Search for: All records

In many applications, Neural Nets (NNs) have classification performance on par or even exceeding human capacity. Moreover, it is likely that NNs leverage underlying features that might differ from those humans perceive to classify. Can we "reverse-engineer" pertinent features to enhance our scientific understanding? Here, we apply this idea to the notoriously difficult task of galaxy classification: NNs have reached high performance for this task, but what does a neural net (NN) "see" when it classifies galaxies? Are there morphological features that the human eye might overlook that could help with the task and provide new insights? Can we visualize tracers of early evolution, or additionally incorporated spectral data? We present a novel way to summarize and visualize galaxy morphology through the lens of neural networks, leveraging Dataset Distillation, a recent deep-learning methodology with the primary objective to distill knowledge from a large dataset and condense it into a compact synthetic dataset, such that a model trained on this synthetic dataset achieves performance comparable to a model trained on the full dataset. We curate a class-balanced, medium-size high-confidence version of the Galaxy Zoo 2 dataset, and proceed with dataset distillation from our accurate NN-classifier to create synthesized prototypical images of galaxy morphological features, demonstrating its effectiveness. Of independent interest, we introduce a self-adaptive version of the state-of-the-art Matching Trajectory algorithm to automate the distillation process, and show enhanced performance on computer vision benchmarks. 

Abstract Mantle‐induced dynamic topography (i.e., subsidence and uplift) has been increasingly recognized as an important process in foreland basin development. However, characterizing and distinguishing the effects (i.e., location, extent and magnitude) of dynamic topography in ancient foreland basins remains challenging because the spatio‐temporal footprint of dynamic topography and flexural topography (i.e., generated by topographic loading) can overlap. This study employs 3D flexural backstripping of Upper Cretaceous strata in the central part of the North American Cordilleran foreland basin (CFB) to better quantify the effects of dynamic topography. The extensive stratigraphic database and good age control of the CFB permit the regional application of 3D flexural backstripping in this basin for the first time. Dynamic topography started to influence the development of the CFB during the late Turonian to middle Campanian (90.2–80.2 Ma) and became the dominant subsidence mechanism during the middle to late Campanian (80.2–74.6 Ma). The area influenced by >100 m dynamic subsidence is approximately 400 by 500 km, within which significant (>200 m) dynamic subsidence occurs in an irregular‐shaped (i.e., lunate) subregion. The maximum magnitude of dynamic subsidence is 300 ± 100 m based on the 80.2–74.6 Ma tectonic subsidence maps. With the maximum magnitude of dynamic uplift being constrained to be 200–300 m, the gross amount of dynamic topography in the Late Cretaceous CFB is 500–600 m. Although the location of dynamic subsidence revealed by tectonic subsidence maps is generally consistent with isopach map trends, tectonic subsidence maps developed through 3D flexural backstripping provide more accurate constraints of the areal extent, magnitude and rate of dynamic topography (as well as flexural topography) in the CFB through the Late Cretaceous. This improved understanding of dynamic topography in the CFB is critical for refining current geodynamic models of foreland basins and understanding the surface expression of mantle processes. 

Dynamic topography refers to the vertical deflection (i.e., uplift and subsidence) of the Earth’s surface generated in response to mantle flow. Although dynamic subsidence has been increasingly invoked to explain the subsidence and migration of depocenters in the Late Cretaceous North American Cordilleran foreland basin (CFB), it remains a challenging task to discriminate the effects of dynamic mantle processes from other subsidence mechanisms, and the spatial and temporal scales of dynamic topography is not well known. To unravel the relationship between sedimentary systems, accommodation, and subsidence mechanisms of the CFB through time and space, a high-resolution chronostratigraphic framework was developed for the Upper Cretaceous strata based on a dense data set integrating >600 well logs from multiple basins/regions in Wyoming, Utah, Colorado, and New Mexico, USA. The newly developed stratigraphic framework divides the Upper Cretaceous strata into four chronostratigraphic packages separated by chronostratigraphic surfaces that can be correlated regionally and constrained by ammonite biozones. Regional isopach patterns and shoreline trends constructed for successive time intervals suggest that dynamic subsidence influenced accommodation creation in the CFB starting from ca. 85 Ma, and this wave of subsidence increasingly affected the CFB by ca. 80 Ma as subsidence migrated from the southwest to northeast. During 100−75 Ma, the depocenter migrated from central Utah (dominantly flexural subsidence) to north-central Colorado (dominantly dynamic subsidence). Subsidence within the CFB during 75−66 Ma was controlled by the combined effects of flexural subsidence induced by local Laramide uplifts and dynamic subsidence. Results from this study provide new constraints on the spatio-temporal footprint and migration of large-scale (>400 km × 400 km) dynamic topography at an average rate ranging from ∼120 to 60 km/m.y. in the CFB through the Late Cretaceous. The wavelength and location of dynamic topography (subsidence and uplift) generated in response to the subduction of the conjugate Shatsky Rise highly varied through both space and time, probably depending on the evolution of the oceanic plateau (e.g., changes in its location, subduction angle and depth, and buoyancy). Careful, high-resolution reconstruction of regional stratigraphic frameworks using three-dimensional data sets is critical to constrain the influence of dynamic topography. The highly transitory effects of dynamic topography need to be incorporated into future foreland basin models to better reconstruct and predict the formation of foreland basins that may have formed under the combined influence of upper crustal flexural loading and dynamic subcrustal loading associated with large-scale mantle flows.