Sequential Neural Likelihood: Fast Likelihood-free Inference with Autoregressive Flows

Abstract

We present Sequential Neural Likelihood (SNL), a new method for Bayesian inference in simulator models, where the likelihood is intractable but simulating data from the model is possible. SNL trains an autoregressive flow on simulated data in order to learn a model of the likelihood in the region of high posterior density. A sequential training procedure guides simulations and reduces simulation cost by orders of magnitude. We show that SNL is more robust, more accurate and requires less tuning than related neural-based methods, and we discuss diagnostics for assessing calibration, convergence and goodness-of-fit.

Accepted for publication at AISTATS 2019

References in corpus (15)

• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• WaveNet: A Generative Model for Raw Audio
• Approximate Bayesian computation scheme for parameter inference and model selection in dynamical systems
• MADE: Masked Autoencoder for Distribution Estimation
• Validating Bayesian Inference Algorithms with Simulation-Based Calibration
• Mining gold from implicit models to improve likelihood-free inference
• Massive optimal data compression and density estimation for scalable, likelihood-free inference in cosmology
• A Guide to Constraining Effective Field Theories with Machine Learning
• Flexible statistical inference for mechanistic models of neural dynamics
• Likelihood-free inference with emulator networks
• Likelihood-free inference with an improved cross-entropy estimator
• Recognition Networks for Approximate Inference in BN20 Networks
• Bootstrapped synthetic likelihood
• Improvements to Inference Compilation for Probabilistic Programming in Large-Scale Scientific Simulators
• Automatic Variational ABC

Cited by in corpus (23)

• Machine learning and the physical sciences
• The frontier of simulation-based inference
• Mining gold from implicit models to improve likelihood-free inference
• Nuisance hardened data compression for fast likelihood-free inference
• Mining for Dark Matter Substructure: Inferring subhalo population properties from strong lenses with machine learning
• Inference of the optical depth to reionization from low multipole temperature and polarisation Planck data
• Conditional Density Estimation Tools in Python and R with Applications to Photometric Redshifts and Likelihood-Free Cosmological Inference
• Simulation-based inference of dynamical galaxy cluster masses with 3D convolutional neural networks
• Real-Time Likelihood-Free Inference of Roman Binary Microlensing Events with Amortized Neural Posterior Estimation
• Likelihood-free MCMC with Amortized Approximate Ratio Estimators
• Black-Box Optimization with Local Generative Surrogates
• Effective LHC measurements with matrix elements and machine learning
• Some models are useful, but how do we know which ones? Towards a unified Bayesian model taxonomy
• Likelihood-free Forward Modeling for Cluster Weak Lensing and Cosmology
• Fully probabilistic quasar continua predictions near Lyman-α with conditional neural spline flows
• Accelerating delayed-acceptance Markov chain Monte Carlo algorithms
• Guided sequential ABC schemes for intractable Bayesian models
• Modeling the Gaia Color-Magnitude Diagram with Bayesian Neural Flows to Constrain Distance Estimates
• Sequentially guided MCMC proposals for synthetic likelihoods and correlated synthetic likelihoods
• Sequential Neural Methods for Likelihood-free Inference
• IV-Posterior: Inverse Value Estimation for Interpretable Policy Certificates
• Optimal Design of Experiments for Simulation-Based Inference of Mechanistic Acyclic Biological Networks
• Lecture notes on Machine Learning applications for global fits