Atomistic modeling of functionalized magnetite surfaces with oxidation states

Understanding the atomic structure of magnetite-carboxylic acid interfaces iscrucial for tailoring nanocomposites involving this interface. We present aMonte Carlo (MC)-based method utilizing iron oxidation state exchange to modelmagnetite interfaces with tens of thousands of atoms - scales typicallyinaccessible by electronic structure calculations. By comparing the bindingsite preferences of carboxylic acids obtained from electronic structurecalculations, we validated the accuracy of our method. We found that theoxidation state distribution, and consequently binding site preference, dependon coverage and surface thickness, with a critical thickness signaling thetransition from layered to bulk-like oxidation states. The method presentedhere needs no interface specific parameterization, ensuring seamlesscompatibility with popular bimolecular force fields providing transferability,and simplifying the study of magnetite interfaces in general.

Adversarial Robustness of Discriminative Self-Supervised Learning in Vision

Self-supervised learning (SSL) has advanced significantly in visualrepresentation learning, yet comprehensive evaluations of its adversarialrobustness remain limited. In this study, we evaluate the adversarialrobustness of seven discriminative self-supervised models and one supervisedmodel across diverse tasks, including ImageNet classification, transferlearning, segmentation, and detection. Our findings suggest that discriminativeSSL models generally exhibit better robustness to adversarial attacks comparedto their supervised counterpart on ImageNet, with this advantage extending totransfer learning when using linear evaluation. However, when fine-tuning isapplied, the robustness gap between SSL and supervised models narrowsconsiderably. Similarly, this robustness advantage diminishes in segmentationand detection tasks. We also investigate how various factors might influenceadversarial robustness, including architectural choices, training duration,data augmentations, and batch sizes. Our analysis contributes to the ongoingexploration of adversarial robustness in visual self-supervised representationsystems.

Chance-constrained Linear Quadratic Gaussian Games for Multi-robot Interaction under Uncertainty

We address safe multi-robot interaction under uncertainty. In particular, weformulate a chance-constrained linear quadratic Gaussian game with couplingconstraints and system uncertainties. We find a tractable reformulation of thegame and propose a dual ascent algorithm. We prove that the algorithm convergesto a generalized Nash equilibrium of the reformulated game, ensuring thesatisfaction of the chance constraints. We test our method in drivingsimulations and real-world robot experiments. Our method ensures safety underuncertainty and generates less conservative trajectories than single-agentmodel predictive control.

HyConEx: Hypernetwork classifier with counterfactual explanations

In recent years, there has been a growing interest in explainable AI methods.We want not only to make accurate predictions using sophisticated neuralnetworks but also to understand what the model's decision is based on. One ofthe fundamental levels of interpretability is to provide counterfactualexamples explaining the rationale behind the decision and identifying whichfeatures, and to what extent, must be modified to alter the model's outcome. Toaddress these requirements, we introduce HyConEx, a classification model basedon deep hypernetworks specifically designed for tabular data. Owing to itsunique architecture, HyConEx not only provides class predictions but alsodelivers local interpretations for individual data samples in the form ofcounterfactual examples that steer a given sample toward an alternative class.While many explainable methods generated counterfactuals for external models,there have been no interpretable classifiers simultaneously producingcounterfactual samples so far. HyConEx achieves competitive performance onseveral metrics assessing classification accuracy and fulfilling the criteriaof a proper counterfactual attack. This makes HyConEx a distinctive deeplearning model, which combines predictions and explainers as an all-in-oneneural network. The code is available at https://github.com/gmum/HyConEx.

Photometric investigation of WY Leo: An Eclipsing binary system with Delta Scuti component

Eclipsing binary stars with Delta Scuti components are exceptional systemsfor gaining a deeper understanding of stellar systems. WY Leo is one suchsystem, exhibiting Delta Scuti-type oscillations. However, it has not beenextensively studied in the literature. WY Leo was observed by TransitingExoplanet Survey Satellite (TESS), providing two sectors of high-qualityphotometric data. This study focuses on the photometric analysis of WY Leo.Binary modeling was performed, and the system's fundamental stellar parameters,such as mass and radius, were determined. The pulsational properties of WY Leowere also investigated, revealing that the more luminous star exhibits DeltaScuti type oscillations with a period of 0.052 days. The position of theprimary component was examined on the HR diagram, and it was found to liewithin the Delta Scuti instability strip.

Graph-Based Floor Separation Using Node Embeddings and Clustering of WiFi Trajectories

Indoor positioning systems (IPSs) are increasingly vital for location-basedservices in complex multi-storey environments. This study proposes a novelgraph-based approach for floor separation using Wi-Fi fingerprint trajectories,addressing the challenge of vertical localization in indoor settings. Weconstruct a graph where nodes represent Wi-Fi fingerprints, and edges areweighted by signal similarity and contextual transitions. Node2Vec is employedto generate low-dimensional embeddings, which are subsequently clustered usingK-means to identify distinct floors. Evaluated on the Huawei UniversityChallenge 2021 dataset, our method outperforms traditional community detectionalgorithms, achieving an accuracy of 68.97%, an F1- score of 61.99%, and anAdjusted Rand Index of 57.19%. By publicly releasing the preprocessed datasetand implementation code, this work contributes to advancing research in indoorpositioning. The proposed approach demonstrates robustness to signal noise andarchitectural complexities, offering a scalable solution for floor-levellocalization.

A Comparison of Ukrainian Hospital Services and Functions Before and During the Russia-Ukraine War

Decoupling Implantation Prediction and Embryo Ranking in Machine Learning: The Impact of Clinical Data and Discarded Embryos

Automated live embryo imaging has transformed in vitro fertilization (IVF) into a data‐intensive field. Unlike clinicians who rank embryos from the same IVF cycle cohort based on the embryos visual quality and determine how many embryos to transfer based on clinical factors, machine learning solutions usually combine these steps by optimizing for implantation prediction and using the same model for ranking the embryos within a cohort. Herein, it is established that this strategy can lead to suboptimal selection of embryos. It is revealed that despite enhancing implantation prediction, inclusion of clinical properties hampers ranking. Moreover, it is found that ambiguous labels of failed implantations, due to either low‐quality embryos or poor clinical factors, confound both the optimal ranking and even implantation prediction. To overcome these limitations, conceptual and practical steps are proposed to enhance machine learning‐driven IVF solutions. These consist of separating the optimizing of implantation from ranking by focusing on visual properties for ranking and reducing label ambiguity.

Decoupling Implantation Prediction and Embryo Ranking in Machine Learning: The Impact of Clinical Data and Discarded Embryos

Decoupling Implantation Prediction and Embryo Ranking in Machine Learning Itay Erlich, Assaf Zaritsky, and co‐workers establish that optimizing a machine learning model to predict in vitro fertilization embryo implantation success by inclusion of clinical properties is not an optimal strategy for the task of embryo ranking (see article number 2400048 ). The reason for this is “shortcut learning”, the model relies on the clinical factor as a proxy for implantation – hampering its ability to approximate the embryo quality. The authors’ practical recommendation is to exclusively focus on the embryo intrinsic features for ranking.

Discovery plasma proteomics analyses identify proteins that interact with plasma ptau231 levels to predict longitudinal episodic memory trajectory over a 9‐year follow‐up period

Abstract Background There has been great progress towards identifying plasma biomarkers for Alzheimer’s disease (AD), though few studies have investigated interactions between proteomic measures and AD pathology. We aimed to identify plasma proteins predictive of episodic memory decline, an early clinical sign of AD, and assess whether such associations were altered by the presence of AD pathology. Method Vanderbilt Memory and Aging Project participants (n=350, 73±7 years, 41% female) without dementia at baseline underwent blood draw and serial neuropsychological assessments over 9‐year follow‐up (mean=6.1 years). Proteins were quantified using Olink® Explore 3072. Linear mixed‐effects regressions related protein levels to longitudinal episodic memory composite scores, including interactions with follow‐up time as the terms of interest. Models adjusted for age, sex, race/ethnicity, education, baseline cognitive status, and apolipoprotein E‐ε4 status. Follow‐up analyses included plasma phosphorylated tau 231 (ptau231) as a covariate, tested proteins x ptau231 interactions, and stratified by ptau231 tertile. Result Higher levels of 15 proteins (GFAP, NEFL, KLK4, SPON1, GH1, CXCL1, OGN, EPHA2, EDA2R, SHBG, CKAP4, DSG2, CD300LG, CLEC5A, NOTCH3) predicted faster decline in episodic memory (pFDR<0.05). With ptau231 included as a covariate, associations between 5 of the proteins (SHBG, CKAP4, DSG2, CD300LG, NOTCH3) and memory attenuated (pFDR>0.05). 38 proteins interacted with ptau231 to predict memory trajectory (pFDR<0.05). Among participants with high ptau23, 24 proteins (LGALS3, FSTL1, SORT1, CHAD, ADAM23, SERPINA3, COL5A1, SYT1, PTPRZ1, CD34, DPP6, POLR2F, PLAUR, SPAG1, PDE4D, PTK7, CSPG4, LRTM2, MEGF10, MYOC, REG4, SPON1, WIF1, EDDM3B) were associated with memory trajectory. Higher protein levels predicted faster decline, except for EDDM3B, for which lower levels predicted faster decline (p<0.05, pFDR>0.05). Among participants with low ptau231, lower levels of 8 proteins (FSTL1, SORT1, IL6R, PTPRM, AHNAK, AMIGO2, IL1RAP, ROBO1) predicted faster decline in memory. Two proteins (FSTL1, SORT1) were associated with memory trajectory in both low and high ptau231 groups, exhibiting opposite directions of effects in the setting of low versus high ptau231 (see Figures). Conclusion Results demonstrate associations between plasma proteomic measures and longitudinal memory trajectory differ by level of AD pathology. To maximize clinical utility of potential biomarkers discovered in plasma proteomic analyses, future studies should assess interactions with AD core biomarkers.