Normalizing Flows for Domain Adaptation when Identifying $Λ$ Hyperon Events

Rowan Kelleher,Anselm Vossen

2024-05-15

Abstract:This study focuses on the novel application of a normalizing flow as a method of domain adaptation. Normalizing flows offer a way to transform data points between two different distributions. The present study investigates a method of transforming latent representations of physics data to a normal distribution and then to a physics distribution again. The final distribution models a simulated distribution. Following the transformation process, the data can be classified by a neural network trained on labeled simulation data. The present study succeeds in training two normalizing flows that can transform between data (or simulation) and a Gaussian distribution.

Data Analysis, Statistics and Probability,High Energy Physics - Experiment

What problem does this paper attempt to address?

This paper discusses how to use Normalizing Flows for domain adaptation in the identification of Hyperon events, such as Lambda hyperons. In the semi-inclusive deep inelastic scattering experiment, researchers need to infer the existence of Lambda hyperons by studying their decay products since the lifetime of Lambda hyperons is too short to be directly detected. However, there are differences between simulated data and actual measured data, which affects the accurate identification of Lambda hyperon events by classifiers.

The paper proposes a new method that uses two Normalizing Flow networks to transform the latent representation of physical data into a Gaussian distribution and then back to the physical distribution. This approach can improve the performance of classifiers in handling measured data. The paper implements a Normalizing Flow model based on the RealNVP architecture and trains it on both measured data and simulated data for bi-directional transformation between data and simulated data distributions. The training results show that the model successfully learns the normalization of the input.

The application section explains that by using this domain adaptation method with Normalizing Flows, the ability of classifiers to identify simulated Lambda events can be improved, thereby enhancing the signal-to-background noise ratio, increasing confidence in signal and background separation, and reducing statistical uncertainty. The results show that the transformed data matches better with the classifier training data, improving the detection performance of Lambda events.

The paper concludes by suggesting that future work can explore other types of flow models to further improve this process.