OptNet: Differentiable Optimization as a Layer in Neural Networks

Abstract

This paper presents OptNet, a network architecture that integrates optimization problems (here, specifically in the form of quadratic programs) as individual layers in larger end-to-end trainable deep networks. These layers encode constraints and complex dependencies between the hidden states that traditional convolutional and fully-connected layers often cannot capture. We explore the foundations for such an architecture: we show how techniques from sensitivity analysis, bilevel optimization, and implicit differentiation can be used to exactly differentiate through these layers and with respect to layer parameters; we develop a highly efficient solver for these layers that exploits fast GPU-based batch solves within a primal-dual interior point method, and which provides backpropagation gradients with virtually no additional cost on top of the solve; and we highlight the application of these approaches in several problems. In one notable example, the method is learns to play mini-Sudoku (4x4) given just input and output games, with no a-priori information about the rules of the game; this highlights the ability of OptNet to learn hard constraints better than other neural architectures.

ICML 2017

References in corpus (2)

• On Differentiating Parameterized Argmin and Argmax Problems with Application to Bi-level Optimization
• End-to-End Learning for Structured Prediction Energy Networks

Cited by in corpus (28)

• Safe Exploration in Continuous Action Spaces
• Lyapunov-based Safe Policy Optimization for Continuous Control
• Universal Planning Networks
• Deep Graph Matching via Blackbox Differentiation of Combinatorial Solvers
• An End-to-End Differentiable Framework for Contact-Aware Robot Design
• SATNet: Bridging deep learning and logical reasoning using a differentiable satisfiability solver
• Simple, Distributed, and Accelerated Probabilistic Programming
• Semi-Amortized Variational Autoencoders
• Learning to Solve Network Flow Problems via Neural Decoding
• Recurrent Relational Networks
• DiffLoop: Tuning PID controllers by differentiating through the feedback loop
• Large-scale Grid Optimization: The Workhorse of Future Grid Computations
• OptLayer - Practical Constrained Optimization for Deep Reinforcement Learning in the Real World
• Melding the Data-Decisions Pipeline: Decision-Focused Learning for Combinatorial Optimization
• End-to-End Reinforcement Learning of Koopman Models for Economic Nonlinear Model Predictive Control
• Differentiable Rendering with Perturbed Optimizers
• Learning to Configure Mathematical Programming Solvers by Mathematical Programming
• Meta Learning for Few-Shot One-class Classification
• Integrating Algorithmic Planning and Deep Learning for Partially Observable Navigation
• Efficient Gradient Approximation Method for Constrained Bilevel Optimization
• Real-Time Visual Object Tracking via Few-Shot Learning
• Large Scale Learning of Agent Rationality in Two-Player Zero-Sum Games
• Proximal Mapping for Deep Regularization
• GPU Accelerated Batch Multi-Convex Trajectory Optimization for a Rectangular Holonomic Mobile Robot
• Optimizing for Generalization in Machine Learning with Cross-Validation Gradients
• Neural Conditional Gradients
• Multicell Power Control under Rate Constraints with Deep Learning
• Convergence Analysis and Design of Multi-block ADMM via Switched Control Theory