Marietta College (MC) is a private liberal arts college in Marietta, Ohio. It offers more than 50 undergraduate majors across the arts, sciences, and engineering, as well as Physician Assistant, Psychology, Clinical Mental Health Counseling, and Athletic training graduate programs. Its campus encompasses approximately three city blocks next to downtown Marietta and enrolls 1,200 full-time students...
Summary Accurate lithology identification from open-hole logging data is fundamental to reservoir characterization, but conventional interpretation depends on core calibration or supervised classifiers that require labelled training data that are rarely available across long, sparsely cored stratigraphic intervals. Unsupervised clustering can group log responses into rock classes without labels, but purely statistical partitioning often produces clusters that do not correspond to true geological units, because both the number of clusters and cluster identity are determined by internal clustering statistics rather than by independent subsurface evidence. This disconnect between mathematical partitioning and geological reality limits the confidence with which clustering outputs can be used in downstream reservoir engineering workflows. We address this gap by developing a lithology classification workflow in which unsupervised clustering is anchored to independently determined well tops, so that the number of clusters and their geological identity are constrained by field evidence rather than by clustering statistics alone. Our objective is to demonstrate that a top-anchored, geology-guided clustering approach can classify lithology consistently in a training well and reproduce petrophysically consistent results in a blind validation well from the same basin, without retraining. Open-hole logging curves, comprising compensated bulk density, photoelectric factor, limestone-matrix neutron porosity, and gamma ray, were compiled for training well TIA-6 (2,606 valid depth samples over an approximately 400 m carbonate-dominated interval) and standardized prior to clustering. We constrained k-means clustering to exactly three clusters, matching the three lithologies identified in the field by the mud logger and wellsite geologist (dolomite, limestone, shale). Instead of random or k-means++ initialization, we initialized cluster centroids from the standardized log responses recorded at four known formation-top depths obtained from logging-while-drilling and office-based gamma-ray correlation, and we subsequently mapped the resulting clusters to lithology names by nearest-centroid matching against these anchors. We then applied the fitted scaler and anchored centroids, without any retraining or re-anchoring, to a second, fully logged well, TALA-7 (2,730 valid samples over an approximately 417 m interval) in the same basin, to provide an independent blind test of model transferability. In the training well, the anchored clustering assigned 69.4% of samples to dolomite, 21.3% to limestone, and 9.4% to shale, and reproduced all four known well-top depths exactly. The resulting lithology classes displayed clear and cross-checkable petrophysical separation, most notably in the photoelectric factor, with mean values of 3.79 b/e for dolomite and 5.24 b/e for limestone, closely matching standard chart-book endpoints for these minerals, and shale was correctly isolated by a markedly elevated gamma-ray response (mean 70.7 API-equivalent units versus 24.0 to 36.7 units for the carbonate classes). Internal cluster-quality metrics in the training well (silhouette coefficient = 0.414; Calinski-Harabasz index = 1,039.2; Davies-Bouldin index = 1.027) indicated moderate-to-good class separation, and a margin-based ambiguity analysis identified the dolomite-shale boundary as the dominant source of near-boundary uncertainty in both wells (47.5% and 61.5% of near-boundary samples in TIA-6 and TALA-7, respectively), consistent with transitional, clay-bearing dolomite facies. When we applied the trained model to the blind well TALA-7 without retraining, it produced petrophysically consistent lithology proportions (60.7% dolomite, 18.1% limestone, 21.3% shale) and preserved the same ordering of petrophysical endpoints across lithologies, with a moderate reduction in separation quality (silhouette = 0.319; Calinski-Harabasz = 987.3; Davies-Bouldin = 1.246). Using subsequently available sonic data, an independent, non-circular test of Gardner’s density-velocity relation showed uniformly weak fit (R-squared at most 0.12 in every lithology and both wells), indicating that density in this carbonate succession is governed primarily by porosity rather than by velocity alone; we also demonstrated a lithology-dependent shear-velocity computation end to end, though this used a deterministic transform from compressional velocity rather than an independently measured shear-sonic curve. These results indicate that constraining unsupervised clustering with a fixed, geologically justified cluster count, informed directly by mud-logger and wellsite-geologist lithology calls, and anchoring cluster identity to independently verified well tops produces a lithology model that is both statistically coherent and physically interpretable, and that transfers to an unseen well within the same basin without retraining. This top-anchored, depth-independent approach differs from conventional clustering by embedding domain knowledge directly into model initialization and label assignment, rather than relying on post hoc, subjective interpretation of anonymous clusters. The resulting continuous lithology log, combined with the diagnostic rock-physics findings reported here, provides a practical and honestly scoped basis for extending lithology-dependent workflows, including porosity-aware density-velocity recalibration, Castagna and Greenberg-Castagna shear-velocity prediction, Eaton-type pore-pressure analysis, and mechanical earth model construction, into long stratigraphic intervals and under-logged sections where core and advanced log coverage are limited by acquisition cost.
The purpose of this review was to examine the risk factors associated with ulnar collateral ligament (UCL) injuries in high-level baseball pitchers and to provide a comprehensive analysis of both nonsurgical and surgical treatment options. Particular emphasis is placed on recent advancements in surgical management of UCL injuries. As medical knowledge and surgical technology continue to evolve, novel techniques have emerged with the potential to optimize return-to-play timelines for competitive athletes while reducing intraoperative and postoperative morbidity. This review compares established reconstructive procedures—including the Jobe, modified Jobe, and docking techniques—with the more recently developed UCL repair with internal brace augmentation. The goal is to evaluate the advantages and limitations of each approach and to provide an in-depth review of both nonsurgical and surgical treatment options in the context of individual player characteristics, specific UCL injury patterns, and surgeon expertise.
Electronic health records (EHRs) capture the complex web of interactions among healthcare providers, patients, and clinical events, offering a rich foundation for predicting critical outcomes in intensive care units (ICUs). However, existing approaches often fail to fully leverage higher-order relational structures and nonlinear dependencies within these networks. To address this gap, we propose LAI-Pool, a persistence-aware pooling method that integrates our self-attention–guided persistence images (SAPI) with graph neural networks (GNNs) via cross attention. First, we model patient–provider interactions as ego networks constructed from EHRs, where nodes represent patients, admissions, diagnoses, procedures, providers, ICU stays, and ICU units, and edges encode the temporal order of care interactions. Then, we apply persistent homology to extract multi-scale topological features from these networks and fuse them with GNN representations through LAI-Pool. We benchmark our approach against persistent homology–only, GNN-only, and hybrid PH+GNN baselines on the MIMIC-IV dataset comprising 65,366 ICU patients. Our results demonstrate that LAI-Pool achieves superior performance in mortality prediction, and highlight the value of combining topological data analysis with deep graph models for advancing patient outcome prediction in critical care.
This chapter reviews the topic of learning by listening, exploring how it pertains to leader development, and how leaders might develop active listening skills (rather than passive listening skills) to truly connect with followers on a productive, meaningful, individual level. Given how much time leaders spend listening in classrooms, seminars, meetings, and so on, the concept of “moving the needle” through listening to build leader capacity is also described.
We present a fully three-dimensional kinetic framework for modeling intense short pulse lasers interacting with dielectric materials. Our work modifies the open-source particle-in-cell code EPOCH to include new models for photoionization and dielectric optical response. We use this framework to model the laser-induced damage of dielectric materials by few-cycle laser pulses. The framework is benchmarked against experimental results for bulk silica targets and then applied to model multi-layer dielectric mirrors with a sequence of simulations with varying laser fluence. This allows us to better understand the laser damage process by providing new insight into energy absorption, excited particle dynamics and nonthermal excited particle distributions. We compare common damage threshold metrics based on the energy density and excited electron density.