High-resolution seismic tomography of Long Beach, CA using machine learning

An UltraMNIST classification benchmark to train CNNs for very large images

Current convolutional neural networks (CNNs) are not designed for large scientific images with rich multi-scale features, such as in satellite and microscopy domain. A new phase of development of CNNs especially designed for large images is awaited. However, application-independent high-quality and challenging datasets needed for such development are still missing. We present the 'UltraMNIST dataset' and associated benchmarks for this new research problem of 'training CNNs for large images'. The dataset is simple, representative of wide-ranging challenges in scientific data, and easily customizable for different levels of complexity, smallest and largest features, and sizes of images. Two variants of the problem are discussed: standard version that facilitates the development of novel CNN methods for effective use of the best available GPU resources and the budget-aware version to promote the development of methods that work under constrained GPU memory. Several baselines are presented and the effect of reduced resolution is studied. The presented benchmark dataset and baselines will hopefully trigger the development of new CNN methods for large scientific images.

Deep learning for high-resolution seismic imaging

Seismic imaging techniques play a crucial role in interpreting subsurface geological structures by analyzing the propagation and reflection of seismic waves. However, traditional methods face challenges in achieving high resolution due to theoretical constraints and computational costs. Leveraging recent advancements in deep learning, this study introduces a neural network framework that integrates Transformer and Convolutional Neural Network (CNN) architectures, enhanced through Adaptive Spatial Feature Fusion (ASFF), to achieve high-resolution seismic imaging. Our approach directly maps seismic data to reflection models, eliminating the need for post-processing low-resolution results. Through extensive numerical experiments, we demonstrate the outstanding ability of this method to accurately infer subsurface structures. Evaluation metrics including Root Mean Square Error (RMSE), Correlation Coefficient (CC), and Structural Similarity Index (SSIM) emphasize the model's capacity to faithfully reconstruct subsurface features. Furthermore, noise injection experiments showcase the reliability of this efficient seismic imaging method, further underscoring the potential of deep learning in seismic imaging.

Physics-informed W-Net GAN for the direct stochastic inversion of fullstack seismic data into facies models

Predicting the subsurface spatial distribution of geological facies from fullstack geophysical data is a main step in the geo-modeling workflow for energy exploration and environmental tasks and requires solving an inverse problem. Generative adversarial networks (GANs) have shown great potential for geologically accurate probabilistic inverse modeling, but existing methods require multiple sequential steps and do not account for the spatial uncertainty of facies-dependent continuous properties, linking the facies to the observed geophysical data. This can lead to biased predictions of facies distributions and inaccurate quantification of the associated uncertainty. To overcome these limitations, we propose a GAN able to learn the physics-based mapping between facies and seismic domains, while accounting for the spatial uncertainty of such facies-dependent properties. During its adversarial training, the network reads the observed geophysical data, providing solutions to the inverse problems directly in a single step. The method is demonstrated on 2-D examples, using both synthetic and real data from the Norne field (Norwegian North Sea). The results show that the trained GAN can model facies patterns matching the spatial continuity patterns observed in the training images, fitting the observed geophysical data, and with a variability proportional to the spatial uncertainty of the facies-dependent properties.

Machine learning in acoustics: Theory and applications

Acoustic data provide scientific and engineering insights in fields ranging from biology and communications to ocean and Earth science. We survey the recent advances and transformative potential of machine learning (ML), including deep learning, in the field of acoustics. ML is a broad family of techniques, which are often based in statistics, for automatically detecting and utilizing patterns in data. Relative to conventional acoustics and signal processing, ML is data-driven. Given sufficient training data, ML can discover complex relationships between features and desired labels or actions, or between features themselves. With large volumes of training data, ML can discover models describing complex acoustic phenomena such as human speech and reverberation. ML in acoustics is rapidly developing with compelling results and significant future promise. We first introduce ML, then highlight ML developments in four acoustics research areas: source localization in speech processing, source localization in ocean acoustics, bioacoustics, and environmental sounds in everyday scenes.