Transfer Learning Research Articles

Residual network improves the prediction accuracy of genomic selection.

Genetic improvement of complex traits in animal and plant breeding depends on the efficient and accurate estimation of breeding values. Deep learning methods have been shown to be not superior over traditional genomic selection (GS) methods, partially due to the degradation problem (i.e. with the increase of the model depth, the performance of the deeper model deteriorates). Since the deep learning method residual network (ResNet) is designed to solve gradient degradation, we examined its performance and factors related to its prediction accuracy in GS. Here we compared the prediction accuracy of conventional genomic best linear unbiased prediction, Bayesian methods (BayesA, BayesB, BayesC, and Bayesian Lasso), and two deep learning methods, convolutional neural network and ResNet, on three datasets (wheat, simulated and real pig data). ResNet outperformed other methods in both Pearson's correlation coefficient (PCC) and mean squared error (MSE) on the wheat and simulated data. For the pig backfat depth trait, ResNet still had the lowest MSE, whereas Bayesian Lasso had the highest PCC. We further clustered the pig data into four groups and, on one separated group, ResNet had the highest prediction accuracy (both PCC and MSE). Transfer learning was adopted and capable of enhancing the performance of both convolutional neural network and ResNet. Taken together, our findings indicate that ResNet could improve GS prediction accuracy, affected potentially by factors such as the genetic architecture of complex traits, data volume, and heterogeneity.

Transfer learning enables identification of multiple types of RNA modifications using nanopore direct RNA sequencing

Nanopore direct RNA sequencing (DRS) has emerged as a powerful tool for RNA modification identification. However, concurrently detecting multiple types of modifications in a single DRS sample remains a challenge. Here, we develop TandemMod, a transferable deep learning framework capable of detecting multiple types of RNA modifications in single DRS data. To train high-performance TandemMod models, we generate in vitro epitranscriptome datasets from cDNA libraries, containing thousands of transcripts labeled with various types of RNA modifications. We validate the performance of TandemMod on both in vitro transcripts and in vivo human cell lines, confirming its high accuracy for profiling m6A and m5C modification sites. Furthermore, we perform transfer learning for identifying other modifications such as m7G, Ψ, and inosine, significantly reducing training data size and running time without compromising performance. Finally, we apply TandemMod to identify 3 types of RNA modifications in rice grown in different environments, demonstrating its applicability across species and conditions. In summary, we provide a resource with ground-truth labels that can serve as benchmark datasets for nanopore-based modification identification methods, and TandemMod for identifying diverse RNA modifications using a single DRS sample.

Speech Recognition for Air Traffic Control Utilizing a Multi-Head State-Space Model and Transfer Learning

Speech Recognition for Air Traffic Control Utilizing a Multi-Head State-Space Model and Transfer Learning

Diagnosis of cervical lymphoma using a YOLO-v7-based model with transfer learning

To investigate the ability of an auxiliary diagnostic model based on the YOLO-v7-based model in the classification of cervical lymphadenopathy images and compare its performance against qualitative visual evaluation by experienced radiologists. Three types of lymph nodes were sampled randomly but not uniformly. The dataset was randomly divided into for training, validation, and testing. The model was constructed with PyTorch. It was trained and weighting parameters were tuned on the validation set. Diagnostic performance was compared with that of the radiologists on the testing set. The mAP of the model was 96.4% at the 50% intersection-over-union threshold. The accuracy values of it were 0.962 for benign lymph nodes, 0.982 for lymphomas, and 0.960 for metastatic lymph nodes. The precision values of it were 0.928 for benign lymph nodes, 0.975 for lymphomas, and 0.927 for metastatic lymph nodes. The accuracy values of radiologists were 0.659 for benign lymph nodes, 0.836 for lymphomas, and 0.580 for metastatic lymph nodes. The precision values of radiologists were 0.478 for benign lymph nodes, 0.329 for lymphomas, and 0.596 for metastatic lymph nodes. The model effectively classifies lymphadenopathies from ultrasound images and outperforms qualitative visual evaluation by experienced radiologists in differential diagnosis.

DLEmotion: Deep learning-based emotion classification using visual attention location information

The correct recognition of each person's emotion can be an important step in knowing mental states and reactions. Because of the complexities of human personality, this is a difficult task and providing a method that can classify emotions with high accuracy can be useful. The purpose of doing this work is to provide a suitable method for recognizing emotions in human subjects. In other words, in this research the aim is to analyze eye tracking data using deep learning approaches and for this purpose, deep neural networks are used to classify fixations in different emotional states. The eye data of 10 women and 7 men who watched 79 emotional images are first collected. In this way, for each fixation point obtained from the subjects, a patch of 80 x 80 pixels was selected from the images and we used these patches of the obtained image as the input to the deep neural networks. The images shown to the subjects have three basic dimensions of emotion, i.e. Valence, Arousal and Dominance. Data classification was done by transfer learning and using Xception, InceptionV3 and DenseNet201 deep neural networks whose weights were trained on ImageNet datasets. We used these three neural networks in our work because they have relatively equal number of parameters and will help us in better analysis of the obtained results. In general, the results show that the points obtained from subjects' fixations location on images with three basic dimensions of emotion can be used to classify emotions. We will also see that changing the size of the patches obtained from the subject's fixation will not have much effect on the test results, which shows that deep neural networks are a powerful and accurate tool for processing people's eye fixation points.

Mechanical strength recognition and classification of thermal protective fabric images after thermal aging based on deep learning

Objectives. Currently, numerous studies have focused on testing or modeling to evaluate the safe service life of thermal protective clothing after thermal aging, reducing the risk to occupational personnel. However, testing will render the garment unsuitable for subsequent use and a series of input parameters for modeling are not readily available. In this study, a novel image recognition strategy was proposed to discriminate the mechanical strength of thermal protective fabric after thermal aging based on transfer learning. Methods. Data augmentation was used to overcome the shortcoming of insufficient training samples. Four pre-trained models were used to explore their performance in three sample classification modes. Results. The experimental results show that the VGG-19 model achieves the best performance in the three-classification mode (accuracy = 91%). The model was more accurate in identifying fabric samples in the early and late stages of strength decline. For fabric samples in the middle stage of strength decline, the three-classification mode was better than the four-classification and six-classification modes. Conclusions. The findings provide novel insights into the image-based mechanical strength evaluation of thermal protective fabrics after aging.

Transfer Learning for Prognostics and Health Management: Advances, Challenges, and Opportunities

Transfer Learning for Prognostics and Health Management: Advances, Challenges, and Opportunities

Cross-modality transfer learning with knowledge infusion for diabetic retinopathy grading

Cross-modality transfer learning with knowledge infusion for diabetic retinopathy grading

Optimizing Breast Cancer Detection: Deep Transfer Learning Empowered by SVM Classifiers

Optimizing Breast Cancer Detection: Deep Transfer Learning Empowered by SVM Classifiers

Efficient Classification on Remote Sensing Image Using Transfer Learning

With a rapid development in aerial technology, applications of Remote Sensing Images (RSI) have become more diverse. Remote sensing image classification plays a crucial role in analyzing and interpreting Earth observation data for various applications, such as land cover mapping, environmental monitoring, and urban planning. However, accurately classifying remote sensing images poses significant challenges due to their complex spatial and spectral characteristics. In recent years, transfer learning has emerged as a promising technique to improve the classification accuracy by leveraging the knowledge learned from pre-trained models on large-scale datasets. The proposed model explores different transfer learning strategies employed in remote sensing image classification, including fine-tuning, feature extraction, and domain adaptation. It discusses popular pre-trained models, such as VGG16, VGG19, and Inceptionv3, and their applicability to remote sensing datasets. The advantages and limitations of each strategy are analyzed, providing insights into their suitability for various remote sensing applications. A comparative study is done on all these techniques to evaluate the performance measures like Accuracy and Loss.

A multifaceted suite of metrics for comparative myoelectric prosthesis controller research.

Upper limb robotic (myoelectric) prostheses are technologically advanced, but challenging to use. In response, substantial research is being done to develop person-specific prosthesis controllers that can predict a user's intended movements. Most studies that test and compare new controllers rely on simple assessment measures such as task scores (e.g., number of objects moved across a barrier) or duration-based measures (e.g., overall task completion time). These assessment measures, however, fail to capture valuable details about: the quality of device arm movements; whether these movements match users' intentions; the timing of specific wrist and hand control functions; and users' opinions regarding overall device reliability and controller training requirements. In this work, we present a comprehensive and novel suite of myoelectric prosthesis control evaluation metrics that better facilitates analysis of device movement details-spanning measures of task performance, control characteristics, and user experience. As a case example of their use and research viability, we applied these metrics in real-time control experimentation. Here, eight participants without upper limb impairment compared device control offered by a deep learning-based controller (recurrent convolutional neural network-based classification with transfer learning, or RCNN-TL) to that of a commonly used controller (linear discriminant analysis, or LDA). The participants wore a simulated prosthesis and performed complex functional tasks across multiple limb positions. Analysis resulting from our suite of metrics identified 16 instances of a user-facing problem known as the "limb position effect". We determined that RCNN-TL performed the same as or significantly better than LDA in four such problem instances. We also confirmed that transfer learning can minimize user training burden. Overall, this study contributes a multifaceted new suite of control evaluation metrics, along with a guide to their application, for use in research and testing of myoelectric controllers today, and potentially for use in broader rehabilitation technologies of the future.

Spoilt for Choice

Constant working conditions are outdated. Consequently, employees are confronted with the challenge of continuing vocational training. In this context, the question is how to select the appropriate trainer or training institute. In recent years, learning transfer research has focussed on researching the determinants of learning transfer. Baldwin and Ford's basic learning transfer model forms the starting point for this. This model differentiates between training input, training output and conditions of transfer. Thus, trainee characteristics, training design and work environment are determinants of training input. The author examines existing standards for training design and trainer competences in practice. In this regard, the author focusses on the question of which are the decisive criteria for the final decision in favour of a trainer or a training institute. No conclusive answer can be given to this question, as no studies are available. Consequently, a future field of research is evident here.

Quantifying the Hardness of Bioactivity Prediction Tasks for Transfer Learning.

Today, machine learning methods are widely employed in drug discovery. However, the chronic lack of data continues to hamper their further development, validation, and application. Several modern strategies aim to mitigate the challenges associated with data scarcity by learning from data on related tasks. These knowledge-sharing approaches encompass transfer learning, multitask learning, and meta-learning. A key question remaining to be answered for these approaches is about the extent to which their performance can benefit from the relatedness of available source (training) tasks; in other words, how difficult ("hard") a test task is to a model, given the available source tasks. This study introduces a new method for quantifying and predicting the hardness of a bioactivity prediction task based on its relation to the available training tasks. The approach involves the generation of protein and chemical representations and the calculation of distances between the bioactivity prediction task and the available training tasks. In the example of meta-learning on the FS-Mol data set, we demonstrate that the proposed task hardness metric is inversely correlated with performance (Pearson's correlation coefficient r = -0.72). The metric will be useful in estimating the task-specific gain in performance that can be achieved through meta-learning.

CNSMolGen: A Bidirectional Recurrent Neural Network-Based Generative Model for De Novo Central Nervous System Drug Design.

Central nervous system (CNS) drugs have had a significant impact on treating a wide range of neurodegenerative and psychiatric disorders. In recent years, deep learning-based generative models have shown great potential for accelerating drug discovery and improving efficacy. However, specific applications of these techniques in CNS drug discovery have not been widely reported. In this study, we developed the CNSMolGen model, which uses a framework of bidirectional recurrent neural networks (Bi-RNNs) for de novo molecular design of CNS drugs. Results showed that the pretrained model was able to generate more than 90% of completely new molecular structures, which possessed the properties of CNS drug molecules and were synthesizable. In addition, transfer learning was performed on small data sets with specific biological activities to evaluate the potential application of the model for CNS drug optimization. Here, we used drugs against the classical CNS disease target serotonin transporter (SERT) as a fine-tuned data set and generated a focused database against the target protein. The potential biological activities of the generated molecules were verified by using the physics-based induced-fit docking study. The success of this model demonstrates its potential in CNS drug design and optimization, which provides a new impetus for future CNS drug development.

Accurate Classification of Tunnel Lining Cracks Using Lightweight ShuffleNetV2-1.0-SE Model with DCGAN-Based Data Augmentation and Transfer Learning

Accurate Classification of Tunnel Lining Cracks Using Lightweight ShuffleNetV2-1.0-SE Model with DCGAN-Based Data Augmentation and Transfer Learning

Transfer-Learning-Enhanced Regression Generative Adversarial Networks for Optimal eVTOL Takeoff Trajectory Prediction

Electric vertical takeoff and landing (eVTOL) aircraft represent a crucial aviation technology to transform future transportation systems. The unique characteristics of eVTOL aircraft include reduced noise, low pollutant emission, efficient operating cost, and flexible maneuverability, which in the meantime pose critical challenges to advanced power retention techniques. Thus, optimal takeoff trajectory design is essential due to immense power demands during eVTOL takeoffs. Conventional design optimizations, however, adopt high-fidelity simulation models in an iterative manner resulting in a computationally intensive mechanism. In this work, we implement a surrogate-enabled inverse mapping optimization architecture, i.e., directly predicting optimal designs from design requirements (including flight conditions and design constraints). A trained inverse mapping surrogate performs real-time optimal eVTOL takeoff trajectory predictions with no need for running optimizations; however, one training sample requires one design optimization in this inverse mapping setup. The excessive training cost of inverse mapping and the characteristics of optimal eVTOL takeoff trajectories necessitate the development of the regression generative adversarial network (regGAN) surrogate. We propose to further enhance regGAN predictive performance through the transfer learning (TL) technique, creating a scheme termed regGAN-TL. In particular, the proposed regGAN-TL scheme leverages the generative adversarial network (GAN) architecture consisting of a generator network and a discriminator network, with a combined loss of the mean squared error (MSE) and binary cross-entropy (BC) losses, for regression tasks. In this work, the generator employs design requirements as input and produces optimal takeoff trajectory profiles, while the discriminator differentiates the generated profiles and real optimal profiles in the training set. The combined loss facilitates the generator training in the dual aspects: the MSE loss targets minimum differences between generated profiles and training counterparts, while the BC loss drives the generated profiles to share analogous patterns with the training set. We demonstrated the utility of regGAN-TL on optimal takeoff trajectory designs for the Airbus A3 Vahana and compared its performance against representative surrogates, including the multi-output Gaussian process, the conditional GAN, and the vanilla regGAN. Results showed that regGAN-TL reached the 99.5% generalization accuracy threshold with only 200 training samples while the best reference surrogate required 400 samples. The 50% reduction in training expense and reduced standard deviations of generalization accuracy achieved by regGAN-TL confirmed its outstanding predictive performance and broad engineering application potential.

Brain tumor X-ray images enhancement and classification using anisotropic diffusion filter and transfer learning models

Brain tumor X-ray images enhancement and classification using anisotropic diffusion filter and transfer learning models

Leveraging Transfer Learning and U-Nets Method for Improved Gap Filling in Himawari Sea Surface Temperature Data Adjacent to Taiwan

Satellite sea surface temperature (SST) images are valuable for various oceanic applications, including climate monitoring, ocean modeling, and marine ecology. However, cloud cover often obscures SST signals, creating gaps in the data that reduce resolution and hinder spatiotemporal analysis, particularly in the waters near Taiwan. Thus, gap-filling methods are crucial for reconstructing missing SST values to provide continuous and consistent data. This study introduces a gap-filling approach using the Double U-Net, a deep neural network model, pretrained on a diverse dataset of Level-4 SST images. These gap-free products are generated by blending satellite observations with numerical models and in situ measurements. The Double U-Net model excels in capturing SST dynamics and detailed spatial patterns, offering sharper representations of ocean current-induced SST patterns than the interpolated outputs of Data Interpolating Empirical Orthogonal Functions (DINEOFs). Comparative analysis with buoy observations shows the Double U-Net model’s enhanced accuracy, with better correlation results and lower error values across most study areas. By analyzing SST at five key locations near Taiwan, the research highlights the Double U-Net’s potential for high-resolution SST reconstruction, thus enhancing our understanding of ocean temperature dynamics. Based on this method, we can combine more high-resolution satellite data in the future to improve the data-filling model and apply it to marine geographic information science.

Detecting Coseismic Landslides in GEE Using Machine Learning Algorithms on Combined Optical and Radar Imagery

Landslides, resulting from disturbances in slope equilibrium, pose a significant threat to landscapes, infrastructure, and human life. Triggered by factors such as intense precipitation, seismic activities, or volcanic eruptions, these events can cause extensive damage and endanger nearby communities. A comprehensive understanding of landslide characteristics, including spatio-temporal patterns, dimensions, and morphology, is vital for effective landslide disaster management. Existing remote sensing approaches mostly use either optical or synthetic aperture radar sensors. Integrating information from both these types of sensors promises greater accuracy for identifying and locating landslides. This study proposes a novel approach, the ML-LaDeCORsat (Machine Learning-based coseismic Landslide Detection using Combined Optical and Radar Satellite Imagery), that integrates freely available Sentinel-1, Palsar-2, and Sentinel-2 imagery data in Google Earth Engine (GEE). The approach also integrates relevant spectral indices and suitable bands used in a machine learning-based classification of coseismic landslides. The approach includes a robust and reproducible training and validation strategy and allows one to choose between five classifiers (CART, Random Forest, GTB, SVM, and Naive Bayes). Using landslides from four different earthquake case studies, we demonstrate the superiority of our approach over existing solutions in coseismic landslide identification and localization, providing a GTB-based detection accuracy of 87–92%. ML-LaDeCORsat can be adapted to other landslide events (GEE script is provided). Transfer learning experiments proved that our model can be applied to other coseismic landslide events without the need for additional training data. Our novel approach therefore facilitates quick and reliable identification of coseismic landslides, highlighting its potential to contribute towards more effective disaster management.

Web-based diagnostic platform for microorganism-induced deterioration on paper-based cultural relics with iterative training from human feedback

PurposePaper-based artifacts hold significant cultural and social values. However, paper is intrinsically fragile to microorganisms, such as mold, due to its cellulose composition, which can serve as a microorganisms’ nutrient source. Mold not only can damage papers’ structural integrity and pose significant challenges to conservation works but also may subject individuals attending the contaminated artifacts to health risks. Current approaches for strain identification usually require extensive training, prolonged time for analysis, expensive operation costs, and higher risks of secondary damage due to sampling. Thus, in current conservation practices with mold-contaminated artifacts, little pre-screening or strain identification was performed before mold removal, and the cleaning techniques are usually broad-spectrum rather than strain-specific. With deep learning showing promising applications across various domains, this study investigated the feasibility of using a convolutional neural network (CNN) for fast in-situ recognition and classification of mold on paper.MethodsMolds were first non-invasively sampled from ancient Xuan Paper-based Chinese books from the Qing and Ming dynasties. Strains were identified using molecular biology methods and the four most prevalent strains were inoculated on Xuan paper to create mockups for image collection. Microscopic images of the molds as well as their stains situated on paper were collected using a compound microscope and commercial microscope lens for cell phone cameras, which were then used for training CNN models with a transfer learning scheme to perform the classification of mold. To enable involvement and contribution from the research community, a web interface that actuates the process while providing interactive features for users to learn about the information of the classified strain was constructed. Moreover, a feedback functionality in the web interface was embedded for catching potential classification errors, adding additional training images, or introducing new strains, all to refine the generalizability and robustness of the model.Results & ConclusionIn the study, we have constructed a suite of high-confidence classification CNN models for the diagnostic process for mold contamination in conservation. At the same time, a web interface was constructed that allows recurrently refining the model with human feedback through engaging the research community. Overall, the proposed framework opens new avenues for effective and timely identification of mold, thus enabling proactive and targeted mold remediation strategies in conservation.