Black-Box Optimization with Local Generative Surrogates

Abstract

We propose a novel method for gradient-based optimization of black-box simulators using differentiable local surrogate models. In fields such as physics and engineering, many processes are modeled with non-differentiable simulators with intractable likelihoods. Optimization of these forward models is particularly challenging, especially when the simulator is stochastic. To address such cases, we introduce the use of deep generative models to iteratively approximate the simulator in local neighborhoods of the parameter space. We demonstrate that these local surrogates can be used to approximate the gradient of the simulator, and thus enable gradient-based optimization of simulator parameters. In cases where the dependence of the simulator on the parameter space is constrained to a low dimensional submanifold, we observe that our method attains minima faster than baseline methods, including Bayesian optimization, numerical optimization, and approaches using score function gradient estimators.

References in corpus (20)

• Adam: A Method for Stochastic Optimization
• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• Practical Bayesian Optimization of Machine Learning Algorithms
• A Brief Introduction to PYTHIA 8.1
• Variational Inference with Normalizing Flows
• A Literature Survey of Benchmark Functions For Global Optimization Problems
• The frontier of simulation-based inference
• Don't Decay the Learning Rate, Increase the Batch Size
• Pyro: Deep Universal Probabilistic Programming
• Improving Generalization Performance by Switching from Adam to SGD
• The Cramer Distance as a Solution to Biased Wasserstein Gradients
• A Compositional Object-Based Approach to Learning Physical Dynamics
• Scalable Global Optimization via Local Bayesian Optimization
• Likelihood-free inference with emulator networks
• Automatic Posterior Transformation for Likelihood-Free Inference
• Sequential Neural Likelihood: Fast Likelihood-free Inference with Autoregressive Flows
• Parallel Bayesian Global Optimization of Expensive Functions
• Deep ReLU network approximation of functions on a manifold
• Bayesian Optimization with Unknown Search Space
• Deep learning is adaptive to intrinsic dimensionality of model smoothness in anisotropic Besov space

Cited by in corpus (6)

• Mind Mappings: Enabling Efficient Algorithm-Accelerator Mapping Space Search
• Deep Generative Models for Detector Signature Simulation: A Taxonomic Review
• BitHEP -- The Limits of Low-Precision ML in HEP
• Programming with Neural Surrogates of Programs
• Automatic optimisation of a Parallel-Plate Avalanche Counter with Optical Readout
• Global Optimisation of Black-Box Functions with Generative Models in the Wasserstein Space