Input Convex Neural Networks

Abstract

This paper presents the input convex neural network architecture. These are scalar-valued (potentially deep) neural networks with constraints on the network parameters such that the output of the network is a convex function of (some of) the inputs. The networks allow for efficient inference via optimization over some inputs to the network given others, and can be applied to settings including structured prediction, data imputation, reinforcement learning, and others. In this paper we lay the basic groundwork for these models, proposing methods for inference, optimization and learning, and analyze their representational power. We show that many existing neural network architectures can be made input-convex with a minor modification, and develop specialized optimization algorithms tailored to this setting. Finally, we highlight the performance of the methods on multi-label prediction, image completion, and reinforcement learning problems, where we show improvement over the existing state of the art in many cases.

arXiv commentICML 2017

Cited by in corpus (71)

• QT-Opt: Scalable Deep Reinforcement Learning for Vision-Based Robotic Manipulation
• Polyconvex anisotropic hyperelasticity with neural networks
• OptNet: Differentiable Optimization as a Layer in Neural Networks
• Deep Learning and Knowledge-Based Methods for Computer Aided Molecular Design -- Toward a Unified Approach: State-of-the-Art and Future Directions
• FACMAC: Factored Multi-Agent Centralised Policy Gradients
• Q-Prop: Sample-Efficient Policy Gradient with An Off-Policy Critic
• Discrete Sequential Prediction of Continuous Actions for Deep RL
• Learning Stable Deep Dynamics Models
• Reinforcement Learning with Combinatorial Actions: An Application to Vehicle Routing
• Learning Approximate Inference Networks for Structured Prediction
• Input Convex Neural Networks for Building MPC
• Optimal Control Via Neural Networks: A Convex Approach
• Input Convex Neural Networks for Optimal Voltage Regulation
• 2-Wasserstein Approximation via Restricted Convex Potentials with Application to Improved Training for GANs
• Wasserstein-2 Generative Networks
• Scalable Computations of Wasserstein Barycenter via Input Convex Neural Networks
• State space models for building control: how deep should you go?
• Learned convex regularizers for inverse problems
• Sampling-based Reachability Analysis: A Random Set Theory Approach with Adversarial Sampling
• CAQL: Continuous Action Q-Learning
• Structured Hammerstein-Wiener Model Learning for Model Predictive Control
• Counterexample-Guided Learning of Monotonic Neural Networks
• $f$-GAIL: Learning $f$-Divergence for Generative Adversarial Imitation Learning
• Transport away your problems: Calibrating stochastic simulations with optimal transport
• Data-Driven Random Access Optimization in Multi-Cell IoT Networks with NOMA
• Large-Scale Wasserstein Gradient Flows
• Do Neural Optimal Transport Solvers Work? A Continuous Wasserstein-2 Benchmark
• Neuro-Symbolic Constraint Programming for Structured Prediction
• Learning Convex Optimization Control Policies
• Offline Reinforcement Learning for Autonomous Driving with Safety and Exploration Enhancement
• Continuous Wasserstein-2 Barycenter Estimation without Minimax Optimization
• Almost Surely Stable Deep Dynamics
• Convex Potential Flows: Universal Probability Distributions with Optimal Transport and Convex Optimization
• Optimizing Functionals on the Space of Probabilities with Input Convex Neural Networks
• Nonstationary Nonparametric Online Learning: Balancing Dynamic Regret and Model Parsimony
• Projection Efficient Subgradient Method and Optimal Nonsmooth Frank-Wolfe Method
• EMaQ: Expected-Max Q-Learning Operator for Simple Yet Effective Offline and Online RL
• The Differentiable Cross-Entropy Method
• Convex Regularization Behind Neural Reconstruction
• Neural Weighted A*: Learning Graph Costs and Heuristics with Differentiable Anytime A*
• Generative Modeling with Optimal Transport Maps
• A Convex Neural Network Solver for DCOPF with Generalization Guarantees
• Implicitly Defined Layers in Neural Networks
• Provable Model-based Nonlinear Bandit and Reinforcement Learning: Shelve Optimism, Embrace Virtual Curvature
• Automatically Learning Compact Quality-aware Surrogates for Optimization Problems
• An Inductive Bias for Distances: Neural Nets that Respect the Triangle Inequality
• NOVAS: Non-convex Optimization via Adaptive Stochastic Search for End-to-End Learning and Control
• Learning Dynamics Models with Stable Invariant Sets
• Machine learning moment closure models for the radiative transfer equation III: enforcing hyperbolicity and physical characteristic speeds
• Continuous Control with Action Quantization from Demonstrations
• On the Existence of Optimal Transport Gradient for Learning Generative Models
• Deep Learning Alternative to Explicit Model Predictive Control for Unknown Nonlinear Systems
• Learning emergent PDEs in a learned emergent space
• Differentiable Forward and Backward Fixed-Point Iteration Layers
• Learning Discriminators as Energy Networks in Adversarial Learning
• Monotonic Neural Network: combining Deep Learning with Domain Knowledge for Chiller Plants Energy Optimization
• Importance Sampling based Exploration in Q Learning
• Overestimation learning with guarantees
• Guide Actor-Critic for Continuous Control
• Extended Koopman Models
• Learning PSD-valued functions using kernel sums-of-squares
• Input Convex Gradient Networks
• Learning Energy-Based Approximate Inference Networks for Structured Applications in NLP
• CDiNN -Convex Difference Neural Networks
• The Expressivity and Training of Deep Neural Networks: toward the Edge of Chaos?
• Structured Prediction in NLP -- A survey
• End-to-End Learning and Intervention in Games
• Learning Deep Energy Shaping Policies for Stability-Guaranteed Manipulation
• Physics Informed Convex Artificial Neural Networks (PICANNs) for Optimal Transport based Density Estimation
• Learning convex regularizers satisfying the variational source condition for inverse problems
• Input Invex Neural Network