Radikal Therapeutics

We introduce a causal modeling framework that captures the input-output behavior of predictive models (e.g., machine learning models). The framework enables us to identify features that directly cause the predictions, which has broad implications for data collection and model evaluation. We then present sound and complete algorithms for discovering direct causes (from data) under some assumptions. Furthermore, we propose a novel independence rule that can be integrated with the algorithms to accelerate the discovery process as we demonstrate both theoretically and empirically.

Abstract Energy efficiency of heat engines (gas and steam turbines) for electricity production and propulsion scale with operating temperature following the Carnot cycle. Commercial Ni- and Co-based superalloys melt near 1500°C and rapidly lose mechanical strength beyond 1000°C. Refractory metals melt well above 2000°C but have inherent manufacturability challenges, like high ductile-to-brittle transition temperatures, that are significant barriers to adoption. Using density-functional theory guided design, we demonstrate tailored local lattice distortions that promote phase-stable, non-equiatomic refractory concentrated solid-solutions with both high ductility and strength. We exemplify this for single-phase, body-centered cubic Nb4Ta4V3Ti that exhibits castability, excellent room-temperature tensile yield strength (>1 GPa) and ductility (>20% uniform elongation), and exceptional high-temperature tensile strength (500 MPa at 1000°C). These findings illustrate a path for designing materials that hold great potential for advancing next-generation technologies like Gen-IV fission reactors, first-generation fusion-plasma reactors, and more efficient gas turbines for electricity generation and propulsion.

The discovery of van der Waals superconductors in recent years has generated a lot of excitement for their potentially novel pairing mechanisms. However, their typical atomic-scale thickness and micrometer-scale lateral dimensions impose severe challenges to investigations of pairing symmetry by conventional methods. We demonstrate an improved technique that employs high-quality-factor superconducting resonators to measure the kinetic inductance—up to one part per million—and loss of a van der Waals superconductor. We analyze the equivalent circuit model to extract the kinetic inductance, superfluid stiffness, penetration depth, and ratio of imaginary and real parts of the complex conductivity. We validate the technique by measuring aluminum and finding excellent agreement in both the zero-temperature superconducting gap as well as the complex conductivity data when compared with BCS theory. We then demonstrate the utility of the technique by measuring the kinetic inductance of multilayered niobium diselenide and discuss the limits to the accuracy of our technique when the transition temperature of the sample, NbSe2 at 7.06 K, approaches our Nb probe resonator at 8.59 K. Our method will be useful for practitioners in the growing fields of superconducting physics, materials science, and quantum sensing, as a means of characterizing superconducting circuit components and studying pairing mechanisms of the novel superconducting states which arise in layered two-dimensional materials and heterostructures. Published by the American Physical Society 2024

In Natural Language Processing (NLP), Machine Reading Comprehension (MRC) is the task of answering a question based on a given context. To handle questions in the medical domain, modern language models such as BioBERT, SciBERT and even ChatGPT are trained on vast amounts of in-domain medical corpora. However, in-domain pre-training is expensive in terms of time and resources. In this paper, we propose a resource-efficient approach for injecting domain knowledge into a model without relying on such domain-specific pre-training. Knowledge graphs are powerful resources for accessing medical information. Building on existing work, we introduce a method using Multi-Layer Perceptrons (MLPs) for aligning and integrating embeddings extracted from medical knowledge graphs with the embedding spaces of pre-trained language models (LMs). The aligned embeddings are fused with open-domain LMs BERT and RoBERTa that are fine-tuned for two MRC tasks, span detection (COVID-QA) and multiple-choice questions (PubMedQA). We compare our method to prior techniques that rely on a vocabulary overlap for embedding alignment and show how our method circumvents this requirement to deliver better performance. On both datasets, our method allows BERT/RoBERTa to either perform on par (occasionally exceeding) with stronger domain-specific models or show improvements in general over prior techniques. With the proposed approach, we signal an alternative method to in-domain pre-training to achieve domain proficiency. Our code is available here 1 .

This is a practitioner paper focused on the use of chat to communicate time-sensitive mission information within a disaggregated distributed battlespace environment. In such environments, operators may have dozens of chat conversations ongoing simultaneously. Although some of these messages are sorted according to key words to help the operators better prioritize them, these keywords are subject to typing errors and other corruption that may lead to the operator missing key decision-making information. Considering the present expansion of the battle space to include multi-mission and multi-domain, the number of chat sessions is greatly increasing. In this environment, the problem the operators face is how to reduce and prioritize the vastly expanding collection of chat messages to focus only on those chat messages that are most critical to meeting mission success and to do so accurately and dynamically. The approach presented here introduces a real-time tasking and fully autonomous system called the Disaggregated Distributed AI Chat Enabler (D2ACE). D2ACE applies AI/ML techniques to correct spelling, typos, and other corruption in chat messages, recognize uncommon language formats, and autonomously prioritize and reduce the quantity of chat messages to only those relevant to specific commander’s objectives and intent for the mission.

This paper discusses pairing the identification of supply obsolescence data with the use of a Readiness Based Spares (RBS) Modeling Tool to expand the capabilities of Spares Availability, Spares Modeling, and Obsolescence Management disciplines to frame all impacts in relation to system or mission readiness. Current state often results in a silo'd approach where obsolescence data is manually integrated with spares inventory into a custom dashboard that is not scalable or tied to a larger Sustainment Digital Thread. While this approach provides a quick reference to the status of a systems' inventory of spares and can be available to individuals working in a crossfunctional sustainment support team, it relies on manual data incorporation, is dependent on manual handoffs and data incorporation, and is moderately expensive to maintain. Most importantly, it lacks a larger connection to the design phases of a product lifecycle and therefore becomes a limiting factor in deploying a Digital Engineering methodology across the product lifecycle. This paper will discuss the benefits and limitations of the current state for supply availability and obsolescence tracking. It will then make recommendations for how a combination of digital thread approach to sustainment data and commercially available system readiness modeling software can be utilized to deploy a scalable capability for reporting changes to obsolescence and/or supply status as it relates to overall system or mission readiness.

In this paper we consider aspects of geometric observability for hypergraphs, extending our earlier work from the uniform to the nonuniform case. Hypergraphs, a generalization of graphs, allow hyperedges to connect multiple nodes and unambiguously represent multi-way relationships which are ubiquitous in many real-world networks including those that arise in biology. We consider polynomial dynamical systems with linear outputs defined according to hypergraph structure, and we propose methods to evaluate local, weak observability.

RTRC’s DAYS projects focused on new flow battery (FB) chemistries for long duration energy storage: one that pairs aqueous polysulfide and manganese active materials in strongly alkaline electrolytes (S/Mn); another explores ligand-modified metal (LMM) chemistries in moderate pH electrolyte conditions. The S/Mn chemistry met durability targets by running for over 500 cycles and 1000 hours. Combined with this durability, power densities of up to 60 mW/cm2 at 80% round-trip efficiency, and exceptionally low-cost S and Mn active materials, the S/Mn FB offers a pathway to meeting the <5¢/kW-hr levelized cost of storage (LCOS) target first proposed by the DAYS Program and accepted by the DOE Energy Storage Grand Challenge. The LMM chemistry made impressive progress as well, including durability tests that retained performance over 600 hours of cycling. Cell optimization allowed this chemistry to operate as high as 500 mW/cm2 at 80% round-trip efficiency, nearly 10x the power density of S/Mn and comparable to the power density achieved in RTRC’s flow battery stacks using acidic vanadium battery solutions. This enabled the LMM chemistry to also offer a pathway to the <5¢/kW-hr LCOS DAYS and DOE targets. The promising results from the LMM chemistry have led to a recent DOE AMO award to advance the technology to a 5 kW, 20 kW-hr prototype.