Abstract
Simulators often provide the best description of real-world phenomena. However, the probability density that they implicitly define is often intractable, leading to challenging inverse problems for inference. Recently, a number of techniques have been introduced in which a surrogate for the intractable density is learned, including normalizing flows and density ratio estimators. We show that additional information that characterizes the latent process can often be extracted from simulators and used to augment the training data for these surrogate models. We introduce several loss functions that leverage these augmented data and demonstrate that these techniques can improve sample efficiency and quality of inference.
Highlights
Simulators often provide the best description of real-world phenomena
The key requirement of our approach is that additional information that characterizes the latent process can be extracted from the simulator as we explain in Extracting More Information from the Simulator
We have presented a family of inference techniques for the setting in which the likelihood is only implicitly defined through a stochastic generative simulator
Summary
We consider a scientific simulator that implements a stochastic generative process that proceeds through a series of latent states zi ∈ Zi and to an output x ∈ Rdx. In analogy to the Galton board toy example, even complicated real-world simulators often allow us to accumulate these factors as they run and to calculate the joint score and joint likelihood ratio conditional on a particular stochastic execution trace z. It may sound as if the joint likelihood ratio and joint score might only be accessible in specially tailored toy examples and are not available in complicated real-life simulators, but that is not the case. 3) the development of simulators within probabilistic programming frameworks is an alternative, more general solution; in this case, the joint likelihood ratio and joint score can be calculated without requiring additional domain knowledge or changes to the simulator code. Our work should be considered as a motivating example to implement access to the joint likelihood ratio and joint score through any of these strategies
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
