RESUMEN
The stochastic synthesis of extreme, rare climate scenarios is vital for risk and resilience models aware of climate change, directly impacting society in different sectors. However, creating high-quality variations of under-represented samples remains a challenge for several generative models. This paper investigates quantizing reconstruction losses for helping variational autoencoders (VAE) better synthesize extreme weather fields from conventional historical training sets. Building on the classical VAE formulation using reconstruction and latent space regularization losses, we propose various histogram-based penalties to the reconstruction loss that explicitly reinforces the model to synthesize under-represented values better. We evaluate our work using precipitation weather fields, where models usually strive to synthesize well extreme precipitation samples. We demonstrate that bringing histogram awareness to the reconstruction loss improves standard VAE performance substantially, especially for extreme weather events.
RESUMEN
Current seismic processing workflows in the oil and gas industry involve several interactions between different experts to optimize the overall data quality in various tasks, such as noise attenuation, velocity analysis and horizon picking. While many machine learning-based approaches have been proposed to support each of those steps, most of them disregard expert interactions to guide the overall optimization. This paper presents geocycles, a cyclic learning approach that mimics this iterative process, which can be applied to different pre-stack seismic processing tasks. Our method refactor these processes considering training, testing, and evaluation sub-tasks, which allow the selection of samples for greedy sequential processes targeting an overall optimum quality for very large seismic datasets. We present encouraging results showing that a cyclic structure and efficient quality metrics improved overall outcomes in up to 128% for two different seismic processing tasks in comparison to a 1-cycle machine learning approach.