improve model performance
Recently Published Documents

Federated Learning (FL) is a setting that allows clients to train a joint global model collaboratively while keeping data locally. Due to FL has advantages of data confidential and distributed computing, interest in this area has increased. In this paper, we designed a new FL algorithm named FedRAD. Random communication and dynamic aggregation methods are proposed for FedRAD. Random communication method enables FL system use the combination of fixed communication interval and constrained variable intervals in a single task. Dynamic aggregation method reforms aggregation weights and makes weights update automately. Both methods aim to improve model performance. We evaluated two proposed methods respectively, and compared FedRAD with three algorithms on three hyperparameters. Results at CIFAR-10 demonstrate that each method has good performance, and FedRAD can achieve higher classification accuracy than state-of-the-art FL algorithms.

To improve model performance and study climate change impacts across physical, biological, and social systems, model intercomparison projects (MIPs) are regularly conducted. MIPs represent a crucial tool for undergraduate researchers to meaningfully contribute to climate change research.

The effectiveness of a machine learning model is impacted by the data representation used. Consequently, it is crucial to investigate robust representations for efficient machine learning methods. In this paper, we explore the link between data representations and model performance for inference tasks on spatial networks. We argue that representations which explicitly encode the relations between spatial entities would improve model performance. Specifically, we consider homogeneous and heterogeneous representations of spatial networks. We recognise that the expressive nature of the heterogeneous representation may benefit spatial networks and could improve model performance on certain tasks. Thus, we carry out an empirical study using Graph Neural Network models for two inference tasks on spatial networks. Our results demonstrate that heterogeneous representations improves model performance for down-stream inference tasks on spatial networks.

Leaf age and plant centre are important phenotypic information of weeds, and accurate identification of them plays an important role in understanding the morphological structure of weeds, guiding precise targeted spraying and reducing the use of herbicides. In this work, a weed segmentation method based on BlendMask is proposed to obtain the phenotypic information of weeds under complex field conditions. This study collected images from different angles (front, side, and top views) of three kinds of weeds (Solanum nigrum, barnyard grass (Echinochloa crus-galli), and Abutilon theophrasti Medicus) in a maize field. Two datasets (with and without data enhancement) and two backbone networks (ResNet50 and ResNet101) were replaced to improve model performance. Finally, seven evaluation indicators are used to evaluate the segmentation results of the model under different angles. The results indicated that data enhancement and ResNet101 as the backbone network could enhance the model performance. The F1 value of the plant centre is 0.9330, and the recognition accuracy of leaf age can reach 0.957. The mIOU value of the top view is 0.642. Therefore, deep learning methods can effectively identify weed leaf age and plant centre, which is of great significance for variable spraying.

The innovative “three comprehensive education” model can mobilize the initiatives of students in higher vocational colleges, whose book borrowing behaviors are affected by “three comprehensive education”. This paper mainly studies the psychological motivations of students for book borrowing in libraries of higher vocational colleges from the perspective of “three comprehensive education”. Firstly, the psychological motivations for book borrowing were analyzed in details under the mechanism of “three comprehensive education”. Then, a dataset was created from the mass data on book borrowing and students. The relevant features were compressed as model inputs, aiming to improve model performance and reduce storage cost. Finally, a perception model was constructed to analyze the psychological motivations of students for book borrowing, and used to solve the five-class classification problem based on the directed acyclic graph. Experimental results prove the scientific of the psychological pattern analysis and the effectiveness of the proposed model. The research results provide a theoretical and practical reference for the effect analysis of “three comprehensive education”.

AbstractObjectivesExisting UK prognostic models for patients admitted to hospital with COVID-19 are limited by reliance on comorbidities, which are under-recorded in secondary care, and lack of imaging data among the candidate predictors. Our aims were to develop and externally validate novel prognostic models for adverse outcomes (death, intensive therapy unit (ITU) admission) in UK secondary care; and externally validate the existing 4C score.DesignCandidate predictors included demographic variables, symptoms, physiological measures, imaging, laboratory tests. Final models used logistic regression with stepwise selection.SettingModel development was performed in data from University Hospitals Birmingham (UHB). External validation was performed in the CovidCollab dataset.ParticipantsPatients with COVID-19 admitted to UHB January-August 2020 were included.Main outcome measuresDeath and ITU admission within 28 days of admission.Results1040 patients with COVID-19 were included in the derivation cohort; 288 (28%) died and 183 (18%) were admitted to ITU within 28 days of admission. Area under the receiver operating curve (AUROC) for mortality was 0.791 (95%CI 0.761-0.822) in UHB and 0.767 (95%CI 0.754-0.780) in CovidCollab; AUROC for ITU admission was 0.906 (95%CI 0.883-0.929) in UHB and 0.811 (95%CI 0.795-0.828) in CovidCollab. Models showed good calibration. Addition of comorbidities to candidate predictors did not improve model performance. AUROC for the 4C score in the UHB dataset was 0.754 (95%CI 0.721-0.786).ConclusionsThe novel prognostic models showed good discrimination and calibration in derivation and external validation datasets, and outperformed the existing 4C score. The models can be integrated into electronic medical records systems to calculate each individual patient’s probability of death or ITU admission at the time of hospital admission. Implementation of the models and clinical utility should be evaluated.Article SummaryStrengths and limitations of this studyWe developed novel prognostic models predicting mortality and ITU admission within 28 days of admission for patients hospitalised with COVID-19, using a large routinely collected dataset gathered at admission with a wide range of possible predictors (demographic variables, symptoms, physiological measures, imaging, laboratory test results).These novel models showed good discrimination and calibration in both derivation and external validation cohorts, and outperformed the existing ISARIC model and 4C score in the derivation dataset. We found that addition of comorbidities to the set of candidate predictors included in model derivation did not improve model performance.If integrated into hospital electronic medical records systems, the model algorithms will provide a predicted probability of mortality or ITU admission for each patient based on their individual data at, or close to, the time of admission, which will support clinicians’ decision making with regard to appropriate patient care pathways and triage. This information might also assist clinicians in explaining complex prognostic assessments and decisions to patients and their relatives.A limitation of the study was that in the external validation cohort we were unable to examine all of the predictors included in the original full UHB model due to only a reduced set of candidate predictors being available in CovidCollab. Nevertheless, the reduced model performed well and the results suggest it may be applicable in a wide range of datasets where only a reduced set of predictor variables is available.Furthermore, it was not possible to carry out stratified analysis by ethnicity as the UHB dataset contained too few patients in most of the strata, and no ethnicity data was available in the CovidCollab dataset.

AbstractWe previously developed an African-ancestry-specific polygenic hazard score (PHS46+African) that substantially improved prostate cancer risk stratification in men with African ancestry. The model consists of 46 SNPs identified in Europeans and 3 SNPs from 8q24 shown to improve model performance in Africans. Herein, we used principal component (PC) analysis to uncover subpopulations of men with African ancestry for whom the utility of PHS46+African may differ. Genotypic data were obtained from PRACTICAL consortium for 6,253 men with African genetic ancestry. Genetic variation in a window spanning 3 African-specific 8q24 SNPs was estimated using 93 PCs. A Cox proportional hazards framework was used to identify the pair of PCs most strongly associated with performance of PHS46+African. A calibration factor (CF) was formulated using estimated Cox coefficients to quantify the extent to which the performance of PHS46+African varies with PC. CF of PHS46+African was strongly associated with the first and twentieth PCs. Predicted CF ranged from 0.41 to 2.94, suggesting that PHS46+African may be up to 7 times more beneficial to some African men than others. The explained relative risk for PHS46+African varied from 3.6% to 9.9% for individuals with low and high CF values, respectively. By cross-referencing our dataset with 1000 Genomes, we identified statistically significant associations between continental and calibration groupings. In conclusion, we identified PCs within 8q24 SNP window that were strongly associated with performance of PHS46+African. Further research to improve clinical utility of polygenic risk scores (or models) is needed to improve health outcomes for men of African ancestry

The Sebastian Inlet Florida Coastal Processes Model computes sediment transport pathways in the nearshore to support sediment management activities. Longshore sediment transport rates are computed by the model and compared with field data. The model is run with two alternative specifications of hard bottom to investigate the impact on computed transport rates. One alternative specifies known hard bottom outcrop locations and the second, a uniform one-meter overburden throughout the model domain. The uniform overburden specification improved longshore sediment transport rate computations throughout the model domain. The goal of this work is to improve upon nearshore sediment transport and morphology by addressing uncertainty in hard bottom locations and ephemeral coverage. This paper documents the modeling effort and the changes necessary to improve model performance in the nearshore.Recorded Presentation from the vICCE (YouTube Link): https://youtu.be/u1bNOca5qUo