Mitigating data center bias in cancer classification: Transfer bias unlearning and feature size reduction via conflict-of-interest free multi-objective optimization.

Healthcare leaders face sustained uncertainty: workforce volatility, financial pressure, and accelerating technology change. In 2024-2025, Ardent Health advanced an AI-enabled virtual care model from pilot to production across multiple markets. The model integrates virtual nursing, virtual attending physicians and providers, virtual sitting, and hospital-to-home remote patient monitoring (RPM) into routine care, with artificial intelligence (AI), providing earlier risk detection and workflow relief. Specifically, AI systems (1) analyze video streams to detect fall risks and unsafe behaviors, prompting earlier alerts to staff; (2) continuously evaluate vital sign trends from wearable sensors to identify clinical deterioration sooner; and (3) support ambient documentation with speech recognition and natural language processing (NLP) that improves note quality and coding accuracy.At Ardent's East Texas location, five months of virtual nursing contributed to reductions in contract labor, a decrease in voluntary RN turnover, and improvements in salaries, wages, and benefits (SWB) per patient day despite an increase in volume. Meanwhile, virtual attending physicians and providers increased virtual patient consultations resulting in patient retention and, generated bed-day capacity; AI-assisted vitals monitoring correlated with lower mortality and shorter length of stay; and the RPM program improved discharge continuity and avoided readmissions.This article presents a case study and playbook to help leaders manage risk, scale safely, and measure value. Readiness includes updating consent form language; data-use and retention policies; training staff to obtain patient consent; establishing algorithm oversight with internal data, analytics, and data science capabilities; and investing in network, data center, and hardware upgrades. We close with lessons learned and an organizational performance tracking accountability structure.

Advanced phase change material-thermal energy storage for low-carbon heat: materials, reliability, and AI-driven system integration for industry, buildings, and data centers

This paper proposes a novel partial power converter based uninterruptible power supply (PPC-UPS) scheme for DC power feeding servers in data centers, in which a server load and its local backup battery are connected in series and compactly integrated to a common DC bus through one PPC. The proposed PPC-UPS scheme is modularized and distributed in multiple DC loads of a data center, with high scalability and reliability. The PPC-UPS scheme can reduce the number of power electronic converters by diverting the battery energy directly to the DC load via the PPC, while power conversion losses can also be significantly reduced as the PPC only undertakes a fraction of the DC load power. A PPC-UPS multi-mode control strategy is designed to autonomously and flexibly secure its continuous operations under various contingencies. Small-signal analysis is carried out to assess the system stability and obtain optimal control parameters. The performances of the PPC-UPS scheme are evaluated through hardware experiments, which illustrate that the PPC-UPS can reliably supply the DC loads under the contingencies of converter or battery outage and facilitate load reconnections.

Microchannel heat sinks for cold plate liquid cooling in data centers: Advances, evaluations and prospects

Improvement strategies based on airflow characteristic in a row-based cooling data center

Bi-level planning of data centers with coupled electricity-heat-computation system using data-driven scenario generation for representing uncertainties

Transient power unit efficiency prediction of a small data center through thermal and energetic analyses

This work proposes a multi-agent system aimed at increasing the computing sustainability of high-performance computing data centers that are distributed among several wind farms. The novel approach of wind turbines housing high-performance computing data centers seeks to maximize renewable energy usage by supplying the data centers with otherwise curtailed wind energy, thus increasing wind farm efficiency as well. To optimize data center operation in this unique environment, job execution should be prioritized during periods of high availability of renewable energy. When wind power generation is low, resource utilization should be continuously adjusted to minimize gray electricity consumption with high carbon intensity or high grid consumption costs. Furthermore, green service-level agreements are introduced allowing for more flexibility in terms of deadline compliance, thereby fostering energy-aware data center operation. The proposed multi-agent system realizes a moving-horizon, multi-objective optimization problem to find the best operational strategy, taking into account both sustainability and performance concerns, and is compared against a selection of baseline job scheduling strategies. • Historic Data-Driven Data Center Placement Strategy. • Scalable Multi-Agent System for Energy-Optimized Data Center Operation. • Two-step Moving-Horizon (Multi-Objective) Optimization Approach for Scheduling. • Implementation of Green Service-Level Agreements. • Thorough Evaluation of the Optimization Approach.

A review of the immersion liquid cooling technology for high-performance data centers

From incident to insight: Fire risk in modern data centers

Dual evaporating temperature steam generation heat pump system for waste heat recovery of air–liquid hybrid cooling data center

Energy optimization for data centers via carbon-aware multi-energy market coordination

Adaptability assessment of air-cooling systems for data center with varied rack power densities

Low volume has been recognized as a problem when benchmarking hospitals due to outcome rate instability. We asked if low-volume hospital outcomes, using matching to control for many clinical and sociodemographic characteristics, would expose quality problems not observed with CMS methods. Matched cohort study. Grades derive from mortality differences between all patients at the low-volume hospital and their matched controls. Medicare patients admitted with Acute Myocardial Infarction, Heart Failure and Pneumonia in 78 low-volume Pennsylvania acute care hospitals (combined condition volume=75≤N≤750 for the 3y, 2017-2019), using Medicare's Virtual Research Data Center. Thirty-day mortality. Using matching, 10 of 78 reportable low-volume hospitals had significantly higher mortality versus matched typical controls and 16 low-volume hospitals displayed significantly higher mortality versus well-resourced controls. In contrast, Medicare reported that only 3 of these same 78 hospitals had significantly higher mortality than "the national rate" on AMI, HF, or pneumonia. We find that some low-volume hospitals performed well. Other low-volume hospitals had significantly worse outcomes than both well-resourced and typical hospitals; and some displayed significantly worse mortality compared with well-resourced controls but did not reach significant differences from typical controls. In short, performing "no different from the national rate," as is almost always reported for low-volume hospitals when using CMS methods, does not imply a low-volume hospital has acceptable outcomes. Reports based on matching can expose low-volume hospital quality problems not apparent using standard methods. Low-volume hospitals have more quality problems than generally reported.

Practical considerations for the parametrization of the transition metal platinum within the third-order density-functional tight-binding (DFTB3) method are presented, enabling straightforward parametrizations of interactions between Pt and elements from the s-, p-, and d-blocks of the periodic table. The newly developed parameter set is fully compatible with the 3ob DFTB3 framework, thereby extending the chemical space accessible to DFTB and enabling rapid and reliable simulations of platinum-containing systems. The parameters were initially benchmarked against more than 1300 Pt-containing structures extracted from the Cambridge Crystallographic Data Centre, as well as over 50 reference systems optimized at the MP2/cc-pVTZ level of theory. Further validation included a challenging binuclear platinum(II) complex, QM/MM molecular dynamics (MD) simulations of Pt(II) complexes in aqueous solution, and 3d-periodic DFTB-based molecular dynamics simulations of cisplatin embedded in metal-organic framework (MOF) hosts. Analysis of the resulting trajectories demonstrates a robust and consistent description of platinum coordination environments. To facilitate reproducibility and adoption, example Python scripts covering each step of the parametrization workflow are provided as part of the Supporting Information.

We present the complete genome sequence of Sphingomonas sp. gentR, a strain exhibiting high-level resistance to gentamicin (MIC = 40 mg/mL). The genome was assembled from hybrid Illumina and Nanopore sequencing data into a gap-free sequence of 4.0 Mbp, comprising one chromosome and two plasmids. A total of 3,692 coding sequences were predicted, with comprehensive functional annotation revealing genes associated with antibiotic resistance, stress adaptation, and metabolic diversity. Three confirmed resistance genes-ANT(2″)-Ia, ANT(3″)-IIa, and Sul1-were co-localized within a genomic island on plasmid B. This dataset provides insight into the genetic basis of high-level aminoglycoside resistance in Sphingomonas and serves as a valuable resource for studying horizontal gene transfer, environmental adaptation, and bioremediation potential. The genome sequence is publicly available under GenBank accessions CP144670-CP144672 and China National Genomics Data Center (accession number GWHDOHA00000000).

ABSTRACT A significant proportion of power consumption in data centers is ultimately converted into low‐grade waste heat (WH), which is typically discharged into the atmosphere, resulting in substantial energy loss and environmental degradation. Utilizing heat pump (HP) systems to recover this WH for district heating presents a promising approach to improving energy efficiency and reducing environmental impact. While previous research primarily focused on the feasibility and technical implementation of such recovery systems, limited attention has been given to the co‐optimization of WH utilization, particularly in systems that integrate HPs with energy storage. To address this gap, this study proposes an intelligent control framework that integrates user‐side demand response with deep reinforcement learning to optimize system performance. Specifically, the twin delayed deep deterministic policy gradient algorithm is employed to generate real‐time, adaptive control strategies. Additionally, a feasible action screening mechanism is introduced to ensure that control actions conform to the physical constraints of the system, thereby enhancing training stability and learning efficiency. Simulation results demonstrate that, compared with a benchmark model, the proposed approach improves system profit by 61.8% and increases renewable energy surplus by 25.7%.

The earthquake doublet on 6 February 2023 served as an important test in Türkiye. It helped assess the vulnerability of Türkiye’s building stock under different seismic loading conditions across a large region. The widespread destruction and casualties observed in heavily damaged cities following the 6 February 2023 earthquakes served as a warning. This urged a re-evaluation of the seismic performance assessment framework and risk mitigation strategies. Seismic isolation technology is considered the best method for earthquake-resilient design. Passive control systems are primarily preferred for use in critical facilities, such as healthcare complexes and data centers. Properly designed seismically isolated hospital buildings exhibited superior performance during the 6 February 2023 earthquakes compared to fixed-base counterparts. However, their use in residential buildings in Türkiye is still limited due to impediments such as stringent code requirements and peer review processes. This study evaluates the effectiveness of the ELF procedure in the Turkish Seismic Design Code-2018, incorporating two site-specific studies and earthquake record scaling in Antakya city center. Moreover, it examines the influence of considering directivity effects for using seismic isolation systems in regions with high seismicity. An effective and rapid evaluation procedure is employed for the inelastic response of seismically isolated residential buildings in accordance with the TSDC-2018 without needing any particular academic or commercial software. A suite of differential equations using the design parameters is arranged to represent the overall dynamics of seismically isolated buildings. Disregarding the directivity effects in site-specific studies for the selected construction site in Antakya city center can result in large earthquake demands and careful attention should be given to reconstruction studies for urban planning and more detailed studies should be carried out including other complex mechanisms experienced during the 6 February 2023 Türkiye earthquake doublet.

A Novel Bidirectional and Crossed AWGR Based Interconnection for Large-Scale Data Centers