General Model Research Articles

Provably Trainable Rotationally Equivariant Quantum Machine Learning

Exploiting the power of quantum computation to realize superior machine learning algorithms has been a major research focus of recent years, but the prospects of quantum machine learning (QML) remain dampened by considerable technical challenges. A particularly significant issue is that generic QML models suffer from so-called barren plateaus in their training landscapes—large regions where cost function gradients vanish exponentially in the number of qubits employed, rendering large models effectively untrainable. A leading strategy for combating this effect is to build problem-specific models that take into account the symmetries of their data in order to focus on a smaller, relevant subset of Hilbert space. In this work, we introduce a family of rotationally equivariant QML models built upon the quantum Fourier transform, and leverage recent insights from the Lie-algebraic study of QML models to prove that (a subset of) our models do not exhibit barren plateaus. In addition to our analytical results we numerically test our rotationally equivariant models on a dataset of simulated scanning tunneling microscope images of phosphorus impurities in silicon, where rotational symmetry naturally arises, and find that they dramatically outperform their generic counterparts in practice. Published by the American Physical Society 2024

Adding metabolic tasks to human GEM models to improve the study of gene targets and their associated toxicities

Genetic minimal cut sets (gMCS) are genes that must be deactivated simultaneously to avoid unwanted states in a metabolic model. The concept of gMCS can be applied to two different scenarios. First, it can be used to identify potential gene toxicities in generic or healthy cell models. Second, it can be used to develop genetic strategies to target cancer cells and prevent their proliferation. Up to now, gMCS have been evaluated using the traditional procedure of preventing biomass production. This paper proposes an additional way: using essential metabolic tasks, which any human cell should perform, to enlarge the set of unwanted states. Including this addition can significantly improve the study of toxicities and reveal targets that can be used to treat unhealthy cells. Excluding metabolic tasks can cause important information to be overlooked, which could impact the study’s success. Regarding toxicities, using the generic Human model, the number of detected generic toxicities with metabolic tasks increases from 106 to 281 (136 gMCSs of length 1 and 49 of length 2). We have used the following context-specific models to evaluate specific toxicities in different healthy tissues: blood, pancreas, liver, heart, and kidney. Again, considering metabolic tasks, we have found new toxicities (lengths 1 and 2) whose inactivation could damage these healthy tissues.Our research strategy has been applied to identify new cancer drug targets in two myeloma cell lines. We obtained new therapeutic targets of lengths 1 and 2 for each cell line. After analyzing the data, we conclude that incorporating metabolic tasks into cancer models can reveal important therapeutic targets previously disregarded by the conventional method of inhibiting biomass production. This approach also improves the evaluation of potential drug toxicities.

Understanding the public stigma of mental illness: a mixed-methods, multi-level, exploratory triangulation study

BackgroundThis study examines the role of themata in understanding mental health-related stigma. It is motivated by the need for alternative theoretical-methodological approaches beyond the dominant frameworks in education and contact-based anti-stigma public health efforts, which have shown mixed effects. Specifically, it addresses the need for a more nuanced framework in stigma research, one that is sensitive to the dialogues through which people relate themselves to mental health and stigma in context.MethodsThe research employs an exploratory mixed-methods approach, including the analysis of 529 news reports, 20 focus group discussions, and 19 one-to-one interviews, all concerning representations of shared living arrangements with someone perceived to have experiences of mental illness. Thematic analysis and natural language processing are used within a convergent triangulation design to analyze the data.ResultsWe found that mental health and illness were communicated through an overarching Self/Other thema and five subordinate themata: normal/abnormal, harm/non-harm, bounded/non-bounded, and moral/immoral. Despite familiarity with psychological distress and ‘modern’ explanations of mental illness, concerns about social identity motivated representations of mental illness as a predominantly permanent, negative form of personhood marked by abnormality, harm, distance, and immorality. Additionally, concerns about personal vulnerability, including historically rooted fears of contagion, motivated distancing representations of mental illness, rather than neutral portrayals.ConclusionsThemata have under-developed theoretical and methodological potential for addressing mental health-related stigma, particularly in their ability to describe the dynamic ways in which culture motivates people to both resist and reproduce stigma, partly through ambivalences, absences, tensions, and ambiguities in representation. A critical discussion is provided on how themata may support ecological strategies in mental health campaigns over generic models, emphasizing the need to understand group knowledge and contact dynamics to mitigate adverse effects. Themata Public Health Unintended Consequences Mixed Methods Behaviour Change Natural Language Processing.

Neutrino anarchy from flavor deconstruction

Flavor deconstruction refers to non-universal gauge extensions where the original gauge symmetry is deconstructed into separate copies, one for each family. A hierarchical chain of symmetry breaking provides an attractive low-scale solution to the flavor puzzle, consistent with flavor-changing neutral currents and finite naturalness. Although successful in explaining the origin of flavor hierarchies in the quark and charged lepton sectors, existing models have struggled with the large and seemingly anarchic mixing observed in neutrino oscillations. This letter identifies conditions under which neutrino anarchy may arise from flavor deconstruction in generic models with right-handed neutrinos. When deconstruction is applied to carefully chosen subgroups of the extended gauge symmetry, hierarchies in the seesaw formula cancel out.

Proactive Drive Failure Prediction for Cloud Storage System Through Semi-Supervised Learning

Proactive drive failure prediction can help operators handle the failing drives in advance, enhancing the storage system dependability. SSD and HDD failure prediction techniques are currently evolving towards a semi-supervised approach. In this paper, we are dedicated to enhancing the methodology for semi-supervised drive failure prediction from these aspects: design more powerful, robust yet generic models, mine drive failure modes and make the prediction model interpretable. Specifically, we propose two semi-supervised drive failure prediction models, DFP-VL and DFP-FL, from the perspective of reconstructing SMART data and learning SMART data's probability density, respectively. They capture the pattern of healthy drives, and failing drives can be detected when they deviate from the normal pattern. We mine two failure modes, slow deterioration and dramatic deterioration. Based on that, we design two failure detectors, ”ThresholdDetector” and ”HybridDetector”, to determine whether a drive deviates from the normal pattern. We evaluate the proposed methods on both SSD and HDD SMART data. DFP-VL is generic yet effective and can interpret the predicted results. DFP-FL has better performance than DFP-VL, but it cannot interpret the results. ThresholdDetector has low complexity and can detect most failing drives. HybridDetector has high complexity and can further improve the detection performance.

Windage loss characterisation for flywheel energy storage system: Model and experimental validation

In this paper, a windage loss characterisation strategy for Flywheel Energy Storage Systems (FESS) is presented. An effective windage loss modeling in FESS is essential for feasible and competitive design. Unlike generic aerodynamic loss models, FESS require particular attention to their unique characteristics i.e., vacuum, small airgaps, high angular speed, and presence of low friction rotor supports.The proposed model is based on several analytical and semi-empirical windage loss solutions for cylindrical and planar surface interactions, harmonised to manage the transition between laminar and turbulent flows. Also, the model is enriched by introducing corrections for rarefied gasses, using kinetic gas theory formulation. Therefore, it is possible the reduce the windage overestimation occurring with Navier–Stokes equation solutions for laminar flow. The model is compared to case studies from the literature featuring different boundary and operating conditions, to check consistency of all the harmonised models. A dedicated experimental test-rig is developed to validate the corrections to the model for rarefied gas at different vacuum levels. Then, a non-invasive characterisation approach for FESS windage losses in self-discharge phase is proposed and validated on the test-rig.

Bidirectional brain image translation using transfer learning from generic pre-trained models

Brain imaging plays a crucial role in the diagnosis and treatment of various neurological disorders, providing valuable insights into the structure and function of the brain. Techniques such as magnetic resonance imaging (MRI) and computed tomography (CT) enable non-invasive visualization of the brain, aiding in the understanding of brain anatomy, abnormalities, and functional connectivity. However, cost and radiation dose may limit the acquisition of specific image modalities, so medical image synthesis can be used to generate required medical images without actual addition. CycleGAN and other GANs are valuable tools for generating synthetic images across various fields. In the medical domain, where obtaining labeled medical images is labor-intensive and expensive, addressing data scarcity is a major challenge. Recent studies propose using transfer learning to overcome this issue. This involves adapting pre-trained CycleGAN models, initially trained on non-medical data, to generate realistic medical images. In this work, transfer learning was applied to the task of MR-CT image translation and vice versa using 18 pre-trained non-medical models, and the models were fine-tuned to have the best result. The models’ performance was evaluated using four widely used image quality metrics: Peak-signal-to-noise-ratio, Structural Similarity Index, Universal Quality Index, and Visual Information Fidelity. Quantitative evaluation and qualitative perceptual analysis by radiologists demonstrate the potential of transfer learning in medical imaging and the effectiveness of the generic pre-trained model. The results provide compelling evidence of the model’s exceptional performance, which can be attributed to the high quality and similarity of the training images to actual human brain images. These results underscore the significance of carefully selecting appropriate and representative training images to optimize performance in brain image analysis tasks.

Energy-dependent boosted dark matter from diffuse supernova neutrino background

Diffuse neutrinos from past supernovae in the Universe present us with a unique opportunity to test dark matter (DM) interactions. These neutrinos can scatter and boost the DM particles in the Milky Way halo to relativistic energies allowing us to detect them in terrestrial laboratories.Focusing on generic models of DM-neutrino and electron interactions, mediated by a vector or a scalar boson, we implement energy-dependent scattering cross-sections and perform detailed numerical analysis of DM attenuation due to electron scattering in-medium while propagating towards terrestrial experiments. We set new limits on DM-neutrino and electron interactions for DM with masses in the range ∼ (0.1, 104) MeV, using recent data from XENONnT, LUX-ZEPLIN, and PandaX-4T direct detection experiments. We demonstrate that consideration of energy-dependent cross-sections for DM interactions can significantly affect constraints previously derived under the assumption of constant cross-sections, modifying them by multiple orders of magnitude.

Model-independent estimates for loop-induced baryon-number-violating nucleon decays

Baryon number is an accidental symmetry of the Standard Model (SM) Lagrangian that so far has been measured to be exactly preserved, although it is expected to be violated at higher energies. In this work we compute order-of-magnitude estimates for the matching contributions of generic ultraviolet models to effective operators that generate nucleon decay processes. This is done in a systematic and automated way using operators constructed from SM fields up to dimension nine and working in a framework that has proved useful in the study of lepton-number violation. For each of the operators we derive estimates for the rates of different nucleon-decay channels. These allow us to establish model-independent lower bounds on the underlying new-physics scale and identify potential correlations between the various decay modes. The results are most relevant for families of models that generate the considered operator. This analysis is especially timely given the expected future sensitivities in numerous experiments such as Hyper-K, DUNE, JUNO and THEIA.

RectanglingGAN: Deep rectangling for stitched image via image inpainting

The deep rectangling task aims to transform edge-irregular stitched images into standardised rectangular formats using deep learning. Existing deep rectangling solutions rely on mesh-based segmentation and deformation, which introduce global distortions detrimental to subsequent analysis and processing. Therefore, we propose RectanglingGAN to handle missing regions in a distortion-free manner via image inpainting. Unlike generic inpainting models, RectanglingGAN focuses on preserving the quality of edge regions. In particular, we present an adaptive distance transform module to improve the attention of the feature maps to the mask boundaries. The module dynamically selects the appropriate distance transformation scheme according to the image content. Moreover, we introduce the distance-weighted decaying reconstruction loss into the pixel-wise reconstruction, which introduces the spatial location information between the inpainting pixels and the mask boundaries by inverting the distance-transformed mask. By incorporating this loss, the model pays more attention to the quality of generation pixels far from the mask boundary. Extensive experiments validate the effectiveness of RectanglingGAN compared to state-of-the-art methods, highlighting its significant advantages in terms of both the quality and the fidelity of the edge region in generation images.

Evaluating the Repeatability of Musculoskeletal Modelling Force Outcomes in Gait among Chronic Stroke Survivors: Implications for Contemporary Clinical Practice

This study aims to evaluate the consistency of musculoskeletal modelling outcomes during walking in chronic post-stroke patients, focusing on both affected and unaffected sides. Understanding the specific muscle forces involved is crucial for designing targeted rehabilitation strategies to improve balance and mobility after a stroke. Musculoskeletal modelling provides valuable insights into muscle and joint loading, aiding clinicians in analysing essential biomarkers and enhancing patients’ functional outcomes. However, the repeatability of these modelling outcomes in stroke gait has not been thoroughly explored until now. Twelve post-stroke, hemiparetic survivors were included in the study, which consisted of a gait analysis protocol to capture kinematic and kinetic variables. Two generic full body MSK models—Hamner (Ham) and Rajagopal (Raj)—were used to compute joint angles and muscle forces during walking, with combinations of two muscle force estimation algorithms (Static Optimisation (SO) and Computed Muscle Control (CMC)) and different joint degrees-of-freedoms (DOF). The multiple correlation coefficient (MCCoef) was used to compute repeatability for all forces, grouped based on anatomical function. Regardless of models and DOFs, the mean minimum (0.75) and maximum (0.94) MCCoefs denote moderate-to-excellent repeatability for all muscle groups. The combination of the Ham model and SO provided the most repeatable muscle force estimations of all the muscle groups except for the hip flexors, adductors and internal rotators. DOF configuration did not generally affect muscle force repeatability in the Ham–SO case, although the 311 seemed to relate to the highest values. Lastly, the DOF setting had a significant effect on some muscle groups’ force output, with the highest magnitudes reported for the 321 and 322 of non-paretic and paretic hip adductors and extensors, knee flexors and ankle dorsiflexors and paretic knee flexors. The primary findings of our study can assist users in selecting the most suitable modelling workflow and encourage the widespread adoption of MSK modelling in clinical practice.

A Hybrid Experimental Modelling Approach to Solar Photovoltaic Cell Temperature Prediction

Solar cell temperature is critical in the determination of solar energy generated by a solar photovoltaic power plant. High temperatures are associated with a reduction in the energy generated and hence prediction of photovoltaic cell temperature is essential in temperature mitigation and solar energy forecasting especially in commercial power plants. The present study focused on the development of a machine learning based predictive model for solar photovoltaic cell temperature prediction in commercial solar photovoltaic power plants. A physical experimental set up was developed to measure solar cell temperature under different weather and other related parameters. Satellite data were also collated for those parameters difficult to measure experimentally and were used to compliment experimental data used in this study. Satellite data used in the study were statistically transformed for each parameter used to mimic experimentally measured data. Statistical approaches were adopted to analyse the influence of both the dependent and independent variables on solar cell temperature. The analysis included multicollinearity and correlation analysis with the aid of heat maps meant to establish the relationships among the independent and dependent variables. Feature selection and dimensionality reduction was also performed to reduce the input variables and maintain relevant data in the modelling process. Parameters with a strong correlation to each other had some of them eliminated in the modelling process. A solar cell temperature predictive model based on selected weather parameters was developed using a machine learning approach (Random Forests), and parameters used were selected from the statistical analysis. The prediction accuracy of the developed model was analysed using the coefficient of determination (R2) and the mean absolute percentage error (MAPE). The results indicated a higher model performance compared to generic models used in cell temperature prediction. The prediction MAPE for the developed model was 0.83 °C while an R2 value of 0.93 was obtained which was indicative of a good model. The developed model was also comparable to other contemporary models developed to predict solar photovoltaic cell temperature. Simulations were also done to determine the annual energy generated with the incorporation of the solar cell temperature prediction model. The results revealed a 3.4% difference in the annual energy generated between a simulation which considered solar cell temperature and that which ignored the solar cell temperature. The study also revealed that this difference is even larger in monetary terms when lifetime energy generation is considered. The present study demonstrated that Random Forests, a machine learning approach to predictive modelling, can handle complex models and can provide models with a higher accuracy compared to statistical modelling approaches. The study recommends the use of solar cell predictive models to improve the accuracy of energy prediction on solar photovoltaic power plants which in turn assists in energy planning and deployment.

A multifluid model with chemically reacting components — Construction of weak solutions

We investigate the existence of weak solutions to a multi-component system, consisting of compressible chemically reacting components, coupled with the compressible Stokes equation for the velocity. Specifically, we consider the case of irreversible chemical reactions and assume a nonlinear relation between the pressure and the particular densities. These assumptions cause the additional difficulties in the mathematical analysis, due to the possible presence of vacuum.It is shown that there exists a global weak solution, satisfying the L∞ bounds for all the components. We obtain strong compactness of the sequence of densities in Lp spaces, under the assumption that all components are strictly positive. The applied method captures the properties of models of high generality, which admit an arbitrary number of components. Furthermore, the framework that we develop can handle models that contain both diffusing and non-diffusing elements.

The new physics case for beam-dump experiments with accelerated muon beams

As the field examines a future muon collider as a possible successor to the LHC, we must consider how to fully utilize not only the high-energy particle collisions, but also any lower-energy staging facilities necessary in the R&D process. An economical and efficient possibility is to use the accelerated muon beam from either the full experiment or from cooling and acceleration tests in beam-dump experiments. Beam-dump experiments are complementary to the main collider as they achieve sensitivity to very small couplings with minimal instrumentation. We demonstrate the utility of muon beam-dump experiments for new physics searches at energies from 10 GeV to 5 TeV. We find that, even at low energies like those accessible at staging or demonstrator facilities, it is possible to probe new regions of parameter space for a variety of generic BSM models, including muonphilic, leptophilic, Lμ − Lτ, and dark photon scenarios. Such experiments could therefore provide opportunities for discovery of new physics well before the completion of the full multi-TeV collider.

Aerodynamic flow field analysis of vortex interactions on multi-swept wing configurations

The underlying flow physics of vortex breakdown and vortex-vortex interactions at moderate to high angles of attack is not well understood. This lack of understanding is a challenge to the development of optimized wing designs for fighter aircraft. The goal of this study is to gain insight into the breakdown of vortices on multi-swept wing configurations. Two generic models were considered, one with a gradual vortex breakdown and one with a rapid (abrupt) breakdown. Aerodynamic forces were calculated from the US Air Force Academy (USAFA) subsonic wind tunnel to validate the experimental setup using computational fluid dynamics (CFD) and previous work. Stereoscopic particle imaging velocimetry (SPIV) was used to visualize vortex-vortex interactions, measure vortex strengths, and identify vortex breakdown locations. The measured lift and drag forces showed flow physics that agreed with previous experiments and CFD, supporting the validity of the new experimental setup. Analysis of SPIV and CFD data improved our understanding of the different types of vortex interactions observed for both models. The only difference between the models was the location of the secondary sweep angle on the chord. The C4L80 model, with a secondary sweep angle at 40% of the root chord, had a large shear layer along the leading edge, which led to vortex merging and gradual vortex breakdown at lower angles of attack than its counterpart. The C6L80 model, with a secondary sweep angle at 60% of the root chord, had no such shear layer and instead developed strong, isolated vortices that were characterized by vortex braiding and abrupt breakdown at high angles of attack. These vortices were entrained into the main vortex at the leading edge of the strake (inboard vortex), creating a balanced vortex system that enhanced lift and delayed the onset of gradual vortex breakdown seen in the C4L80 model.

Temperature compensation through kinetic regulation in biochemical oscillators

Nearly all circadian clocks maintain a period that is insensitive to temperature changes, a phenomenon known as temperature compensation (TC). Yet, it is unclear whether there is any common feature among different systems that exhibit TC. From a general timescale invariance, we show that TC relies on the existence of certain period-lengthening reactions wherein the period of the system increases strongly with the rates in these reactions. By studying several generic oscillator models, we show that this counterintuitive dependence is nonetheless a common feature of oscillators in the nonlinear (far-from-onset) regime where the oscillation can be separated into fast and slow phases. The increase of the period with the period-lengthening reaction rates occurs when the amplitude of the slow phase in the oscillation increases with these rates while the progression speed in the slow phase is controlled by other rates of the system. The positive dependence of the period on the period-lengthening rates balances its inverse dependence on other kinetic rates in the system, which gives rise to robust TC in a wide range of parameters. We demonstrate the existence of such period-lengthening reactions and their relevance for TC in all four model systems we considered. Theoretical results for a model of the Kai system are supported by experimental data. A study of the energy dissipation also shows that better TC performance requires higher energy consumption. Our study unveils a general mechanism by which a biochemical oscillator achieves TC by operating in parameter regimes far from the onset where period-lengthening reactions exist.

ProkDBP: Toward more precise identification of prokaryotic DNA binding proteins.

Prokaryotic DNA binding proteins (DBPs) play pivotal roles in governing gene regulation, DNA replication, and various cellular functions. Accurate computational models for predicting prokaryotic DBPs hold immense promise in accelerating the discovery of novel proteins, fostering a deeper understanding of prokaryotic biology, and facilitating the development of therapeutics targeting for potential disease interventions. However, existing generic prediction models often exhibit lower accuracy in predicting prokaryotic DBPs. To address this gap, we introduce ProkDBP, a novel machine learning-driven computational model for prediction of prokaryotic DBPs. For prediction, a total of nine shallow learning algorithms and five deep learning models were utilized, with the shallow learning models demonstrating higher performance metrics compared to their deep learning counterparts. The light gradient boosting machine (LGBM), coupled with evolutionarily significant features selected via random forest variable importance measure (RF-VIM) yielded the highest five-fold cross-validation accuracy. The model achieved the highest auROC (0.9534) and auPRC (0.9575) among the 14 machine learning models evaluated. Additionally, ProkDBP demonstrated substantial performance with an independent dataset, exhibiting higher values of auROC (0.9332) and auPRC (0.9371). Notably, when benchmarked against several cutting-edge existing models, ProkDBP showcased superior predictive accuracy. Furthermore, to promote accessibility and usability, ProkDBP (https://iasri-sg.icar.gov.in/prokdbp/) is available as an online prediction tool, enabling free access to interested users. This tool stands as a significant contribution, enhancing the repertoire of resources for accurate and efficient prediction of prokaryotic DBPs.

Deformation and Plasticity of Reinforced Concrete

Statement of the problem.Development of the model of plasticity of reinforced concrete structures is based on the intensity of the "stress-strain" relation to project tensors using transitions, as well as the coefficients of the steps of forces and axes (principal angle and strain-strain, total shear strain, etc.). Results. The dependences of plasticity modulus of concrete, transverse coefficient have been obtained. Complex functions of generalized hypotheses of deplanation of linear and angular deformations and jumps for crack formation breakaway and stiffness are constructed. It includes force flows, i.e., blocks of compressed and tensile concrete for isotropic medium (the first object), "main cracks" from fracture mechanics, functional and two-cone element for the deformation effect of reinforced concrete. The resistance of the tensile concrete is transferred to the working reinforcement using the total average longitudinal and transverse forces for the transversal-isotropic environment, including the average modulus of reinforcement and the reduced tensile concrete coefficient of the second limit state group. The special dowel effect is derived from the second level model of structural mechanics for a rebar with two pinched ends, as well as crack opening (a development of the Thomas –– author hypothesis), crack-track bank shear, and buckling. The main vector of forces in the reinforcement is characterized by two values of longitudinal and transverse displacements and the width of opening and shear of the crack-track (the third object). Conclusions. The energetic theory on the surface of the sphere and the determination of the integral of the mean square value of the tangential stresses of plasticity theory allowed to develop an alternative to the ge-neral model in the form of a paraboloid by summing the volume sectors, radius-levels from the matrix of sliding planes (including octahedral and pure shear) in concrete medium from microcracks to macrocracks.

Predicting mixed venous oxygen saturation (SvO2) impairment in COPD patients using clinical-CT radiomics data: A preliminary study.

Chronic obstructive pulmonary disease (COPD) is one of the most common chronic airway diseases in the world. To predict the degree of mixed venous oxygen saturation (SvO2) impairment in patients with COPD by modeling using clinical-CT radiomics data and to provide reference for clinical decision-making. A total of 236 patients with COPD diagnosed by CT and clinical data at Xiangyang No. 1 People's Hospital (n= 157) and Xiangyang Central Hospital (n= 79) from June 2018 to September 2021 were retrospectively analyzed. The patients were divided into group A (SvO⩾2 62%, N= 107) and group B (SvO<2 62%, N= 129). We set up training set and test set at a ratio of 7/3 and time cutoff spot; In training set, Logistic regression was conducted to analyze the differences in general data (e.g. height, weight, systolic blood pressure), laboratory indicators (e.g. arterial oxygen saturation and pulmonary artery systolic pressure), and CT radiomics (radscore generated using chest CT texture parameters from 3D slicer software and LASSO regression) between these two groups. Further the risk factors screened by the above method were used to establish models for predicting the degree of hypoxia in COPD, conduct verification in test set and create a nomogram. Univariate analysis demonstrated that age, smoking history, drinking history, systemic systolic pressure, digestive symptoms, right ventricular diameter (RV), mean systolic pulmonary artery pressure (sPAP), cardiac index (CI), pulmonary vascular resistance (PVR), 6-min walking distance (6MWD), WHO functional classification of pulmonary hypertension (WHOPHFC), the ratio of forced expiratory volume in the first second to the forced vital capacity (FEV1%), and radscore in group B were all significantly different from those in group A (P< 0.05). Multivariate regression demonstrated that age, smoking history, digestive symptoms, 6MWD, and radscore were independent risk factors for SvO2 impairment. The combined model established based on the abovementioned indicators exhibited a good prediction effect [AUC: 0.903; 95%CI (0.858-0.937)], higher than the general clinical model [AUC: 0.760; 95%CI (0.701-0.813), P< 0.05] and laboratory examination-radiomics model [AUC: 0.868; 95%CI (0.818-0.908), P= 0.012]. The newly created nomogram may be helpful for clinical decision-making and benefit COPD patients. SvO2 is an important indicator of hypoxia in COPD, and it is highly related to age, 6MWD, and radscore. The combined model is helpful for early identification of SvO2 impairment and adjustment of COPD treatment strategies.

Deep Class-Incremental Learning From Decentralized Data.

In this article, we focus on a new and challenging decentralized machine learning paradigm in which there are continuous inflows of data to be addressed and the data are stored in multiple repositories. We initiate the study of data-decentralized class-incremental learning (DCIL) by making the following contributions. First, we formulate the DCIL problem and develop the experimental protocol. Second, we introduce a paradigm to create a basic decentralized counterpart of typical (centralized) CIL approaches, and as a result, establish a benchmark for the DCIL study. Third, we further propose a decentralized composite knowledge incremental distillation (DCID) framework to transfer knowledge from historical models and multiple local sites to the general model continually. DCID consists of three main components, namely, local CIL, collaborated knowledge distillation (KD) among local models, and aggregated KD from local models to the general one. We comprehensively investigate our DCID framework by using a different implementation of the three components. Extensive experimental results demonstrate the effectiveness of our DCID framework. The source code of the baseline methods and the proposed DCIL is available at https://github.com/Vision-Intelligence-and-Robots-Group/DCIL.