Turbo-Sim: a generalised generative model with a physical latent space

Abstract

We present Turbo-Sim, a generalised autoencoder framework derived from principles of information theory that can be used as a generative model. By maximising the mutual information between the input and the output of both the encoder and the decoder, we are able to rediscover the loss terms usually found in adversarial autoencoders and generative adversarial networks, as well as various more sophisticated related models. Our generalised framework makes these models mathematically interpretable and allows for a diversity of new ones by setting the weight of each loss term separately. The framework is also independent of the intrinsic architecture of the encoder and the decoder thus leaving a wide choice for the building blocks of the whole network. We apply Turbo-Sim to a collider physics generation problem: the transformation of the properties of several particles from a theory space, right after the collision, to an observation space, right after the detection in an experiment.

PC-JeDi: Diffusion for particle cloud generation in high energy physics

“…Parametrised models have been studied as replacements for expensive Monte Carlo (MC) simulation and detailed detector simulation, but in recent years deep generative models using Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and Normalising Flows have been used for detector simulation [8][9][10][11][12][13][14][15][16][17][18], event simulation [19][20][21][22][23][24][25][26][27][28][29][30], and the generation of jet constituents [31,32]. Typically the particles and particle showers generated in these approaches are represented by images or ordered vectors, and as such are not preserve permutation invariance.…”

“…In Figs. 13 to 16 we compare the target and generated m 30 jet and p 30 T for gluon jets and top jets. In all cases, the DDIM solver shows a more linear correspondence between target and generation.…”