Neural ordinary differential equations (NODEs) have recently regained popularity as large-depth limits of a large class of neural networks. In particular, residual neural networks (ResNets) are equivalent to an explicit Euler discretization of an underlying NODE, where the transition from one layer to the next is one time step of the discretization. The relationship between continuous and discrete neural networks has been of particular interest. Notably, analysis from the ordinary differential equation viewpoint can potentially lead to new insights for understanding the behavior of neural networks in general. In this work, we take inspiration from differential equations to define the concept of stiffness for a ResNet via the interpretation of a ResNet as the discretization of a NODE. Here, we then examine the effects of stiffness on the ability of a ResNet to generalize, via computational studies on example problems coming from climate and chemistry models. We find that penalizing stiffness does have a unique regularizing effect, but we see no benefit to penalizing stiffness over L2 regularization (penalization of network parameter norms) in terms of predictive performance.
The Big Hill SPR site has a rich data set consisting of multi-arm caliper (MAC) logs collected from the cavern wells. This data set provides insight into the on-going casing deformation at the Big Hill site. This report summarizes the MAC surveys for each well and presents well longevity estimates where possible. Included in the report is an examination of the well twins for each cavern and a discussion on what may or may not be responsible for the different levels of deformation between some of the well twins. The report also takes a systematic view of the MAC data presenting spatial patterns of casing deformation and deformation orientation in an effort to better understand the underlying causes. The conclusions present a hypothesis suggesting the small-scale variations in casing deformation are attributable to similar scale variations in the character of the salt-caprock interface. These variations do not appear directly related to shear zones or faults.
Direct air capture (DAC) of CO2 is a negative emission technology under development to limit the impacts of climate change. The dilute concentration of CO2 in the atmosphere (~400 ppm) requires new materials for carbon capture with increased CO2 selectivity that is not met with current carbon capture materials. Porous liquids (PLs) are an emerging candidate for carbon capture and consists of a combination of solvents and porous hosts that creates a liquid with permanent porosity. The fundamental mechanisms of carbon capture in a PL are relatively unknown. To uncover these mechanisms, PLs were synthesized consisting of three different zeolitic-imidazolate framework (ZIF-8, ZIF-67, or ZIF-69) porous host in a water/glycol/2-methylimidazole solvent. The most stable composition was based on ZIF-8 and exhibited carbon capture following exposure to CO2. Density functional theory identified a three-step carbon capture mechanism based on (i) reaction of OH- with ethylene glycol in the solution followed by (ii) formation of 2-hydroxyethyl carbonate, which (iii) further react with OH- to form a carbonate species. This mechanism was validated with experimental nuclear magnetic resonance spectroscopy (NMR) to identify the dissolved carbonate phases and the decrease in the pH during CO2 exposure. Deuterated samples of the ZIF-8 PLs were synthesized and analyzed via neutron diffraction at the Spallation Neutron Sources at Oak Ridge National Laboratory. Results identified differences in diffraction for PLs pre- and post-CO2 exposure that will be combined with ab initio molecular dynamics data of the same PL composition to identify how the presence of a solvent-porous host interfaces results in carbon capture.
On May 16-20, 2022, federal mission partners (e.g., DOE Consequence Management, CDC, FDA, FBI, DHS) as well as integrated state, local, tribal, and territorial governments took part in Cobalt Magnet 22 (CM22), a large-scale, week-long radiological incident exercise in Austin, Texas, that linked several important national assets (National Search Program, Radiological Assistance Program, and Consequence Management [CM] personnel) into a single response effort. The exercise had nine (9) overarching Objectives and an additional 162 associated Critical Tasks for all the participating organizations. In total, 13 National Core Capabilities spanning 5 Mission Areas were represented in the final exercise. This exercise enabled a full range of capabilities to be fielded together and examine the operational connection between major assets, discover any resource shortages associated with conducting multiple mission areas simultaneously or in close succession, and identify any challenges related to leadership. This report summarizes nearly 100 successes and observations provided from players and controllers supporting the LA Division, Fly Away Laboratory (FAL) and Gamma Spectroscopist operations. The observations were categorized to align with the FRMAC programmatic functional areas to consider for future improvements: Logistics, CBRN Responder, Laboratory Analysis, Sampling and Monitoring, Health and Safety, Gamma Spectroscopist Operations, Fly Away Laboratory, and the FRMAC Interdivision Interoperability Group (FIIG).
