Learning To Simulate

Abstract

Simulation is a useful tool in situations where training data for machine learning models is costly to annotate or even hard to acquire. In this work, we propose a reinforcement learning-based method for automatically adjusting the parameters of any (non-differentiable) simulator, thereby controlling the distribution of synthesized data in order to maximize the accuracy of a model trained on that data. In contrast to prior art that hand-crafts these simulation parameters or adjusts only parts of the available parameters, our approach fully controls the simulator with the actual underlying goal of maximizing accuracy, rather than mimicking the real data distribution or randomly generating a large volume of data. We find that our approach (i) quickly converges to the optimal simulation parameters in controlled experiments and (ii) can indeed discover good sets of parameters for an image rendering simulator in actual computer vision applications.

arXiv commentPublished at International Conference on Learning Representations (ICLR) 2019

References in corpus (9)

• Neural Architecture Search with Reinforcement Learning
• Semantic Foggy Scene Understanding with Synthetic Data
• Evolution Strategies as a Scalable Alternative to Reinforcement Learning
• CARLA: An Open Urban Driving Simulator
• Neural Optimizer Search with Reinforcement Learning
• Sim4CV: A Photo-Realistic Simulator for Computer Vision Applications
• Synthesizing Programs for Images using Reinforced Adversarial Learning
• Learning to Teach
• Adversarial Variational Optimization of Non-Differentiable Simulators

Cited by in corpus (8)

• Unsupervised Doodling and Painting with Improved SPIRAL
• StyleGAN2 Distillation for Feed-forward Image Manipulation
• Multimodal Safety-Critical Scenarios Generation for Decision-Making Algorithms Evaluation
• Neural Data Server: A Large-Scale Search Engine for Transfer Learning Data
• Less is More: Learning from Synthetic Data with Fine-grained Attributes for Person Re-Identification
• Learning to simulate complex scenes
• Population-Based Evolution Optimizes a Meta-Learning Objective
• Man versus Machine: AutoML and Human Experts' Role in Phishing Detection