machine learning
- Time discretizations of Wasserstein-Hamiltonian flows
- Large Sample Mean-Field Stochastic Optimization
- Deep learning for ranking response surfaces with applications to optimal stopping problems
- Controlling propagation of epidemics via mean-field control
- Shared Prior Learning of Energy-Based Models for Image Reconstruction
- Disordered high-dimensional optimal control
- Dynamical Systems andOptimal Control Approach to Deep Learning
- The Continuous Formulation of Shallow Neural Networks as Wasserstein-Type Gradient Flows
- Value-Gradient Based Formulation of Optimal Control Problem and Machine Learning Algorithm
- Implicit integration of nonlinear evolution equations on tensor manifolds
- Optimal control of nonlocal continuity equations: numerical solution
- Approximation capabilities of measure-preserving neural networks
- Deep learning approximation of diffeomorphisms via linear-control systems
- An ODE-based neural network with Bayesian optimization
- A fast proximal gradient method and convergence analysis for dynamic mean field planning
- The Mori-Zwanzig formulation of deep learning
- Deep learning via dynamical systems: an approximation perspective
- Pontryagin's maximum principle and indirect descent method for optimal impulsive control of nonlocal transport equation
- Optimal control using to approximate probability distribution of observation set
- Optimal Dirichlet boundary control by Fourier neural operators applied to nonlinear optics
- Leveraging Multi-time Hamilton-Jacobi PDEs for Certain Scientific Machine Learning Problems
- Efficient and stable SAV-based methods for gradient flows arising from deep learning
- On dynamical system modeling of learned primal-dual with a linear operator \(\mathcal{K}\): stability and convergence properties
- An optimal control framework for adaptive neural ODEs
- Operator learning using random features: a tool for scientific computing
- An accurate numerical scheme for mean-field forward and backward SDEs with jumps
- A large multi-agent system with noise both in position and control
- Recent progress in Carleman estimates for mean field games
- Understanding the training of infinitely deep and wide ResNets with conditional optimal transport
- Cluster-based classification with neural ODEs via control
- Variational formulations of ODE-Net as a mean-field optimal control problem and existence results
- Primal-dual hybrid gradient algorithms for computing time-implicit Hamilton-Jacobi equations
- Directions in rough analysis. Abstracts from the workshop held November 3--8, 2024
- Scaling ResNets in the large-depth regime
- From NeurODEs to AutoencODEs: a mean-field control framework for width-varying neural networks
- Interpolation, approximation, and controllability of deep neural networks
- Convexification for a coefficient inverse problem for a system of two coupled nonlinear parabolic equations
- Convexification numerical method for a coefficient inverse problem for the system of nonlinear parabolic equations governing mean field games
- Optimal protocols for continual learning via statistical physics and control theory
- PDE models for deep neural networks: learning theory, calculus of variations and optimal control
- Stability analysis of hierarchical tensor methods for time-dependent PDEs
- Machine learning from a continuous viewpoint. I
This page was built for publication: A mean-field optimal control formulation of deep learning
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2319864)
