Achieved Prediction Accuracy Research Articles

Forecasting rare earth stock prices with machine learning

Rare earth elements (REEs) are indispensable for producing green technologies and electronics. Demand for REEs in clean energy technologies in 2040 are projected to be three to seven times higher than today and will be critical to the clean technology transition needed to stave off catastrophic climate change. Forecasting rare earth stock prices is critical for making well informed investment decisions concerning this important asset class. Despite the latter, the literature on forecasting rare earth stock prices is scarce. We use machine learning techniques to forecast daily rare earth stock price direction. The analysis reveals that random forests, extremely randomized trees, RNN, and support vector machine have higher prediction accuracy than Lasso or Naïve Bayes. We find that the 10- to 20-day forecasts using random forests, extremely randomized trees, and support vector machine achieve prediction accuracies greater than 85% with some prediction accuracy reaching 90%. Lasso prediction accuracy is higher than Naïve Bayes but never greater than 67%. The MA200, MA50, on balance volume, VIX, and WAD are the most important predictive features of rare earth stock price direction. A switching portfolio that uses trading signals from an Extra Trees model impressively outperforms a buy and hold portfolio. Our results reveal the high prediction accuracy of using machine learning methods in forecasting rare earth stock price direction which should be useful to investors, policy makers and venture capitalists.

Geographical origin identification of Hainan camellia oil based on fatty acid composition and near infrared spectroscopy combined with chemometrics

Ensuring the authenticity of the original production region is of utmost importance in safeguarding the reputation and ensuring the quality and safety of Hainan camellia oil, which possesses unique quality and commands a higher price than camellia oil from other main producing areas in China. This study explored the potential of fatty acid composition and near infrared (NIR) spectra for geographical traceability of Hainan camellia oil. The relative content of 16 fatty acids in camellia oil samples was determined using gas chromatography (GC), and the spectral information of the samples was obtained using NIR spectroscopy. The data were then analyzed using chemometrics methods, comparing the classification abilities of partial least squares discriminant analysis (PLS-DA), random forest (RF), support vector machine (SVM), and convolutional neural network (CNN) algorithms. The results demonstrated that the SVM model based on the fatty acid composition, the CNN model based on the NIR spectra, and the CNN model based on data fusion achieved prediction accuracies of 97.08%, 97.92%, and 98.75%, respectively, enabling high-precision identification of the geographical origin of Hainan camellia oil. This study reveals that the fatty acid composition and NIR spectra can serve as accurate tools for identifying the geographical origin of camellia oil.

Data-driven interpretable analysis for polysaccharide yield prediction

Cornstalks show promise as a raw material for polysaccharide production through xylanase. Rapid and accurate prediction of polysaccharide yield can facilitate process optimization, eliminating the need for extensive experimentation in actual production to refine reaction conditions, thereby saving time and costs. However, the intricate interplay of enzymatic factors poses challenges in predicting and optimizing polysaccharide yield accurately. Here, we introduce an innovative data-driven approach leveraging multiple artificial intelligence techniques to enhance polysaccharide production. We propose a machine learning framework to identify highly accurate polysaccharide yield prediction modeling methods and uncover optimal enzymatic parameter combinations. Notably, Random Forest (RF) and eXtreme Gradient Boost (XGB) demonstrate robust performance, achieving prediction accuracies of 93.0% and 95.6%, respectively, while an independently developed deep neural network (DNN) model achieves 91.1% accuracy. A feature importance analysis of XGB reveals the enzyme solution volume's dominant role (43.7%), followed by time (20.7%), substrate concentration (15%), temperature (15%), and pH (5.6%). Further interpretability analysis unveils complex parameter interactions and potential optimization strategies. This data-driven approach, incorporating machine learning, deep learning, and interpretable analysis, offers a viable pathway for polysaccharide yield prediction and the potential recovery of various agricultural residues.

Comparison of decision tree and naïve Bayes algorithms in detecting trace residue of gasoline based on gas chromatography-mass spectrometry data.

Fire debris analysis aims to detect and identify any ignitable liquid residues in burnt residues collected at a fire scene. Typically, the burnt residues are analysed using gas chromatography-mass spectrometry (GC-MS) and are manually interpreted. The interpretation process can be laborious due to the complexity and high dimensionality of the GC-MS data. Therefore, this study aims to compare the potential of classification and regression tree (CART) and naïve Bayes (NB) algorithms in analysing the pixel-level GC-MS data of fire debris. The data comprise 14 positive (i.e. fire debris with traces of gasoline) and 24 negative (i.e. fire debris without traces of gasoline) samples. The differences between the positive and negative samples were first inspected based on the mean chromatograms and scores plots of the principal component analysis technique. Then, CART and NB algorithms were independently applied to the GC-MS data. Stratified random resampling was applied to prepare three sets of 200 pairs of training and testing samples (i.e. split ratio of 7:3, 8:2, and 9:1) for estimating the prediction accuracies. Although both the positive and negative samples were hardly differentiated based on the mean chromatograms and scores plots of principal component analysis, the respective NB and CART predictive models produced satisfactory performances with the normalized GC-MS data, i.e. majority achieved prediction accuracy >70%. NB consistently outperformed CART based on the prediction accuracies of testing samples and the corresponding risk of overfitting except when evaluated using only 10% of samples. The accuracy of CART was found to be inversely proportional to the number of testing samples; meanwhile, NB demonstrated rather consistent performances across the three split ratios. In conclusion, NB seems to be much better than CART based on the robustness against the number of testing samples and the consistent lower risk of overfitting.

Optimizing electric vehicle charging schedules and energy management in smart grids using an integrated GA-GRU-RL approach

Introduction: Smart grid technology is a crucial direction for the future development of power systems, with electric vehicles, especially new energy vehicles, serving as important carriers for smart grids. However, the main challenge faced by smart grids is the efficient scheduling of electric vehicle charging and effective energy management within the grid.Methods: To address this issue, we propose a novel approach for intelligent grid electric vehicle charging scheduling and energy management, integrating three powerful technologies: Genetic Algorithm (GA), Gated Recurrent Unit (GRU) neural network, and Reinforcement Learning (RL) algorithm. This integrated approach enables global search, sequence prediction, and intelligent decision-making to optimize electric vehicle charging scheduling and energy management. Firstly, the Genetic Algorithm optimizes electric vehicle charging demands while minimizing peak grid loads. Secondly, the GRU model accurately predicts electric vehicle charging demands and grid load conditions, facilitating the optimization of electric vehicle charging schedules. Lastly, the Reinforcement Learning algorithm focuses on energy management, aiming to minimize grid energy costs while meeting electric vehicle charging demands.Results and discussion: Experimental results demonstrate that the method achieves prediction accuracy and recall rates of 97.56% and 95.17%, respectively, with parameters (M) and triggers (G) at 210.04 M and 115.65G, significantly outperforming traditional models. The approach significantly reduces peak grid loads and energy costs while ensuring the fulfilment of electric vehicle charging demands and promoting the adoption of green energy in smart city environments.

Predicting tensile strength of material extrusion parts during the pre-process using neural networks

Quantitative quality characteristics of additive manufactured parts are influenced by parameters selected in the preparation process (pre-process), especially in the material extrusion process. As a result, a prediction of the tensile strength of manufactured parts is hardly possible, which significantly reduces the usability of the process. In this paper a neural network approach is used to predict the tensile strength during the pre-process. The parameters investigated are print speed, number of shells, layer thickness, nozzle temperature and infill density. A prediction with a mean absolute percentage error (MAPE) of 2.54% could be achieved for randomly generated process parameters using a training data set of 243 samples. This exceeds the best prediction accuracies of the current literature which is between 2.56 and 3.34%. However, this research is particularly different in that, unlike the existing literature, the developed prediction models were tested with untrained random parameter values in a properly conducted test. With a data reduction to a data volume of 32 samples the used approach achieved already a MAPE of 4.15%. The neural network approach outperformed a multiple linear regression even at low training data volume. This publication differs from previously published research activities due to the achieved prediction accuracies on random parameter sets, the number of investigated parameters and the sample size. Users are provided with an algorithm and its procedure to predict the tensile strength which can be adapted to the respective application with the help of company data.

A Comprehensive Prediction Method for Pore Pressure in Abnormally High-Pressure Blocks Based on Machine Learning

In recent years, there has been significant research and practical application of machine learning methods for predicting reservoir pore pressure. However, these studies frequently concentrate solely on reservoir blocks exhibiting normal-pressure conditions. Currently, there exists a scarcity of research addressing the prediction of pore pressure within reservoir blocks characterized by abnormally high pressures. In light of this, the present paper introduces a machine learning-based approach to predict pore pressure within reservoir blocks exhibiting abnormally high pressures. The methodology is demonstrated using the X block as a case study. Initially, the combination of the density–sonic velocity crossplot and the Bowers method is favored for elucidating the overpressure-to-compact mechanism within the X block. The elevated pressure within the lower reservoir is primarily attributed to the pressure generated during hydrocarbon formation. The Bowers method has been chosen to forecast the pore pressure in well X-1. Upon comparison with real pore pressure data, the prediction error is found to be under 5%, thus establishing it as a representative measure of the reservoir’s pore pressure. Intelligent prediction models for pore pressure were developed using the KNN, Extra Trees, Random Forest, and LightGBM algorithms. The models utilized five categories of well logging data, sonic time difference (DT), gamma ray (GR), density (ZDEN), neutron porosity (CNCF), and well diameter (CAL), as input. After training and comparison, the results demonstrate that the LightGBM model exhibits significantly superior performance compared to the other models. Specifically, it achieves R2 values of 0.935 and 0.647 on the training and test sets, respectively. The LightGBM model is employed to predict the pore pressure of two wells neighboring well X-1. Subsequently, the predicted data are juxtaposed with the actual pore pressure measurements to conduct error analysis. The achieved prediction accuracy exceeds 90%. This study delivers a comprehensive analysis of pore pressure prediction within sections exhibiting anomalously high pressure, consequently furnishing scientific insights to facilitate both secure and efficient drilling operations within the X block.

A cusp-capturing PINN for elliptic interface problems

In this paper, we propose a cusp-capturing physics-informed neural network (PINN) to solve discontinuous-coefficient elliptic interface problems whose solution is continuous but has discontinuous first derivatives on the interface. To find such a solution using neural network representation, we introduce a cusp-enforced level set function as an additional feature input to the network to retain the inherent solution properties; that is, capturing the solution cusps (where the derivatives are discontinuous) sharply. In addition, the proposed neural network has the advantage of being mesh-free, so it can easily handle problems in irregular domains. We train the network using the physics-informed framework in which the loss function comprises the residual of the differential equation together with certain interface and boundary conditions. We conduct a series of numerical experiments to demonstrate the effectiveness of the cusp-capturing technique and the accuracy of the present network model. Numerical results show that even using a one-hidden-layer (shallow) network with a moderate number of neurons and sufficient training data points, the present network model can achieve prediction accuracy comparable with traditional methods. Besides, if the solution is discontinuous across the interface, we can simply incorporate an additional supervised learning task for solution jump approximation into the present network without much difficulty.

An interpretable machine learning model for predicting cavity water depth and cavity length based on XGBoost–SHAP

Abstract In contrast to the traditional black box machine learning model, the white box model can achieve higher prediction accuracy and accurately evaluate and explain the prediction results. Cavity water depth and cavity length of aeration facilities are predicted in this research based on Extreme Gradient Boosting (XGBoost) and a Bayesian optimization technique. The Shapley Additive Explanation (SHAP) method is then utilized to explain the prediction results. This study demonstrates how SHAP may order all features and feature interaction terms in accordance with the significance of the input features. The XGBoost–SHAP white box model can reasonably explain the prediction results of XGBoost both globally and locally and can achieve prediction accuracy comparable to the black box model. The cavity water depth and cavity length white box model developed in this study have a promising future application in the shape optimization of aeration facilities and the improvement of model experiments.

Quality evaluation of preschool video games based on RNN

With the rapid development of video multimedia technology, video games have started to be applied in the field of education. A video game quality evaluation method needs to be designed to help teaching staff select video materials. In view of this, a Recurrent Neural Network (RNN)-based video game quality evaluation method for preschool education is proposed. Firstly, an improved attention mechanism null domain weight assignment method is proposed for the video quality evaluation problem. Then a video game quality evaluation model based on RNN is proposed. Finally, performance testing is conducted on the proposed quality evaluation method. The results show that in the comparison of fitness values, when the number of iterations is 4 or 16, the maximum fitness value and the average fitness value of the research method are 97 and 96 respectively. In the prediction of different video datasets, the research methods can achieve prediction accuracy of 0.958 and 0.947 for LIVE and CSIQ videos respectively. In the analysis of practical application effects, all students show significant improvement in language reading, writing, understanding, and spelling. The above results indicate that the research method has a faster convergence speed and prediction accuracy. It can effectively improve students' learning efficiency and provide new references for the education and teaching evaluation of preschool children.

Different-Classification-Scheme-Based Machine Learning Model of Building Seismic Resilience Assessment in a Mountainous Region

This study aims to develop different-classification-scheme-based building-seismic-resilience (BSR)-mapping models using random forest (RF) and a support vector machine (SVM). Based on a field survey of earthquake-damaged buildings in Shuanghe Town, the epicenter of the Changning M 5.8 earthquake that occurred on 17 June 2019, we selected 19 influencing factors for BSR assessment to establish a database. Based on three classification schemes for the description of BSR, we developed six machine learning assessment models for BSR mapping using RF and an SVM after optimizing the hyper-parameters. The validation indicators of model performance include precision, recall, accuracy, and F1-score as determined from the test sub-dataset. The results indicate that the RF- and SVM-based BSR models achieved prediction accuracies of approximately 0.64–0.94 for different classification schemes applied to the test sub-dataset. Additionally, the precision, recall, and F1-score indicators showed satisfactory values with respect to the BSR levels with relatively large sample sizes. The RF-based models had a lower tendency for overfitting compared to the SVM-based models. The performance of the BSR models was influenced by the quantity of total datasets, the classification schemes, and imbalanced data. Overall, the RF- and SVM-based BSR models can improve the evaluation efficiency of earthquake-damaged buildings in mountainous areas.

Prediction of Mechanical Properties and Failure Behavior of Nano-Alumina Reinforced Poly(Lactic) Acid Composite for Fused Deposition Modeling 3D Printing via Finite Element Analysis

Mechanical properties and failure behavior of 3D printed poly (lactic) acid (PLA) reinforced with varying loadings of nano-alumina (Al2O3) (0, 2.5, 5.0, and 10.0 wt.%) were investigated through simulation using a finite element analysis (FEA)-based software. Tensile test specimens were 3D printed via fused deposition modeling (FDM) technique and underwent actual testing. The mechanical properties determined were then used as parameters for the FEA simulation to achieve prediction accuracy. Specifically, this study utilized MSC Patran and Nastran software to simulate the tensile test on the modeled test specimen with tetrahedron mesh. The finite element model was verified by comparing the simulated values with the results of actual experimental testing. Upon calculation, the average percentage differences for the tensile strength, elastic modulus, and displacement were 5.86%, 12.07%, and 10.57%, respectively. Although percentage differences were obtained, using FEA as an initial analysis for the prediction of mechanical properties and failure behavior could serve as a solution for better design and materials optimization.

Acute coronary event (ACE) prediction following breast radiotherapy by features extracted from 3D CT, dose, and cardiac structures.

Heart toxicity, such as major acute coronary events (ACE), following breast radiation therapy (RT) is of utmost concern. Thus, many studies have been investigating the effect of mean heart dose (MHD) and dose received in heart sub-structures on toxicity. Most studies focused on the dose thresholds in the heart and its sub-structures, while few studies adopted such computational methods as deep neural networks (DNN) and radiomics. This work aims to construct a feature-driven predictive model for ACE after breast RT. A recently proposed two-step predictive model that extracts a number of features from a deep auto-segmentation network and processes the selected features for prediction was adopted. This work refined the auto-segmenting network and feature processing algorithms to enhance performance in cardiac toxicity prediction. In the predictive model, the deep convolutional neural network (CNN) extracted features from 3D computed tomography (CT) images and dose distributions in three automatically segmented heart sub-structures, including the left anterior descending artery (LAD), right coronary artery (RCA), and left ventricle (LV). The optimal feature processing workflow for the extracted features was explored to enhance the prediction accuracy. The regions associated with toxicity were visualized using a class activation map (CAM)-based technique. Our proposed model was validated against a conventional DNN (convolutional and fully connected layers) and radiomics with a patient cohort of 84 cases, including 29 and 55 patient cases with and without ACE. Of the entire 84 cases, 12 randomly chosen cases (5 toxicity and 7 non-toxicity cases) were set aside for independent test, and the remaining 72 cases were applied to 4-fold stratified cross-validation. Our predictive model outperformed the conventional DNN by 38% and 10% and radiomics-based predictive models by 9% and 10% in AUC for 4-fold cross-validations and independent test, respectively. The degree of enhancement was greater when incorporating dose information and heart sub-structures into feature extraction. The model whose inputs were CT, dose, and three sub-structures (LV, LAD, and RCA) reached 96% prediction accuracy on average and 0.94 area under the curve (AUC) on average in the cross-validation, and also achieved prediction accuracy of 83% and AUC of 0.83 in the independent test. On 10 correctly predicted cases out of 12 for the independent test, the activation maps implied that for cases of ACE toxicity, the higher intensity was more likely to be observed inside the LV. The proposed model characterized by modifications in model input with dose distributions and cardiac sub-structures, and serial processing of feature extraction and feature selection techniques can improve the predictive performance in ACE following breast RT.

Semiquantitative Fingerprinting Based on Pseudotargeted Metabolomics and Deep Learning for the Identification of Listeria monocytogenes and Its Major Serotypes.

The rapid identification of pathogenic microorganism serotypes is still a bottleneck problem to be solved urgently. Compared with proteomics technology, metabolomics technology is directly related to phenotypes and has higher specificity in identifying pathogenic microorganism serotypes. Our study combines pseudotargeted metabolomics with deep learning techniques to obtain a new deep semiquantitative fingerprinting method for Listeria monocytogenes identification at the serotype levels. We prescreened 396 features with orthogonal partial least-squares discrimination analysis (OPLS-DA), and 200 features were selected for deep learning model building. A residual learning framework for L. monocytogenes identification was established. There were 256 convolutional filters in the initial convolution layer, and each hidden layer contained 128 filters. The total depth included seven layers, consisting of an initial convolution layer, a residual layer, and two final fully connected classification layers, with each residual layer containing four convolutional layers. In addition, transfer learning was used to predict new isolates that did not participate in model training to verify the method's feasibility. Finally, we achieved prediction accuracies of L. monocytogenes at the serotype level exceeding 99%. The prediction accuracy of the new strain validation set was greater than 97%, further demonstrating the feasibility of this method. Therefore, this technology will be a powerful tool for the rapid and accurate identification of pathogens.

Classification of Tree Species in the Process of Timber-Harvesting Operations Using Machine-Learning Methods

This article presents the constraining factors that limit the increase in the efficiency of logging production by modern multi-operation machines operating on the Scandinavian cut-to-length technology in the felling phase, namely the selection and registration of wood species. The factors for creating a complete architecture of a fully connected neural network (NN) are given. The dependence of the prediction accuracy of a fully connected NN on a test sample on the size of the training dataset, and an image of the dependence of the prediction accuracy on the number of trees in the random forest method for image classification is shown. For a fully connected NN, a sufficient number of images and a test sample size were established for training, using tree-trunk breed-class labels as target values. A selected list of trees was given, with the size of the training sample of images presenting a problem for the classification of tree trunks using the random forest method. The aim was the discovery of the optimal number of trees necessary to achieve prediction accuracy.

Multiparameter prediction model of immune checkpoint inhibitors combined with chemotherapy for non-small cell lung cancer based on support vector machine learning

The reliable predictive markers to identify which patients with advanced non-small cell lung cancer tumors (NSCLC) will achieve durable clinical benefit (DCB) for chemo-immunotherapy are needed. In this retrospective study, we collected radiomics and clinical signatures from 94 patients with advanced NSCLC treated with anti-PD-1/PD-L1 combined with chemotherapy from January 1, 2018 to May 31, 2022. Radiomics variables were extracted from pretreatment CT and selected by Spearman correlation coefficients and clinical features by Logistics regression analysis. We performed effective diagnostic algorithms principal components analysis (PCA) and support vector machine (SVM) to develop an early classification model among DCB and non-durable benefit (NDB) groups. A total of 26 radiomics features and 6 clinical features were selected, and then principal component analysis was used to obtain 6 principal components for SVM building. RC-SVM achieved prediction accuracy with AUC of 0.91 (95% CI 0.87–0.94) in the training set, 0.73 (95% CI 0.61–0.85) in the cross-validation set, 0.84 (95% CI 0.80–0.89) in the external validation set. The new method of RC-SVM model based on radiomics-clinical signatures provides a significant additive value on response prediction in patients with NSCLC preceding chemo-immunotherapy.

Physics-informed machine learning for MPC: Application to a batch crystallization process

This work presents a framework for developing physics-informed recurrent neural network (PIRNN) models and PIRNN-based predictive control schemes for batch crystallization processes. The population balance model of the aspirin crystallization process is first developed to describe the formation of crystals through nucleation and growth. Then, the PIRNN modeling scheme is introduced to integrate observational data and mechanistic models for the development of machine learning models. Additionally, the physical constraints on process states are embedded in the machine learning models to prevent physically unreasonable predictions. Subsequently, the PIRNN model that captures the dynamic behavior of the batch crystallization process is utilized in the design of model predictive controller that optimizes the operation of the crystallizer. Through open-loop and closed-loop simulations, it is demonstrated that the PIRNN models using less training data achieve prediction accuracy and closed-loop performance comparable to the purely data-driven model.

Development and validation of ischemic heart disease and stroke prognostic models using large-scale real-world data from Japan.

Previous cardiovascular risk prediction models in Japan have utilized prospective cohort studies with concise data. As the health information including health check-up records and administrative claims becomes digitalized and publicly available, application of large datasets based on such real-world data can achieve prediction accuracy and support social implementation of cardiovascular disease risk prediction models in preventive and clinical practice. In this study, classical regression and machine learning methods were explored to develop ischemic heart disease (IHD) and stroke prognostic models using real-world data. IQVIA Japan Claims Database was searched to include 691,160 individuals (predominantly corporate employees and their families working in secondary and tertiary industries) with at least one annual health check-up record during the identification period (April 2013-December 2018). The primary outcome of the study was the first recorded IHD or stroke event. Predictors were annual health check-up records at the index year-month, comprising demographic characteristics, laboratory tests, and questionnaire features. Four prediction models (Cox, Elnet-Cox, XGBoost, and Ensemble) were assessed in the present study to develop a cardiovascular disease risk prediction model for Japan. The analysis cohort consisted of 572,971 invididuals. All prediction models showed similarly good performance. The Harrell's C-index was close to 0.9 for all IHD models, and above 0.7 for stroke models. In IHD models, age, sex, high-density lipoprotein, low-density lipoprotein, cholesterol, and systolic blood pressure had higher importance, while in stroke models systolic blood pressure and age had higher importance. Our study analyzed classical regression and machine learning algorithms to develop cardiovascular disease risk prediction models for IHD and stroke in Japan that can be applied to practical use in a large population with predictive accuracy.

Machine Learning-Based MPC of Batch Crystallization Process Using Physics-Informed RNNs

This work presents a framework for developing physics-informed recurrent neural network (PIRNN) models and PIRNN-based predictive control schemes for batch crystallization processes. The population balance model of aspirin crystallization process is first developed to describe the formation of crystals through nucleation and growth. Then, the PIRNN modeling scheme is introduced to integrate observational data and mechanistic models for the development of machine learning models. Subsequently, the PIRNN model that captures the dynamic behavior of the batch crystallization process is utilized in the design of model predictive control. Through open-loop and closed-loop simulations, it is demonstrated that the PIRNN models using less training data achieve prediction accuracy and closed-loop performance comparable to the purely data-driven model.

Graph Neural Networks for High-Level Synthesis Design Space Exploration

High-level Synthesis (HLS) Design-Space Exploration (DSE) aims at identifying Pareto-optimal synthesis configurations whose exhaustive search is unfeasible due to the design-space dimensionality and the prohibitive computational cost of the synthesis process. Within this framework, we address the design automation problem by proposing graph neural networks that jointly predict acceleration performance and hardware costs of a synthesized behavioral specification given optimization directives. Learned models can be used to rapidly approach the Pareto curve by guiding the DSE, taking into account performance and cost estimates. The proposed method outperforms traditional HLS-driven DSE approaches, by accounting for the arbitrary length of computer programs and the invariant properties of the input. We propose a novel hybrid control and dataflow graph representation that enables training the graph neural network on specifications of different hardware accelerators. Our approach achieves prediction accuracy comparable with that of state-of-the-art simulators without having access to analytical models of the HLS compiler. Finally, the learned representation can be exploited for DSE in unexplored configuration spaces by fine-tuning on a small number of samples from the new target domain. The outcome of the empirical evaluation of this transfer learning shows strong results against state-of-the-art baselines in relevant benchmarks.