Best-performing Method Research Articles

Assessing informative tract segmentation and nTMS for pre-operative planning

BackgroundDeep learning-based (DL) methods are the best-performing methods for white matter tract segmentation in anatomically healthy subjects. However, tract annotations are variable or absent in clinical data and manual annotations are especially difficult in patients with tumors where normal anatomy may be distorted. Direct cortical and subcortical stimulation is the gold standard ground truth to determine the cortical and sub-cortical lo- cation of motor-eloquent areas intra-operatively. Nonetheless, this technique is invasive, prolongs the surgical procedure, and may cause patient fatigue. Navigated Transcranial Magnetic Stimulation (nTMS) has a well-established correlation to direct cortical stimulation for motor mapping and the added advantage of being able to be acquired pre-operatively. New methodIn this work, we evaluate the feasibility of using nTMS motor responses as a method to assess corticospinal tract (CST) binary masks and estimated uncertainty generated by a DL-based tract segmentation in patients with diffuse gliomas. ResultsOur results show CST binary masks have a high overlap coefficient (OC) with nTMS response masks. A strong negative correlation is found between estimated uncertainty and nTMS response mask distance to the CST binary mask. Comparison with existing methodsWe compare our approach (UncSeg) with the state-of-the-art TractSeg in terms of OC between the CST binary masks and nTMS response masks. ConclusionsIn this study, we demonstrate that estimated uncertainty from UncSeg is a good measure of the agreement between the CST binary masks and nTMS response masks distance to the CST binary mask boundary.

Approaches for Hybrid Coregistration of Marker-Based and Markerless Coordinates Describing Complex Body/Object Interactions.

Full-body motion capture is essential for the study of body movement. Video-based, markerless, mocap systems are, in some cases, replacing marker-based systems, but hybrid systems are less explored. We develop methods for coregistration between 2D video and 3D marker positions when precise spatial relationships are not known a priori. We illustrate these methods on three-ball cascade juggling in which it was not possible to use marker-based tracking of the balls, and no tracking of the hands was possible due to occlusion. Using recorded video and motion capture, we aimed to transform 2D ball coordinates into 3D body space as well as recover details of hand motion. We proposed four linear coregistration methods that differ in how they optimize ball-motion constraints during hold and flight phases, using an initial estimate of hand position based on arm and wrist markers. We found that minimizing the error between ball and hand estimate was globally suboptimal, distorting ball flight trajectories. The best-performing method used gravitational constraints to transform vertical coordinates and ball-hold constraints to transform lateral coordinates. This method enabled an accurate description of ball flight as well as a reconstruction of wrist movements. We discuss these findings in the broader context of video/motion capture coregistration.

Segmentation methods applied to MRI-derived radiomic analysis for the prediction of placenta accreta spectrum in patients with placenta previa.

To retrospectively evaluate the performance of different manual segmentation methods of placenta MR images for predicting Placenta Accreta Spectrum (PAS) disorders in patients with placenta previa (PP) using a Machine Learning (ML) Radiomics analysis. 64 patients (n=41 with PAS and n= 23 without PAS) with PP who underwent MRI examination for suspicion of PAS were retrospectively selected. All MRI examinations were acquired on a 1.5 T using T2-weighted (T2w) sequences on axial, sagittal and coronal planes. Ten different manual segmentation methods were performed on sagittal placental T2-weighted images obtaining five sets of 2D regions of interest (ROIs) and five sets of 3D volumes of interest (VOIs) from each patient. In detail, ROIs and VOIs were positioned on the following areas: placental tissue, retroplacental myometrium, cervix, placenta with underneath myometrium, placenta with underneath myometrium and cervix. For feature stability testing, the same process was repeated on 30 randomly selected placental MRI examinations by two additional radiologists, working independently and blinded to the original segmentation. Radiomic features were extracted from all available ROIs and VOIs. 100 iterations of 5-fold cross-validation with nested feature selection, based on recursive feature elimination, were subsequently run on each ROI/VOI to identify the best-performing method to classify instances correctly. Among the segmentation methods, the best performance in predicting PAS was obtained by the VOIs covering the retroplacental myometrium (Mean validation score: 0.761, standard deviation: 0.116). Our preliminary results show that the VOI including the retroplacental myometrium using T2w images seems to be the best method when segmenting images for the development of ML radiomics predictive models to identify PAS in patients with PP.

Landmark-Assisted Anatomy-Sensitive Retinal Vessel Segmentation Network.

Automatic retinal vessel segmentation is important for assisting clinicians in diagnosing ophthalmic diseases. The existing deep learning methods remain constrained in instance connectivity and thin vessel detection. To this end, we propose a novel anatomy-sensitive retinal vessel segmentation framework to preserve instance connectivity and improve the segmentation accuracy of thin vessels. This framework uses TransUNet as its backbone and utilizes self-supervised extracted landmarks to guide network learning. TransUNet is designed to simultaneously benefit from the advantages of convolutional and multi-head attention mechanisms in extracting local features and modeling global dependencies. In particular, we introduce contrastive learning-based self-supervised extraction anatomical landmarks to guide the model to focus on learning the morphological information of retinal vessels. We evaluated the proposed method on three public datasets: DRIVE, CHASE-DB1, and STARE. Our method demonstrates promising results on the DRIVE and CHASE-DB1 datasets, outperforming state-of-the-art methods by improving the F1 scores by 0.36% and 0.31%, respectively. On the STARE dataset, our method achieves results close to the best-performing methods. Visualizations of the results highlight the potential of our method in maintaining topological continuity and identifying thin blood vessels. Furthermore, we conducted a series of ablation experiments to validate the effectiveness of each module in our model and considered the impact of image resolution on the results.

Reliability of smart noise pollution map

AbstractThe problem of noise pollution in Baghdad, the capital city of Iraq, is getting worse every day as a result of the increased volume of traffic. This presents a considerable risk, particularly on the main roads that connect densely populated neighborhoods such as the Al-Sadr City district with the central neighborhoods of the capital. In order to inform decision-makers in urban development and environmental policy about the high values of noise pollution that require remediation and regulation, noise maps are produced. However, two fundamental problems are generally faced in creating a more reliable noise map in the shortest possible time: the excessive time requirements for measuring noise and determining the method of map creation. Therefore, the role of geographic information system (GIS) software in producing noise maps is evident due to the difficulty of increasing the spatial density of measurements and integrating them with spatial information. Hence, an appropriate interpolation method is required. In this article, Moran’sIindex was calculated to assess the spatial autocorrelation of measured traffic noise points. A comparison was made between the Smart Map Plugin ordinary kriging (OK) and the inverse distance weighting (IDW) deterministic interpolation method to determine the best method for producing noise maps for the main entrance and exit roads of Al-Sadr City. The noise values were modeled using the best-performing method. Furthermore, the predictive raster data are displayed in the spatial context as a starting point and reference for identifying and understanding the levels of traffic noise in the selected study area. The locations of selected points for measuring traffic noise values were determined in an organized and homogeneous manner, where noise points for the main entrance and exit roads were opposite each other, and the distance between consecutive noise points on each road was 100 m. Traffic noise measurements were carried out at each selected point using the SVAN977 sound and vibration analyzer. At each measurement point, three noise values (LAeq, Max, Min) were obtained during the three peak times, 7–9 AM, 12–2 PM, and 4–6 PM. QGIS software was used to compare the two interpolation methods, with its strength lying in the use of plugins that facilitate spatial analysis, processing tools, and algorithms. The Smart Map Plugin provided facilities to choose the appropriate semi-variogram in the OK interpolation method. The root mean square error was used to compare the two interpolation methods in order to determine the most suitable method for producing traffic noise maps in the study area. The results indicated that the Smart Map Plugin using OK outperformed the IDW method, as spatial distribution pattern and homogeneity affect the accuracy of interpolation. Moreover, based on the analysis of the three noise attributes (LAeq, Max, Min), the performance of the Smart Map Plugin (OK) was found to be better than IDW when the Moran’sIvalue was high.

More than just pattern recognition: Prediction of uncommon protein structure features by AI methods

The CASP14 experiment demonstrated the extraordinary structure modeling capabilities of artificial intelligence (AI) methods. That result has ignited a fierce debate about what these methods are actually doing. One of the criticisms has been that the AI does not have any sense of the underlying physics but is merely performing pattern recognition. Here, we address that issue by analyzing the extent to which the methods identify rare structural motifs. The rationale underlying the approach is that a pattern recognition machine tends to choose the more frequently occurring motifs, whereas some sense of subtle energetic factors is required to choose infrequently occurring ones. To reduce the possibility of bias from related experimental structures and to minimize the effect of experimental errors, we examined only CASP14 target protein crystal structures determined to a resolution limit better than 2 Å, which lacked significant amino acid sequence homology to proteins of known structure. In those experimental structures and in the corresponding models, we track cis peptides, π-helices, 310-helices, and other small 3D motifs that occur in the PDB database at a frequency of lower than 1% of total amino acid residues. The best-performing AI method, AlphaFold2, captured these uncommon structural elements exquisitely well. All discrepancies appeared to be a consequence of crystal environment effects. We propose that the neural network learned a protein structure potential of mean force, enabling it to correctly identify situations where unusual structural features represent the lowest local free energy because of subtle influences from the atomic environment.

Deep learning approaches for MIMO time-series analysis

This study presents a comparative analysis of various deep learning (DL) methods for multi-input and multi-output (MIMO) time-series forecasting of stock prices. The analysis is conducted on a dataset comprising the stock price of Bitcoin. The dataset consists of 2950 rows from December 2017 to December 2021. This study aims to evaluate the performance of multiple DL methods, including Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Long Short Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), and Gated Recurrent Unit (GRU). The evaluation criteria for selecting the best-performing methods in this research are based on two performance metrics: Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE). These metrics were chosen for specific reasons related to assessing the accuracy and reliability of the forecasting models. MAPE is used to assess accuracy, while RMSE helps detect outliers in the system. Results show that the LSTM method achieves the best performance, outperforming other methods with an average MAPE value of 8.73% and Bi-LSTM has the best average RMSE value of 0.02216. The findings of this study have practical implications for time-series forecasting in the field of stock trading. The superior performance of LSTM highlights its potential as a reliable method for accurately predicting stock prices. The Bi-LSTM model's ability to detect outliers can aid in identifying abnormal stock market behavior. In summary, this research provides insights into the performance of various DL models of MIMO for stock price forecasting. The results contribute to the field of time-series forecasting and offer valuable guidance for decision-making in stock trading by identifying the most effective methods for predicting stock prices accurately and detecting unusual market behavior.

Mitigating Negative Transfer in Multi-Task Learning with Exponential Moving Average Loss Weighting Strategies (Student Abstract)

Multi-Task Learning (MTL) is a growing subject of interest in deep learning, due to its ability to train models more efficiently on multiple tasks compared to using a group of conventional single-task models. However, MTL can be impractical as certain tasks can dominate training and hurt performance in others, thus making some tasks perform better in a single-task model compared to a multi-task one. Such problems are broadly classified as negative transfer, and many prior approaches in the literature have been made to mitigate these issues. One such current approach to alleviate negative transfer is to weight each of the losses so that they are on the same scale. Whereas current loss balancing approaches rely on either optimization or complex numerical analysis, none directly scale the losses based on their observed magnitudes. We propose multiple techniques for loss balancing based on scaling by the exponential moving average and benchmark them against current best-performing methods on three established datasets. On these datasets, they achieve comparable, if not higher, performance compared to current best-performing methods.

A Question-Answering Approach to Key Value Pair Extraction from Form-Like Document Images

In this paper, we present a new question-answering (QA) based key-value pair extraction approach, called KVPFormer, to robustly extracting key-value relationships between entities from form-like document images. Specifically, KVPFormer first identifies key entities from all entities in an image with a Transformer encoder, then takes these key entities as questions and feeds them into a Transformer decoder to predict their corresponding answers (i.e., value entities) in parallel. To achieve higher answer prediction accuracy, we propose a coarse-to-fine answer prediction approach further, which first extracts multiple answer candidates for each identified question in the coarse stage and then selects the most likely one among these candidates in the fine stage. In this way, the learning difficulty of answer prediction can be effectively reduced so that the prediction accuracy can be improved. Moreover, we introduce a spatial compatibility attention bias into the self-attention/cross-attention mechanism for KVPFormer to better model the spatial interactions between entities. With these new techniques, our proposed KVPFormer achieves state-of-the-art results on FUNSD and XFUND datasets, outperforming the previous best-performing method by 7.2% and 13.2% in F1 score, respectively.

Probability Density Estimation through Nonparametric Adaptive Partitioning and Stitching

We present a novel nonparametric adaptive partitioning and stitching (NAPS) algorithm to estimate a probability density function (PDF) of a single variable. Sampled data is partitioned into blocks using a branching tree algorithm that minimizes deviations from a uniform density within blocks of various sample sizes arranged in a staggered format. The block sizes are constructed to balance the load in parallel computing as the PDF for each block is independently estimated using the nonparametric maximum entropy method (NMEM) previously developed for automated high throughput analysis. Once all block PDFs are calculated, they are stitched together to provide a smooth estimate throughout the sample range. Each stitch is an averaging process over weight factors based on the estimated cumulative distribution function (CDF) and a complementary CDF that characterize how data from flanking blocks overlap. Benchmarks on synthetic data show that our PDF estimates are fast and accurate for sample sizes ranging from 29 to 227, across a diverse set of distributions that account for single and multi-modal distributions with heavy tails or singularities. We also generate estimates by replacing NMEM with kernel density estimation (KDE) within blocks. Our results indicate that NAPS(NMEM) is the best-performing method overall, while NAPS(KDE) improves estimates near boundaries compared to standard KDE.

Electricity Day-Ahead Market Conditions and Their Effect on the Different Supervised Algorithms for Market Price Forecasting

Participants in deregulated electricity markets face risks from price volatility due to various factors, including fuel prices, renewable energy production, electricity demand, and crises such as COVID-19 and energy-related issues. Price forecasting is used to mitigate risk in markets trading goods which have high price volatility. Forecasting in electricity markets is difficult and challenging as volatility is attributed to many unpredictable factors. This work studies and reports the performance both in terms of forecasting error and of computational time of forecasting algorithms that are based on Extreme Learning Machine, Artificial Neural Network, XGBoost and random forest. All these machine learning techniques are combined with the Bootstrap technique of creating new samples from the available ones in order to improve the forecasting errors. In order to assess the performance of these methodologies, the Day-Ahead market prices are divided into three classes, namely normal, extremely high and negative, and these algorithms are subsequently used to provide forecasts for the whole year 2020 of the German and Finnish Day-Ahead markets. The average yearly forecasting errors along with the computation time required by each methodology are reported. The findings indicate that the random forest algorithm performs best for the normal and extremely high price categories, while XGBoost demonstrates better results for the negative price category. The methodology based on Extreme Learning Machine requires the least computational time and achieves forecasting errors that are comparable to the best-performing methods.

An evaluation framework for downscaling and bias correction in climate change impact studies

Climate change impact studies commonly use impact models (such as hydrological or crop models) forced with corrected climate input data from global climate models. A range of downscaling and bias correction methods have been developed to increase the spatial resolution and remove systematic biases in climate model outputs to be applied before use in impact models. Many studies have focused on evaluating such approaches for the climate variables they aim to correct. However, due to nonlinear error propagation there can be large remaining biases in model outputs, even when ingesting bias corrected climate forcings.Here we propose an impact-centric evaluation framework for downscaling and bias correction methods to be used in climate change risk assessments. This framework evaluates and compares the strengths and limitations of downscaling and bias correction in the impact domain, highlighting approaches that lead to reduced biases in impact variables of interest. We demonstrate the evaluation framework in the context of assessing downscaling and bias correction methods for hydrological projections in Australia. Our results show that although all downscaling and bias correction methods evaluated perform adequately for the input climate variables, their errors vary markedly when the impact is modelled. Our proposed evaluation framework involves selecting a number of key performance metrics, and ranking the four downscaling and bias correction methods to compute an overall ranking, highlighting the best-performing methods for each statistical metric and the overall best-performing approach. We present an application of this approach using performance metrics relevant to hydrological applications, relating to mean biases, variability, heavy precipitation and peak runoff days, and dry conditions.For impact studies related to hydrological applications, we find that multi-variate bias correction that considers cross-correlations, temporal auto-correlations and biases at multiple time scales (daily to annual) performs best in reducing biases in hydrological output variables for Australia. Our proposed evaluation approach can be applied to a wide range of climate change applications where downscaling and bias correction are required, including impacts on agricultural production, wildfires, energy generation, human health, ecosystem functioning, and water resource management.

Exploring Tree-Based Machine Learning Models to Estimate the Ultimate Pile Capacity From Cone Penetration Test Data

Several approaches have been developed to estimate the ultimate capacity of piles, such as static and dynamic load tests, static analysis from soil borings, and directly utilizing in-situ test results. Recently, there has been increased interest in using in-situ cone penetration test (CPT) to estimate pile capacity. Several analytical pile-CPT methods have been developed, which involve several correlation assumptions that can affect their accuracy. In this paper, three tree-based machine learning (ML) models, namely decision tree (DT), random forest (RF), and gradient boosted tree (GBT), are developed for estimating the ultimate capacity of piles from CPT data. A database that contains 80 pile load tests and associated CPT data collected in Louisiana was used to develop these ML models. The measured ultimate pile capacity (Qm) was determined using Davisson’s interpretation method from the load–settlement curve of each pile load test. Among the developed ML models, GBT demonstrated the most accurate ML model compared with the others. The estimation of ultimate pile capacity from the GBT model is compared with those obtained from the four best-performing direct pile-CPT methods (based on a previous study): the University of Florida (UF), probabilistic, European Regional Technical Committee 3 (ERTC3), and Laboratoire Central des Ponts et Chaussées (LCPC) methods. The GBT and pile-CPT methods were evaluated and ranked based on analysis of multiple statistical criteria. The results clearly showed that the GBT model outperforms the four direct pile-CPT methods for estimating the ultimate capacity of piles.

A Review and Comparative Study of Differential Evolution Algorithms in Solving Inverse Kinematics of Mobile Manipulator

Mobile manipulator robots have become important pieces of equipment due to the high mobility of mobile subsystems and the high flexibility of manipulator subsystems. Considering the increasing degrees of freedom and the need to avoid singular locations, one of the most challenging problems is solving the inverse kinematics problem of mobile manipulator robots (IKMM). Of all the popular optimization algorithms, the differential evolution (DE) algorithm is the most effective method for quickly solving the IKMM problem with sufficient solutions. Currently, many strategies have been proposed for DE algorithms to improve the performance of solving mathematical problems; some symmetry strategies or symmetry functions have been introduced to DE algorithms. However, the effects of various DE algorithms on solving the actual IKMM lack a comprehensive explanation. Therefore, we divide various DE algorithms into three categories considering the control parameter selection and compare the specific optimization of various DE algorithms. Then, we compare the performance of various DE algorithms when solving the inverse kinematics problems of mobile manipulators with different degrees of freedom. Considering the effectiveness and the speed of the DE algorithm on the IKMM problem, we determine the best DE algorithm by comparing the error and time required to reach 100 random mission points and tracking the typical trajectories. Finally, the best-performing DE method is further improved by studying the selection of fundamental parameters in the best DE algorithm. Valuable conclusions are obtained from these experimental simulations, which can help with choosing an algorithm that is suitable for solving the inverse kinematics problem of mobile manipulator robots in practice.

Predicting nutritional status for women of childbearing age from their economic, health, and demographic features: A supervised machine learning approach.

Malnutrition imposes enormous costs resulting from lost investments in human capital and increased healthcare expenditures. There is a dearth of research focusing on the prediction of women's body mass index (BMI) and malnutrition outcomes (underweight, overweight, and obesity) in developing countries. This paper attempts to fill out this knowledge gap by predicting the BMI and the risks of malnutrition outcomes for Bangladeshi women of childbearing age from their economic, health, and demographic features. Data from the 2017-18 Bangladesh Demographic and Health Survey and a series of supervised machine learning (SML) techniques are used. Additionally, this study circumvents the imbalanced distribution problem in obesity classification by utilizing an oversampling approach. Study findings demonstrate that the support vector machine and k-nearest neighbor are the two best-performing methods in BMI prediction based on the coefficient of determination (R2), root mean square error (RMSE), and mean absolute error (MAE). The combined predictor algorithms consistently yield top specificity, Cohen's kappa, F1-score, and AUC in classifying the malnutrition status, and their performance is robust to alternative standards. The feature importance ranking based on several nonparametric and combined predictors indicates that socioeconomic status, women's age, and breastfeeding status are the most important features in predicting women's nutritional outcomes. Furthermore, the conditional inference trees corroborate that those three features, along with the partner's educational attainment and employment status, significantly predict malnutrition risks. To the best of our knowledge, this is the first study that predicts BMI and one of the pioneer studies to classify all three malnutrition outcomes for women of childbearing age in Bangladesh, let alone in any lower-middle income country, using SML techniques. Moreover, in the context of Bangladesh, this paper is the first to identify and rank features that are critical in predicting nutritional outcomes using several feature selection algorithms. The estimators from this study predict the outcomes of interest most accurately and efficiently compared to other existing studies in the relevant literature. Therefore, study findings can aid policymakers in designing policy and programmatic approaches to address the double burden of malnutrition among Bangladeshi women, thereby reducing the country's economic burden.

Using econometric and machine learning models to forecast crude oil prices: Insights from economic history

The volatility of the crude oil market and its effects on the global economy increased the concerns of individual investors, states/governments, and corporations. Forecasting the price of crude oil is difficult owing to its complicated, nonlinear, and chaotic nature in economic history. Multiple variables influence crude oil prices, such as the economic history, economic cycle, international relations, and geopolitics. Predicting the price of crude oil is a complex but valuable endeavor. Crude oil price forecasting is done using historical data (time series method) or dependent variables/factors (regression method) using traditional econometric or machine learning models. In this study, we use both methods (regression and time series) to examine the prediction performance of both models (econometric and machine learning models) for daily WTI crude oil prices covering the period December 18, 2011, through December 31, 2018. We present a performance analysis of conventional econometric models (ARIMA, GARCH, and OLS), Artificial Neural Network (ANN) regression models, and ANN Time Series models to compare their results to find out the best-performing method (time series or regression) and the best model (econometric or machine learning model). Based on our study results, we propose a novel Artificial Neural Network model to improve the prediction performance of existing models by adjusting the bias and weights of ANN hidden layers. We used historical prices of 14 different variables, including gold, silver, S&P500, USD Index price, and US-EU conversion rates for regression models, whereas historical time series data of WTI crude oil for time series models. Analysis of the results reveals that the performance of our proposed model remained better than all tested models. The comparative results of existing models show that the overall performance of Neural Networks remained better than econometric models. Our results have substantial implications for governments, businesses, and investors, and for the sustainable growth of economies that rely on energy.

SBERTiment: A New Pipeline to Solve Aspect Based Sentiment Analysis in the Zero-Shot Setting

The field of Natural Language Processing is gaining increased attention for the Aspect Based Sentiment Analysis task due to its ability to provide fine-grained information. This paper introduces SBERTiment, a novel approach to perform Aspect Based Sentiment Analysis. The method extracts relevant topics along with their sentiments from the input text by using a 2-step pipeline. In the first step, a token classification model is used to identify the relevant aspect terms and their sentiments. In the second step, a Sentence-BERT embedding model maps each aspect term to a predefined aspect category. Our approach has been tested on benchmark datasets and has achieved scores that are comparable to the best-performing methods. The pipeline is also able to perform zero-shot classification, which means it can extract information in unseen domains without additional training. When evaluated on a dataset with unseen aspect categories, SBERTiment achieved the best score among benchmark approaches.

Predicting flow in porous media: a comparison of physics-driven neural network approaches

This paper presents the development of physics-informed machine learning models for subsurface flows, specifically for determining pressure variation in the subsurface without the use of numerical modeling schemes. The numerical elliptic operator is replaced with a neural network operator and includes comparisons of several different machine learning models, along with linear regression, support vector regression, lasso, random forest regression, decision tree regression, light weight gradient boosting, eXtreme gradient boosting, convolution neural network, artificial neural network, and perceptron. The mean of absolute error of all models is compared, and error residual plots are used as a measure of error to determine the best-performing method.

Efficient Point Cloud Object Classifications with GhostMLP

Efficient models capable of handling large numbers of data points in point cloud research are in high demand in computer vision. Despite recent advancements in 3D classification and segmentation tasks in point cloud processing, the deep learning PointNeXt and PointMLP models are plagued with heavy computation requirements with limited efficiencies. In this paper, a novel GhostMLP model for point clouds is thus introduced. It takes the advantages of the GhostNet design modules and uses them to replace the MLP layers in the existing PointMLP model. The resulting GhostMLP architecture achieves superior classification performance with lower computation requirements. Compared to the PointMLP, GhostMLP maintains sustainable performance with fewer parameters and lower FLOPs computations. Indeed, it outperforms PointMLP on the ScanObjectNN dataset, achieving 88.7% overall accuracy and 87.6% mean accuracy with only 6 million parameters and 7.2 GFLOPs—about half the resources required by PointMLP. At the same time, GhostMLP-S is introduced as a lightweight version which also outperforms PointMLP in performance. GhostMLP completes faster training and inference with GPU and is the best-performing method that does not require any extra training data in the ScanObjectNN benchmark. Efficient point cloud analysis is essential in computer vision, and we believe that GhostMLP has the potential to become a powerful tool for large-scale point cloud analysis.

Evaluation of bias correction methods for a multivariate drought index: case study of the Upper Jhelum Basin

Abstract. Bias correction (BC) is often a necessity to improve the applicability of global and regional climate model (GCM and RCM, respectively) outputs to impact assessment studies, which usually depend on multiple potentially dependent variables. To date, various BC methods have been developed which adjust climate variables separately (univariate BC) or jointly (multivariate BC) prior to their application in impact studies (i.e., the component-wise approach). Another possible approach is to first calculate the multivariate hazard index from the original, biased simulations and bias-correct the impact model output or index itself using univariate methods (direct approach). This has the advantage of circumventing the difficulties associated with correcting the inter-variable dependence of climate variables which is not considered by univariate BC methods. Using a multivariate drought index (i.e., standardized precipitation evapotranspiration index – SPEI) as an example, the present study compares different state-of-the-art BC methods (univariate and multivariate) and BC approaches (direct and component-wise) applied to climate model simulations stemming from different experiments at different spatial resolutions (namely Coordinated Regional Climate Downscaling Experiment (CORDEX), CORDEX Coordinated Output for Regional Evaluations (CORDEX-CORE), and 6th Coupled Intercomparison Project (CMIP6)). The BC methods are calibrated and evaluated over the same historical period (1986–2005). The proposed framework is demonstrated as a case study over a transboundary watershed, i.e., the Upper Jhelum Basin (UJB) in the Western Himalayas. Results show that (1) there is some added value of multivariate BC methods over the univariate methods in adjusting the inter-variable relationship; however, comparable performance is found for SPEI indices. (2) The best-performing BC methods exhibit a comparable performance under both approaches with a slightly better performance for the direct approach. (3) The added value of the high-resolution experiments (CORDEX-CORE) compared to their coarser-resolution counterparts (CORDEX) is not apparent in this study.