An Introduction to Transfer Learning

Currently, numerous machine learning (ML) techniques are being applied in the field of renewable energy (RE). These techniques may not perform well if they do not have enough training data. Additionally, the main assumption in most of the ML algorithms is that the training and testing data are from the same feature space and have similar distributions. However, in many practical applications, this assumption is false. Recently, transfer learning (TL) has been introduced as a promising machine-learning framework to mitigate these issues by preparing extra-domain data so that knowledge may be transferred across domains. This learning technique improves performance and avoids the resource expensive collection and labeling of domain-centric datasets; furthermore, it saves computing resources that are needed for re-training new ML models from scratch. Lately, TL has drawn the attention of researchers in the field of RE in terms of forecasting and fault diagnosis tasks. Owing to the rapid progress of this technique, a comprehensive survey of the related advances in RE is needed to show the critical issues that have been solved and the challenges that remain unsolved. To the best of our knowledge, few or no comprehensive surveys have reviewed the applications of TL in the RE field, especially those pertaining to forecasting solar and wind power, load forecasting, and predicting failures in power systems. This survey fills this gap in RE classification and forecasting problems, and helps researchers and practitioners better understand the state of the art technology in the field while identifying areas for more focused study. In addition, this survey identifies the main issues and challenges of using TL for REs, and concludes with a discussion of future perspectives.

Active Learning is a machine learning and data mining technique that selects the most informative samples for labeling and uses them as training data; it is especially useful when there are large amount of unlabeled data and labeling them is expensive. Recently, batch-mode active learning, where a set of samples are selected concurrently for labeling, based on their collective merit, has attracted a lot of attention. The objective of batch-mode active learning is to select a set of informative samples so that a classifier learned on these samples has good generalization performance on the unlabeled data. Most of the existing batch-mode active learning methodologies try to achieve this by selecting samples based on certain criteria. In this article we propose a novel criterion which achieves good generalization performance of a classifier by specifically selecting a set of query samples that minimize the difference in distribution between the labeled and the unlabeled data, after annotation. We explicitly measure this difference based on all candidate subsets of the unlabeled data and select the best subset. The proposed objective is an NP-hard integer programming optimization problem. We provide two optimization techniques to solve this problem. In the first one, the problem is transformed into a convex quadratic programming problem and in the second method the problem is transformed into a linear programming problem. Our empirical studies using publicly available UCI datasets and two biomedical image databases demonstrate the effectiveness of the proposed approach in comparison with the state-of-the-art batch-mode active learning methods. We also present two extensions of the proposed approach, which incorporate uncertainty of the predicted labels of the unlabeled data and transfer learning in the proposed formulation. In addition, we present a joint optimization framework for performing both transfer and active learning simultaneously unlike the existing approaches of learning in two separate stages, that is, typically, transfer learning followed by active learning. We specifically minimize a common objective of reducing distribution difference between the domain adapted source, the queried and labeled samples and the rest of the unlabeled target domain data. Our empirical studies on two biomedical image databases and on a publicly available 20 Newsgroups dataset show that incorporation of uncertainty information and transfer learning further improves the performance of the proposed active learning based classifier. Our empirical studies also show that the proposed transfer-active method based on the joint optimization framework performs significantly better than a framework which implements transfer and active learning in two separate stages.

A dose deposition matrix (DDM) prediction method using several voxel features and a machine learning (ML) approach is proposed for plan optimization in radiation therapy. Head and lung cases with the inhomogeneous medium are used as training and testing data. The prediction model is a cascade forward backprop neural network where the input is the features of the voxel, including 1) voxel to body surface distance along the beamlet axis, 2) voxel to beamlet axis distance, 3) voxel density, 4) heterogeneity corrected voxel to body surface distance, 5) heterogeneity corrected voxel to beamlet axis, and (6) the dose of voxel obtained from the pencil beam (PB) algorithm. The output is the predicted voxel dose corresponding to a beamlet. The predicted DDM was used for plan optimization (ML method) and compared with the dose of MC-based plan optimization (MC method) and the dose of pencil beam-based plan optimization (PB method). The mean absolute error (MAE) value was calculated for full volume relative to the dose of the MC method to evaluate the overall dose performance of the final plan. For patient with head tumor, the ML method achieves MAE value 0.49 × 10-4 and PB has MAE 1.86 × 10-4. For patient with lung tumor, the ML method has MAE 1.42 × 10-4 and PB has MAE 3.72 × 10-4. The maximum percentage difference in PTV dose coverage (D98) between ML and MC methods is no more than 1.2% for patient with head tumor, while the difference is larger than 10% using the PB method. For patient with lung tumor, the maximum percentage difference in PTV dose coverage (D98) between ML and MC methods is no more than 2.1%, while the difference is larger than 16% using the PB method. In this work, a reliable DDM prediction method is established for plan optimization by applying several voxel features and the ML approach. The results show that the ML method based on voxel features can obtain plans comparable to the MC method and is better than the PB method in achieving accurate dose to the patient, which is helpful for rapid plan optimization and accurate dose calculation. Establishment of a new machine learning method based on the relationship between the voxel and beamlet features for dose deposition matrix prediction in radiation therapy.

Passive sensing applications are limited by their inability to determine who is using a device, a critical concern in child mobile device use research, where devices are often shared between siblings or between a child and their parent. Our previous work leveraged behavioral biometrics to identify a target child user; however, it is unknown what type of training data is necessary for optimal model performance. This study evaluated model performance across different characteristics of training data. Thirty-six children (11.3 ± 0.9 years, 56% female) self-selected a video or a game on iPads for 10 min while laying and for another 5 min while sitting. The SensorLog application captured iPad accelerometer and gyroscope data while the child interacted with the device. Machine learning algorithms including Neural Network (NN), Random Forest (RF), k-Nearest Neighbors (k-NN), and SwipeFormer were applied to determine the most important aspects of training data to optimize model performance. The aspects of training data evaluated included (1) varying the length (i.e., seconds of training data), (2) varying the user position (i.e., sitting, laying), and (3) varying the time proximity between training and testing data. F1 score was used to evaluate model performance. The SwipeFormer F1 scores were lowest when the training data was further from the test data (0 when training data was 11 min away from test data) and highest when training data was close to test data (0.91 when training data was the minute preceding test data). The SwipeFormer F1 scores were highest when predicting the user laying from laying (0.97) and sitting from sitting (0.94), and lowest when predicting the user sitting from laying (0) and laying from sitting (0). The length of training data had little impact on performance, with a SwipeFormer F1 score of 0.91 when training on one minute of data and a SwipeFormer F1 score of 0.94 when training on twelve minutes of data. Because researchers would likely be predicting users at different timepoints than their training data, research should focus on improving model performance for identifying users independent of time proximity for training and test data.

Spatial application of Random Forest models for fine-scale coastal vegetation classification using object based analysis of aerial orthophoto and DEM data

Financial institutions have a large amount of data on their borrowers, which can be used to predict the probability of borrowers defaulting their loan or not. Some of the models that have been used to predict individual loan defaults include linear discriminant analysis models and extreme value theory models. These models are parametric in nature since they assume that the response being investigated takes a particular functional form. However, there is a possibility that the functional form used to estimate the response is very different from the actual functional form of the response. The purpose of this research was to analyze individual loan defaults in Kenya using the logistic regression model. The data used in this study was obtained from equity bank of Kenya for the period between 2006 to 2016. A random sample of 1000 loan applicants whose loans had been approved by equity bank of Kenya during this period was obtained. Data obtained was on the credit history, purpose of the loan, loan amount, nature of the saving account, employment status, sex of the applicant, age of the applicant, security used when acquiring the loan and the area of residence of the applicant (rural or urban). This study employed a quantitative research design, it deals with individual loans defaults as group characteristics of a borrower. The data was pre-processed by seeding using R- Software and then split into training dataset and test data set. The train data was used to train the logistic regression model by employing Supervised machine learning approach. The R-statistical software was used for the analysis of the data. The test data set was used to do cross-validation of the developed logistic model which later was used for analysis prediction of individual loan defaults. This study focused on the analysis of individual loan defaults in Kenya using the logistic regression model in Machine learning. The logistic regression model predicted 303 defaults from train data set, 122 non-defaults and misclassified loans were 56 and 69. The model had an accuracy of 0.7727 with the train data and 0.7333 with the test data. The logistic regression model showed a precision of 0.8440 and 0.8244 with the train and test data respectively. The performance of the model with both the train and test data was illustrated using a plot of train errors and test errors against sample size on the same axes. The plot showed that the performance of the model increases with an increase in sample size. The study recommended the use of logistic regression in conjunction with supervised machine learning approach in loan default prediction in financial institutions and also more research should be carried out on ensemble methods of loan defaults prediction in order to increase the prediction accuracy.

Machine learning (ML) techniques are increasingly applied to decision-making and control problems in Cyber-Physical Systems among which many are safety-critical, e.g., chemical plants, robotics, autonomous vehicles. Despite the significant benefits brought by ML techniques, they also raise additional safety issues because 1) most expressive and powerful ML models are not transparent and behave as a black box and 2) the training data which plays a crucial role in ML safety is usually incomplete. An important technique to achieve safety for ML models is "Safe Fail", i.e., a model selects a reject option and applies the backup solution, a traditional controller or a human operator for example, when it has low confidence in a prediction. Data-driven models produced by ML algorithms learn from training data, and hence they are only as good as the examples they have learnt. As pointed in [17], ML models work well in the "training space" (i.e., feature space with sufficient training data), but they could not extrapolate beyond the training space. As observed in many previous studies, a feature space that lacks training data generally has a much higher error rate than the one that contains sufficient training samples [31]. Therefore, it is essential to identify the training space and avoid extrapolating beyond the training space. In this paper, we propose an efficient Feature Space Partitioning Tree (FSPT) to address this problem. Using experiments, we also show that, a strong relationship exists between model performance and FSPT score.

SummaryAccurate prediction of the solubility of chemical substances in solvents remains a challenge. The sparsity of high-quality solubility data is recognized as the biggest hurdle in the development of robust data-driven methods for practical use. Nonetheless, the effects of the quality and quantity of data on aqueous solubility predictions have not yet been scrutinized. In this study, the roles of the size and the quality of data sets on the performances of the solubility prediction models are unraveled, and the concepts of actual and observed performances are introduced. In an effort to curtail the gap between actual and observed performances, a quality-oriented data selection method, which evaluates the quality of data and extracts the most accurate part of it through statistical validation, is designed. Applying this method on the largest publicly available solubility database and using a consensus machine learning approach, a top-performing solubility prediction model is achieved.

In the field of supervised machine learning, a transfer learning environment is defined as the training data having different distribution characteristics than the testing data. This is due to the lack of available labeled data for the domain of interest, which prompts an alternate domain to be used as the training data. Because there is insufficient labeled data from the domain of interest, validation techniques cannot be reliably used for the algorithm selection process in a transfer learning environment. A transfer learning algorithm is typically comprised of a domain adaptation step followed by a learning step. The learning step is usually implemented using a traditional machine learning algorithm. In this paper, we examine and analyze the impact that the traditional machine learning algorithm (the learning step) has on the overall performance of a transfer learning algorithm. Using the transfer learning test framework, we test five state-of-the-art transfer learning algorithms coupled with seven different traditional learning algorithms for a total of 35 unique transfer learning algorithms. For our experiment, no labeled data from the domain of interest is available for the training process. Since validation techniques cannot be reliably used for the algorithm selection process in a transfer learning environment, it is important for machine learning researchers and practitioners to understand the impact of a traditional machine learner on the overall performance of a transfer learning algorithm.

Purpose: The purpose of this study is to address the limited use of transfer learning techniques in radio frequency machine learning and to propose a customized taxonomy for radio frequency applications. The aim is to enable performance gains, improved generalization, and cost-effective training data solutions in this specific domain.
 Methodology: The research design employed in this study involves a comprehensive review of existing literature on transfer learning in radio frequency machine learning. The researchers collected relevant papers from reputable sources and analyzed them to identify patterns, trends, and insights. The method of data collection primarily relied on examining and synthesizing existing literature. Data analysis involved identifying key findings and developing a customized taxonomy for radio frequency applications.
 Findings: The study's findings highlight the limited utilization of transfer learning techniques in radio frequency machine learning. While transfer learning has shown significant performance improvements in computer vision and natural language processing, its potential in the wireless communications domain has yet to be fully explored. The customized taxonomy proposed in this study provides a consistent framework for analyzing and comparing existing and future efforts in this field.
 Recommendations: Based on the findings, the study recommends further research and experimentation to explore the potential of transfer learning techniques in radio frequency machine learning. This includes investigating performance gains, improving generalization capabilities, and addressing concerns related to training data costs. Additionally, collaborations between researchers and practitioners in the field are encouraged to facilitate knowledge exchange and foster innovation. Practice: To practitioners in the field of radio frequency machine learning, this study emphasizes the potential benefits of incorporating transfer learning techniques. It encourages practitioners to explore the application of transfer learning in their specific domain, leveraging prior knowledge to enhance performance and address training data challenges. It also highlights the importance of staying informed about the latest developments and collaborating with experts in the field. Policy: To policy makers, the study underscores the need for supportive policies that promote research and development in radio frequency machine learning. It recommends creating an environment that fosters innovation, encourages collaborations between academia and industry, and provides resources and incentives for further exploration of transfer learning techniques. Policy makers should consider the potential impact of transfer learning on the wireless communications industry and support initiatives that enhance its adoption and implementation.

Forecasting streamflow in stream basin systems plays a crucial role in facilitating effective urban planning to mitigate floods. In addition to employing intricate hydrological modeling systems, machine learning and statistical techniques offer an alternative means for streamflow forecasts. Nonetheless, the precision and dependability of these methods are subjects of paramount importance. This study rigorously investigates the effectiveness of three distinct machine learning techniques and two statistical approaches when applied to streamflow data from the Göksu Stream in the Marmara Region of Turkey, spanning from 1984 to 2022. Through a comparative analysis of these methodologies, this examination aims to contribute innovative advancements to the existing methodologies used in the prediction of streamflow data. The methodologies employed include machine learning methods such as Extreme Gradient Boosting (XGBoost), Random Forest (RF), and Support Vector Machine (SVM) and statistical methods such as Simple Exponential Smoothing (SES) and Autoregressive Integrated Moving Average (ARIMA) model. In the study, 444 data points between 1984 and 2020 were used as training data, and the remaining data points for the period 2021–2022 were used for streamflow forecasting in the test validation period. The results were evaluated using various metrics, such as the correlation coefficient (r), mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE), coefficient of determination (R2), and Nash–Sutcliffe efficiency (NSE). Upon analyzing the results, it was found that the model generated using the XGBoost algorithm outperformed other machine learning and statistical techniques. Consequently, the models implemented in this study demonstrate a high level of accuracy in predicting potential streamflow in the river basin system.

In the last few decades, statistical methods and machine learning (ML) algorithms have become efficient in medical decision-making. Coronary artery disease (CAD) is a common type of cardiovascular disease that causes many deaths each year. In this study, two CAD datasets from different countries (TRNC and Iran) are tested to understand the classification efficiency of different supervised machine learning algorithms. The Z-Alizadeh Sani dataset contained 303 individuals (216 patient, 87 control), while the Near East University (NEU) Hospital dataset contained 475 individuals (305 patients, 170 control). This study was conducted in three stages: (1) Each dataset, as well as their merged version, was subject to review separately with a random sampling method to obtain train-test subsets. (2) The NEU Hospital dataset was assigned as the training data, while the Z-Alizadeh Sani dataset was the test data. (3) The Z-Alizadeh Sani dataset was assigned as the training data, while the NEU hospital dataset was the test data. Among all ML algorithms, the Random Forest showed successful results for its classification performance at each stage. The least successful ML method was kNN which underperformed at all pitches. Other methods, including logistic regression, have varying classification performances at every step.

Automated monitoring of the movements and behaviour of animals is a valuable research tool. Recently, machine learning tools were applied to many species to classify units of behaviour. For the monitoring of wild species, collecting enough data for training models might be problematic, thus we examine how machine learning models trained on one species can be applied to another closely related species with similar behavioural conformation. We contrast two ways to calculate accuracies, termed here as overall and threshold accuracy, because the field has yet to define solid standards for reporting and measuring classification performances. We measure 21 dogs and 7 wolves, and find that overall accuracies are between 51 and 60% for classifying 8 behaviours (lay, sit, stand, walk, trot, run, eat, drink) when training and testing data are from the same species and between 41 and 51% when training and testing is cross-species. We show that using data from dogs to predict the behaviour of wolves is feasible. We also show that optimising the model for overall accuracy leads to similar overall and threshold accuracies, while optimizing for threshold accuracy leads to threshold accuracies well above 80%, but yielding very low overall accuracies, often below the chance level. Moreover, we show that the most common method for dividing the data between training and testing data (random selection of test data) overestimates the accuracy of models when applied to data of new specimens. Consequently, we argue that for the most common goals of animal behaviour recognition overall accuracy should be the preferred metric. Considering, that often the goal is to collect movement data without other methods of observation, we argue that training data and testing data should be divided by individual and not randomly.

Recent progress in machine learning has been mainly due to the availability of large amounts of annotated data used for training complex models with deep architectures. Annotating this training data becomes burdensome and creates a major bottleneck in maintaining machine-learning databases. Moreover, these trained models fail to generalize to new categories or new varieties of the same categories. This is because new categories or new varieties have data distribution different from the training data distribution. To tackle these problems, this thesis proposes to develop a family of transfer-learning techniques that can deal with different training (source) and testing (target) distributions with the assumption that the availability of annotated data is limited in the testing domain. This is done by using the auxiliary data-abundant source domain from which useful knowledge is transferred that can be applied to data-scarce target domain. This transferable knowledge serves as a prior that biases target-domain predictions and prevents the target-domain model from overfitting. Specifically, we explore structural priors that encode relational knowledge between different data entities, which provides more informative bias than traditional priors. The choice of the structural prior depends on the information availability and the similarity between the two domains. Depending on the domain similarity and the information availability, we divide the transfer learning problem into four major categories and propose different structural priors to solve each of these sub-problems. This thesis first focuses on the unsupervised-domain-adaptation problem, where we propose to minimize domain discrepancy by transforming labeled source-domain data to be close to unlabeled target-domain data. For this problem, the categories remain the same across the two domains and hence we assume that the structural relationship between the source-domain samples is carried over to the target domain. Thus, graph or hyper-graph is constructed as the structural prior from both domains and a graph/hyper-graph matching formulation is used to transform samples in the source domain to be closer to samples in the target domain. An efficient optimization scheme is then proposed to tackle the time and memory inefficiencies associated with the matching problem. The few-shot learning problem is studied next, where we propose to transfer knowledge from source-domain categories containing abundantly labeled data to novel categories in the target domain that contains only few labeled data. The knowledge transfer biases the novel category predictions and prevents the model from overfitting. The knowledge is encoded using a neural-network-based prior that transforms a data sample to its corresponding class prototype. This neural network is trained from the source-domain data and applied to the target-domain data, where it transforms the few-shot samples to the novel-class prototypes for better recognition performance. The few-shot learning problem is then extended to the situation, where we do not have access to the source-domain data but only have access to the source-domain class prototypes. In this limited information setting, parametric neural-network-based priors would overfit to the source-class prototypes and hence we seek a non-parametric-based prior using manifolds. A piecewise linear manifold is used as a structural prior to fit the source-domain-class prototypes. This structure is extended to the target domain, where the novel-class prototypes are found by projecting the few-shot samples onto the manifold. Finally, the zero-shot learning problem is addressed, which is an extreme case of the few-shot learning problem where we do not have any labeled data in the target domain. However, we have high-level information for both the source and target domain categories in the form of semantic descriptors. We learn the relation between the sample space and the semantic space, using a regularized neural network so that classification of the novel categories can be carried out in a common representation space. This same neural network is then used in the target domain to relate the two spaces. In case we want to generate data for the novel categories in the target domain, we can use a constrained generative adversarial network instead of a traditional neural network. Thus, we use structural priors like graphs, neural networks and manifolds to relate various data entities like samples, prototypes and semantics for these different transfer learning sub-problems. We explore additional post-processing steps like pseudo-labeling, domain adaptation and calibration and enforce algorithmic and architectural constraints to further improve recognition performance. Experimental results on standard transfer learning image recognition datasets produced competitive results with respect to previous work. Further experimentation and analyses of these methods provided better understanding of machine learning as well.

Applying data mining techniques to classify patients with suspected hepatitis C virus infection