Model Predictions Research Articles

Disentangling Neurodegeneration From Aging in Multiple Sclerosis Using Deep Learning: The Brain-Predicted Disease Duration Gap.

Disentangling brain aging from disease-related neurodegeneration in patients with multiple sclerosis (PwMS) is increasingly topical. The brain-age paradigm offers a window into this problem but may miss disease-specific effects. In this study, we investigated whether a disease-specific model might complement the brain-age gap (BAG) by capturing aspects unique to MS. In this retrospective study, we collected 3D T1-weighted brain MRI scans of PwMS to build (1) a cross-sectional multicentric cohort for age and disease duration (DD) modeling and (2) a longitudinal single-center cohort of patients with early MS as a clinical use case. We trained and evaluated a 3D DenseNet architecture to predict DD from minimally preprocessed images while age predictions were obtained with the DeepBrainNet model. The brain-predicted DD gap (the difference between predicted and actual duration) was proposed as a DD-adjusted global measure of MS-specific brain damage. Model predictions were scrutinized to assess the influence of lesions and brain volumes while the DD gap was biologically and clinically validated within a linear model framework assessing its relationship with BAG and physical disability measured with the Expanded Disability Status Scale (EDSS). We gathered MRI scans of 4,392 PwMS (69.7% female, age: 42.8 ± 10.6 years, DD: 11.4 ± 9.3 years) from 15 centers while the early MS cohort included 749 sessions from 252 patients (64.7% female, age: 34.5 ± 8.3 years, DD: 0.7 ± 1.2 years). Our model predicted DD better than chance (mean absolute error = 5.63 years, R2 = 0.34) and was nearly orthogonal to the brain-age model (correlation between DD and BAGs: r = 0.06 [0.00-0.13], p = 0.07). Predictions were influenced by distributed variations in brain volume and, unlike brain-predicted age, were sensitive to MS lesions (difference between unfilled and filled scans: 0.55 years [0.51-0.59], p < 0.001). DD gap significantly explained EDSS changes (B = 0.060 [0.038-0.082], p < 0.001), adding to BAG (ΔR2 = 0.012, p < 0.001). Longitudinally, increasing DD gap was associated with greater annualized EDSS change (r = 0.50 [0.39-0.60], p < 0.001), with an incremental contribution in explaining disability worsening compared with changes in BAG alone (ΔR2 = 0.064, p < 0.001). The brain-predicted DD gap is sensitive to MS-related lesions and brain atrophy, adds to the brain-age paradigm in explaining physical disability both cross-sectionally and longitudinally, and may be used as an MS-specific biomarker of disease severity and progression.

Abstract A026: Assessing the impact of obesity on postoperative outcomes in RCC using machine learning

Abstract Introduction Obesity is associated with an increased risk of developing renal cell carcinoma (RCC) but paradoxically correlates with improved outcomes in metastatic cases. Moreover, the data on the relationship between obesity and postoperative outcomes following nephrectomy are inconsistent. Accurate prediction of postoperative outcomes, including complications, 30-day readmissions, and mortality, is essential for improving patient outcomes in surgical procedures. In this study, we utilized ML models to predict such outcomes in RCC patients undergoing nephroureterectomy, radical nephrectomy, partial nephrectomy, and other excision procedures on the kidney using data from the National Surgical Quality Improvement Program (NSQIP; 2016- 2021). Methods A gradient-boosted tree (GBT) ML model was developed and trained to predict the primary outcomes of interest: major complications, minor complications, 30-day readmission, and mortality. Patients with a BMI ≥30 kg/m2 were considered obese, and patients with a BMI &amp;lt;30 kg/m2 were used as controls. The model's performance was rigorously evaluated using various measures, such as AUROC, generalized R-square, and misclassification rate. The model's predictions were validated using separate training (70%) and validation (30%) cohorts to ensure generalizability and applicability in diverse patient populations. Results In the analysis, 36,284 cases were included. Bivariate analysis revealed that obesity was associated with an increased rate of minor complications (5.48% vs. 4.41%, p&amp;lt;0.0001) compared to non-obese individuals with similar major complications (3.33 vs. 3.21, p=0.49), mortality (0.54 vs. 0.64, p=0.22) and 30-day readmission rates (5.74 vs. 5.83, p=0.70). The GBT model demonstrated substantial predictive power across all four outcomes. In predicting mortality, the model achieved an AUROC of 0.80 in the validation set, with a misclassification rate of 0.68%. The prediction of major complications was also robust, with an AUROC of 0.91 in the validation set and a misclassification rate of 2.2%. Similarly, the model's performance in predicting minor complications and 30-day readmissions was strong, with AUROC values of 0.70 and 0.66, respectively, in the validation sets. Conclusions Obesity was linked to a higher rate of minor complications but did not affect other postoperative outcomes in RCC patients undergoing nephrectomy. By integrating a wide range of patient-specific variables, ML models are powerful tools for clinicians to identify high-risk patients and tailor perioperative care accordingly. The interaction between RCC and the surrounding adipose tissue microenvironment may have clinical relevance, warranting further research to elucidate the role of BMI in postoperative complications. Our findings advocate for the expanded use of ML techniques in surgical decision-making, which could enhance patient safety, optimize resource use, and improve overall surgical outcomes. Citation Format: Atulya A. Khosla, Manas Pustake, Sufal Chhabra, Yanjia Zhang, Mukesh Roy, Muni Rubens, Venkataraghavan Ramamoorthy, Anshul Saxena, Ishmael A. Jaiyesimi. Assessing the impact of obesity on postoperative outcomes in RCC using machine learning [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Tumor-body Interactions: The Roles of Micro- and Macroenvironment in Cancer; 2024 Nov 17-20; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2024;84(22_Suppl):Abstract nr A026.

Automating Abercrombie: Machine‐learning trademark distinctiveness

AbstractTrademark law protects marks to enable firms to signal their products' qualities to consumers. To qualify for protection, a mark must be able to identify and distinguish goods. US courts typically locate a mark on a “spectrum of distinctiveness”—known as the Abercrombie spectrum—that categorizes marks as fanciful, arbitrary, or suggestive, and thus as “inherently distinctive,” or as descriptive or generic, and thus as not inherently distinctive. This article explores whether locating trademarks on the Abercrombie spectrum can be automated using current natural‐language processing techniques. Using about 1.5 million US trademark registrations between 2012 and 2019 as well as 2.2 million related USPTO office actions, the article presents a machine‐learning model that learns semantic features of trademark applications and predicts whether a mark is inherently distinctive. Our model can predict trademark actions with 86% accuracy overall, and it can identify subsets of trademark applications where it is highly certain in its predictions of distinctiveness. Using an eXplainable AI (XAI) algorithm, we further analyze which features in trademark applications drive our model's predictions. We then explore the practical and normative implications of our approach. On a practical level, we outline a decision‐support system that could, as a “robot trademark clerk,” assist trademark experts in their determination of a trademark's distinctiveness. Such a system could also help trademark experts understand which features of a trademark application contribute the most toward a trademark's distinctiveness. On a theoretical level, we discuss the normative limits of the Abercrombie spectrum and propose to move beyond Abercrombie for trademarks whose distinctiveness is uncertain. We discuss how machine‐learning projects in the law not only inform us about the aspects of the legal system that may be automated in the future, but also force us to tackle normative tradeoffs that may be invisible otherwise.

Effects of Red Vinasse on Physicochemical Qualities of Blue Round Scad (Decapterus maruadsi) During Storage, and Shelf Life Prediction

A fish processed with red vinasse is a type of Fujian cuisine with regional characteristics. In order to monitor the effect of red vinasse on storage quality and shelf life of blue round scad (Decapterus maruadsi) during storage, the changes in fat content, thiobarbituric acid reactive substances (TBARS), composition of polyunsaturated fatty acids (PUFAs), pH value, texture, and sensory quality were studied at different storage temperatures (4 °C, 25 °C, and 37 °C). By analyzing the correlation between changes in sensory qualities and physical and chemical indexes, a first-order kinetic model and the Arrhenius equation were used to build a shelf-life prediction model for blue round scad during storage. The results showed that processing with red vinasse can significantly reduce the malondialdehyde (MDA) production and the decrease in PUFAs, eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA) (p &lt; 0.05) during storage. Based on partial least squares regression (PLSR), the storage temperature and time have a significant impact on the PUFA composition in blue round scad, which changed less when the samples were stored at 4 °C and 25 °C, and they had better nutritional composition of fatty acids at lower temperatures. Among the PUFAs, DHA (C22:6n-3) and EPA (C20:5n-3) had higher relative contents and significantly decreased during storage (p &lt; 0.05). Additionally, the processing with red vinasse can slow down the increase in total volatile basic nitrogen (TVB-N) value and pH of blue round scad, maintain the appropriate hardness, elasticity, cohesion and chewability, and improve the overall sensory quality of the fish. In addition, according to the results of model prediction based on TBARS value, the storage shelf life of blue round scad with red vinasse added was 55 d, 2.7 d and 28 h at 4 °C, 25 °C and 37 °C, respectively. The accuracy of the forecast model was high, and the relative errors of the measured values and predicted values were less than 10%. Thus, it not only provided a theoretical basis for the processing and application of red vinasse to Chinese traditional food, but also provided innovative ideas for the safe storage and high-value utilization of blue round scad.

Ensemble Transfer Learning for Image Classification

The deep learning (DL) techniques used for image classification might not deliver the desired level of classification accuracy as some features belonging to some class of a dataset are missed during feature extraction. The ensemble learning (EL) based model improves classification accuracy by combining the strengths of individual classifiers. As a result, those features that were missed during feature extraction by a specific DL technique will be taken care of by another DL technique in an ensemble DL approach. In this paper, averaging EL (AENet), weighted averaging EL (WAENet), and stacking EL (StackedNet) approaches are proposed, considering the DenseNet201, EcientNetB0, and ResNetRS101 as base models. The predictions of the base models are averaged to generate the AENet. The WAENet is constructed by assigning weights to each base model based on their prediction and then taking their average. Similarly, the Stacked-Net is developed by considering the DenseNet201, EcientNetB0, and ResNetRS101 as base-learners and ResNetRS101 as meta-learner. Analysed performance of the considered pre-trained base models and the developed EL models on the standard and application-specific datasets such as MiniImageNet, CIFAR10, CIFAR100, Plant Village (PV), Tomato, Covid-19 and 9IndianFood. 80% of the datasets were used to train and 20% to test the base and proposed models. The models are trained for an epoch of 30, considering a learning rate of 0.001 and adam optimizer. The stackedNet delivered better results than others.

Time-Efficacy Relationship of Semaglutide in the Treatment of Type 2 Diabetes Mellitus: A Model-Based Meta-Analysis.

Our objective was to quantify the efficacy of subcutaneous once-weekly semaglutide in treating type 2 diabetes mellitus (T2DM) over time. Based on a literature search of the PubMed, Embase, Cochrane, and Web of Science databases, a modified maximum effect (Emax) model including rebound effects was built using model-based meta-analysis with change from baseline in glycated hemoglobin as the efficacy endpoint. This was combined with the covariate model to form a final model, and then theoretical values of Emax and time to reach 50% of Emax (ET50) were obtained for each dose. Model fit and prediction were assessed using goodness-of-fit plots and visual prediction checking. Emax and ET50 were influenced by the proportion of males and the baseline values, respectively. There was no evidence of a placebo effect with semaglutide. The efficacy of other doses became more significant over time, and a rebound effect was observed after maximum efficacy, at a rate of 0.018. Simulation of the typical efficacy at the different doses yielded a maximum efficacy of -1.58% with 0.5 mg and a maximum efficacy of -1.87% with 1 mg. In addition, the six simulated doses (0, 0.1, 0.2, 0.5, 1, and 2 mg) showed a dose-dependent relationship between dose and efficacy except for 0.4 mg and 0.8 mg. A higher dose would result in greater efficacy and a faster onset of action. The efficacy of semaglutide in glucose control was investigated using the model-based meta-analysis method, which yields new insights into the treatment of T2DM with semaglutide.

Optimising bearing degradation prediction: a bayesian vector autoregressive-assisted DeepAR model approach

Abstract Accurate prediction of the remaining useful life of bearings is crucial for the maintenance of rotating machinery. Current research often involves extracting multidimensional features from vibration signals to construct health degradation indicator, and health indicator is directly used for degradation prediction. However, constructed health indicator exhibit complex nonlinearity and fail to utilize the correlations with the original multidimensional features. To address this limitation, we propose an integrated prediction framework, known as the bayesian vector autoregression-assisted DeepAR method. The proposed two step prediction framework combines the advantages of covariance prediction and temporal correlation prediction, and realizes the full utilization of multidimensional features correlation information. Additionally, the proposed approach utilizes the intrinsic properties of statistical models to constrain improve the prediction results of deep learning methods, achieving nonlinear modeling of the degradation indicator while maintaining the stability of the prediction results. This process of prediction, re-estimation and adaptive adjustment achieves a balance in model performance. Finally, experimental studies using the PRONOSTIA and XJTU-SY datasets validate the effectiveness of the proposed method in improving the prediction accuracy of bearing health degradation indicators and enhancing the stability of multistep prediction in deep model.

Model-independent analysis of new physics effects in $$B\rightarrow K^*_2(1430)\mu ^+ \mu ^-$$ decay

AbstractRecently, the LHCb Collaboration provided updated measurements for the lepton flavour ratios $$R_K$$ R K and $$R_{K^*}$$ R K ∗ . The currently observed values align with the predictions of the standard model. In light of these recent updates, our investigation delves into the repercussions of new physics characterized by universal couplings to electrons and muons. We specifically focus on their impact on various observables within the $$B\rightarrow K_2^*(1430)(\rightarrow K\pi )\mu ^+ \mu ^-$$ B → K 2 ∗ ( 1430 ) ( → K π ) μ + μ - decay. These observables include the differential branching ratio, forward-backward asymmetry ($$A_{FB}$$ A FB ), longitudinal polarization asymmetry ($$F_L$$ F L ), and a set of optimized observables ($$P_i$$ P i ). Our findings indicate that the branching ratio of $$B\rightarrow K_2^*(\rightarrow K\pi )\mu ^+ \mu ^-$$ B → K 2 ∗ ( → K π ) μ + μ - decay can be suppressed up to $$25\%$$ 25 % for various new physics solutions. Furthermore, all permissible new physics scenarios demonstrate finite enhancement in the muon forward-backward asymmetry $$(A_{FB})$$ ( A FB ) as well as an increase in the value of the optimized angular observable $$P_2$$ P 2 . Moreover, in the presence of new physics zero crossing points for $$A_{FB}$$ A FB and $$P_2$$ P 2 shift towards higher $$q^2$$ q 2 . The current data have a mild deviation from SM predictions in $$P_5'$$ P 5 ′ observable in the low-$$q^2$$ q 2 bin. We also explored massive $$Z'$$ Z ′ models, which can generate universal 1D new physics scenarios, characterized by $$C_9^{NP}&lt;0$$ C 9 NP &lt; 0 , $$C_9^{NP}=-C_{10}^{NP}$$ C 9 NP = - C 10 NP , and $$C_9^{NP}=-C_9'$$ C 9 NP = - C 9 ′ . Using additional constraints coming from $$B_s-\overline{B_s}$$ B s - B s ¯ mixing and neutrino trident process, we find that the conclusions of the model-independent analysis remain valid.

Integrating deep learning and discrete cosine transform for surface waves full waveform inversion

Summary Accurate estimations of near-surface S-wave velocity (Vs) models hold particular significance in geological and engineering investigations. On the one hand, the popular Multichannel Analysis of Surface Waves (MASW) is limited to the 1D and the plane wave assumptions. On the other hand, the more advanced and computationally expensive full-waveform inversion (FWI) approach is often solved within a deterministic framework that hampers an accurate uncertainty assessment and makes the final predictions heavily reliant on the starting model. Here we combine deep learning with Discrete Cosine Transforms (DCT) to solve the FWI of surface waves and to efficiently estimate the inversion uncertainties. Our neural network approach effectively learns the inverse non-linear mapping between DCT-compressed seismograms and DCT-compressed S-velocity models. The incorporation of DCT into the deep learning framework provides several advantages: it notably reduces parameter space dimensionality and alleviates the ill-conditioning of the problem. Additionally, it decreases the complexity of the network architecture and the computational cost for the training phase compared to training in the full domain. A Monte Carlo simulation is also used to propagate the uncertainties from the data to the model space. We first test the implemented inversion method on synthetic data to showcase the generalization capabilities of the trained network and to explore the implications of incorrect noise assumptions in the recorded seismograms and inaccurate wavelet estimations. Further, we demonstrate the applicability of the implemented method to field data. In this case, available borehole information is used to validate our predictions. In both the synthetic and field applications, the predictions provided by the proposed method are compared with those of a deterministic FWI and the outcomes of a network trained in the full data and model spaces. Our experiments confirm that the implemented deep-learning inversion efficiently and successfully solves the FWI problem and yields more accurate and stable results than a network trained without the DCT compression. This opens the possibility to efficiently train a neural network that provides accurate instantaneous predictions of Vs near-surface models and related uncertainties.

Characterization of Long‐Term Municipal Solid Waste Constitutive Behavior With Coupled Biodegradation and Fibrous Reinforcing Effects

ABSTRACTTo appropriately simulate the long‐term mechanical behavior of municipal solid waste (MSW), a constitutive model coupling the effects of biodegradation and fibrous reinforcement was developed. In the proposed model, the compressive deformation due to biodegradation was regarded as being caused by an additional equivalent stress. Considering the effect of biodegradation, an evolution equation of the equivalent stress was proposed, and a plastic volumetric strain hardening law was developed. A fibrous reinforcement parameter was introduced, which was associated with the fiber content, stress state, and plastic shear strain of MSW. A plastic shear strain hardening law was developed to model the fibrous reinforcement. Based on the associated flow rule and two plastic strain hardening laws, the proposed model was established. The proposed model well simulated the hardening properties of MSW, as evidenced by the stress‒strain curves and the consistent, nonlinear increase in volumetric strain with axial strain. The differences in the shear strength and volumetric deformation due to the confining stress and fiber content were also well simulated by the model. Furthermore, the model predictions accurately reflected the findings of experiments conducted over a period of 10 years. Finally, parametric investigations were used to calibrate this proposed model, which can well characterize the long‐term MSW mechanical behavior.

Backdoor Attacks and Defenses Targeting Multi-Domain AI Models: A Comprehensive Review

Since the emergence of security concerns in artificial intelligence (AI), there has been significant attention devoted to the examination of backdoor attacks. Attackers can utilize backdoor attacks to manipulate model predictions, leading to significant potential harm. However, current research on backdoor attacks and defenses in both theoretical and practical fields still has many shortcomings. To systematically analyze these shortcomings and address the lack of comprehensive reviews, this paper presents a comprehensive and systematic summary of both backdoor attacks and defenses targeting multi-domain AI models. Simultaneously, based on the design principles and shared characteristics of triggers in different domains and the implementation stages of backdoor defense, this paper proposes a new classification method for backdoor attacks and defenses. We use this method to extensively review backdoor attacks in the fields of computer vision and natural language processing, and also examine the current applications of backdoor attacks in audio recognition, video action recognition, multimodal tasks, time series tasks, generative learning, and reinforcement learning, while critically analyzing the open problems of various backdoor attack techniques and defense strategies. Finally, this paper builds upon the analysis of the current state of AI security to further explore potential future research directions for backdoor attacks and defenses.

A Visco-Elasto-Plastic Constitutive Law for Deformation Prediction of High Concrete Face Rockfill Dams

Deformation predictions in high Concrete Face Rockfill Dams tend to underestimate observed settlements due to scale effect and breakage phenomena that cannot be adequately captured by laboratory tests. This paper presents a Visco-Elasto-Perfectly Plastic (VEPP) model for predicting deformations in high Concrete Face Rockfill Dams (CFRDs) that addresses these challenges incorporating explicitly key rockfill parameters like grain size and post-compaction porosity, which influence both the non-linear elastic and plastic behaviors of rockfill. The VEPP model enables deformation prediction while using standard laboratory test results. The model’s effectiveness was demonstrated through its application to the 233 m high Shuibuya Dam, the tallest CFRD in the world. The VEPP model predictions closely align with observed deformations throughout the dam’s construction, impoundment, and early operational stages. By using physically meaningful parameters, the model reduces the uncertainty associated with the empirical assessment of model parameters using back-analysis from similar projects. While the VEPP model offers improved predictive accuracy, particularly during early design phases, further advancements could be achieved by refining the creep formulation and accounting for grain size evolution during construction. This approach has the potential to optimize the design and construction of future high CFRD construction.

Accurate and efficient prediction of atmospheric PM1, PM2.5, PM10, and O3 concentrations using a customized software package based on a machine-learning algorithm.

Particulate matter (PM) and ozone (O3) pollution have been attracting increasing attention recently due to their severe harm to human health. PM and O3 are secondary pollutants, and there remain significant challenges in accurately and efficiently predicting their concentrations in the atmosphere. In this study, one-year monitoring data of PM1, PM2.5, PM10, and O3 concentrations as well as meteorological parameters and concentrations of various precursors (i.e., nitrogen oxides, SO2, CO, alkanes, aldehydes, and ketones) are obtained at a monitoring site in central China's Hunan province. The eXtreme Gradient Boosting model is trained and tested to achieve efficient and accurate predictions of the concentrations of the four pollutants. The effects of different datasets, input features, and model parameters on the prediction accuracy are investigated. Principal component analysis is employed to further reduce the dimensions of features, increasing the prediction efficiency. Finally, all model training and prediction processes are incorporated in an executable application using the PyQt5 framework to build a user interface. The customized software supports user-defined modeling. The software can simultaneously predict PM1, PM2.5, PM10, and O3 concentrations, making the prediction process convenient.

Improving computational models of deep brain stimulation through experimental calibration.

Deep brain stimulation has become a well-established clinical tool to treat movement disorders. Nevertheless, the knowledge of processes initiated by the stimulation remains limited. To address this knowledge gap, computational models are developed to gain deeper insight. However, their predictive power remains constrained by model uncertainties and a lack of validation and calibration. Exemplified with rodent microelectrodes, we present a workflow for validating electrode model geometry using microscopy and impedance spectroscopy in vitro before implantation. We address uncertainties in the tissue distribution and dielectric properties and outline a concept for calibrating the computational model based on in vivo impedance spectroscopy measurements. The standard deviation of the volume of tissue activated across the 18 characterized electrodes was approximately 32.93%, underscoring the importance of electrode characterization. Thus, the workflow significantly enhances the model predictions' credibility of neural activation exemplified in a rodent model. Computational models are frequently employed without validation or calibration, relying instead on manufacturers' specifications. Our approach provides an accessible method to obtain a validated and calibrated electrode geometry, which significantly enhances the reliability of the computational model that relies on this electrode. By reducing the uncertainties of the model, the accuracy in predicting neural activation is increased. The entire workflow is realized in open-source software, making it adaptable for other use cases, such as deep brain stimulation in humans. Additionally, the framework allows for the integration of further experiments, enabling live updates and refinements to computational models.

Explainable Machine Learning Models for Real-Time Threat Detection in Cybersecurity

In the rapidly evolving landscape of cybersecurity, traditional machine learning models often operate as "black boxes," providing high accuracy but lacking transparency in decision-making. This lack of explainability poses challenges for trust and accountability, especially in critical areas like threat detection and incident response. Explainable machine learning models aim to address this by making the model's predictions more understandable and interpretable to users. This research integrates explainable machine learning models for real-time threat detection in cybersecurity. Data from multiple sources, including network traffic, system logs, and user behavior, undergo preprocessing such as cleaning, feature extraction, and normalization. The processed data is passed through various machine learning models, including traditional approaches like SVM and decision trees, as well as deep learning models like CNN and RNN. Explainability techniques such as LIME, SHAP, and attention mechanisms provide transparency, ensuring interpretable predictions. The explanations are delivered through a user interface that generates alerts, visualizations, and reports, facilitating effective threat assessment and incident response in decision support systems. This framework enhances model performance, trust, and reliability in complex cybersecurity scenarios.

Simulation of soil phosphorus dynamics and crop yield for organic and mineral fertilization treatments at two long-term field sites.

The efficacy of phosphorus (P) based fertilizers is frequently compromised by soil dynamics that render much of the applied P unavailable for crops. This study aimed to: (i) validate a new P model's prediction of plant-available P; (ii) analyze the effects of organic versus mineral fertilization on P availability and crop yield; and (iii) examine temporal changes in P pools under various fertilization regimes. Data were collected from two long-term field trials, Dikopshof and Bad Lauchstädt, in Germany, using organic (FYM), mineral (MIN), a combination of organic and mineral (MIX) fertilizers, and unfertilized treatments. The AgroC model, incorporating a new P module, accurately predicted P dynamics in cropped plots. At both sites, MIX presented the highest yield, P removal, total P and available soil P. After 120 years of repeated P fertilization, simulations at Dikopshof revealed a positive P balance in MIN (11.1 % with observed 13 %) and in MIX (15 % with observed 15 %), but negative in FYM (-4.9 % with observed -5 %). However, at Bad Lauchstädt, the P balance was negative in all treatments except in MIN (+1.04 %), indicating P depletion. Among crops, cereals showed the most variated yields, with P-use efficiency ranging from 50 % to 99 %, while sugar beet presented the highest P-use efficiency (up to 122 %). The lowest P application rates exhibited, FYM treatment, the highest P-use efficiency for all crops. Model pools were successfully linked to field-measured soil P fractions using CAL and DGT methods, providing initial predictions of various soil P fractions across different fertilization strategies.

A framework for categorizing sources of uncertainty in in silico toxicology methods: considerations for chemical toxicity predictions.

Improving regulatory confidence and acceptance of in silico toxicology methods for chemical risk assessment requires assessment of associated uncertainties. Therefore, there is a need to identify and systematically categorize sources of uncertainty relevant to the methods and their predictions. In the present study, we analyzed studies that have characterized sources of uncertainty across commonly applied in silico toxicology methods. Our study reveals variations in the kind and number of uncertainty sources these studies cover. Additionally, the studies use different terminologies to describe similar sources of uncertainty; consequently, a majority of the sources considerably overlap. Building on an existing framework, we developed a new uncertainty categorization framework that systematically consolidates and categorizes the different uncertainty sources described in the analyzed studies. We then illustrate the importance of the developed framework through a case study involving QSAR prediction of the toxicity of five compounds, as well as compare it with the QSAR Assessment Framework (QAF). The framework can provide a structured (and potentially more transparent) understanding of where the uncertainties reside within in silico toxicology models and model predictions, thus promoting critical reflection on appropriate strategies to address the uncertainties.

Target Structure Learning Framework for Unsupervised Multi-Class Domain Adaptation

Unsupervised multi-class domain adaptation (multi-class UDA) has recently been proposed to fill the gap between empirically practical methods for multi-class classification and well-founded theory with the setting of binary classification. Nevertheless, the multi-class UDA methods use model predictions to characterize the disagreement of multi-class scoring hypotheses, which is used to optimize the divergence between domain distributions. Such self-training manner may bring inaccurate model predictions, which would damage the target structure due to the absence of labels of the target domain, leading to sub-optimal performance. On the other hand, this disagreement between multi-class scoring hypotheses does not involve the relationships among all of the multiple classes. It causes that multi-class UDA cannot properly connect the advanced practical UDA methods that consider class-conditional distribution alignment. Thus, we propose to exploit the target structure information and then incorporate it into multi-class UDA to achieve class-conditional distribution alignment. We theoretically and experimentally explain the importance of accurate target structure information to reduce the expected error on the target domain. Notably, our method achieves state-of-the-art results on three commonly-used benchmarks with different scales. In addition, using the target structure information, we propose a variant to cope with noisy open-world source domains such as noisy labels and out-of-distribution samples, enhancing the robustness of our method. The source code is available at https://github.com/jingzhengli/Multi_Class_UDA .

Intellectual capital forecasting for invention patent through machine learning model

PurposeThis research explores support vector machine (SVM) with Gaussian radial basis function kernel (RBF) as the model and Analysis of Variance (ANOVA) for forecasting the invalidation re-examination decisions of China invention patents, it is beneficial to support patent monetization for corporate intellectual capital.Design/methodology/approachThere were 8,666 China invention patents with their existing invalidation re-examination decisions during 2000∼2021 chosen to conduct classification model training and prediction for the accuracy of invalidation re-examination decisions through SVM with RBF. Statistical significance was performed by ANOVA to identify indicators for these invention patents selected in this research. These selected 8,666 China invention patents were divided into two groups based on their invalidation re-examination decisions during 2000∼2021 in Table 1, which Group 1 included 5,974 invention patents with all valid or partially valid claims, and Group 0 included 2,692 invention patents with all invalid claims. Thereafter, each group was further divided into sub-groups based on 13 major regions where the applicants filed invalidation re-examination. The training sets for Group 1, Group 0 and the sub-groups were selected based on the patent issued in January, February, April, May, July, August, October and November; while the prediction sets were selected from the invention patents issued in March, June, September and December.FindingsThe training and prediction accuracies were compared to the existing invalidation re-examination decisions. Accuracies of training sets were ranged from 100% in region 7 (Beijing) and region 9 (Shanghai) to 95.95% in region 1 (US), and the average accuracy of invalidation re-examination decisions was 98.95%. While the accuracies of prediction sets for Group 1 were ranged from 100.00% in region 7 (Beijing) to 90.78% in region 13 (Overseas-others), and the average accuracy of classification was 95.96%, this research’s outcomes confirmed the purpose of applying SVM with RBF to predict the patentability sustainability.Originality/valueThis research developed an empirical method through SVM with RBF to predict patentability sustainability which is crucial for corporate intellectual capital on patents. In particular, the investments on patents are huge, including the patent cultivation and maintenance, developments into products or services, patent litigations and dispute managements. Therefore, this research is beneficial not only for corporation, but also for research organisations to perform cost-effective and profitable patent strategies on intellectual capital.

SAL Method Applied in Grid Forecasting Product Verification with Three-Source Fusion Product

Quantitative precipitation forecast (QPF) verification stands out as one of the most formidable endeavors in the realm of forecast verification. Traditional verification methods are not suitable for high-resolution forecasting products in some cases. Therefore, the SAL (structure, amplitude and location) method was proposed as a method of object-based spatial verification that studies precipitation verification in a certain range, which is combined with factors including structure, amplitude and location of the targets. However, the setting of the precipitation threshold would affect the result of the verification. This paper presented an improved method for determining the precipitation threshold using the QPF from ECMWF, which is an ensemble forecast model and three-source fusion product that was used in China from 1 July to 31 August 2020, and then the results obtained with this method were compared with the other two traditional methods. Furthermore, the SAL and the traditional verification methods were carried out for geometric, simulated and real cases, respectively. The results showed the following: (1) The proposed method in this paper for determining the threshold was more accurate at identifying the precipitation objects. (2) The verification area size was critical for SAL calculation. If the area selected was too large, the calculated SAL value had little significance. (3) ME (Mean Error) could not identify the displacement between prediction and observation, while HSS (Heidke Skill Score) was sensitive to the displacement of the prediction field. (4) Compared with the traditional verification methods, the SAL method was more straight forward and simple, and it could give a better representation of prediction ability. Therefore, forecasters can better understand the model prediction effect and what needs to be improved.