Multi-source Transfer Learning Research Articles

Research on sound quality of roller chain transmission system based on multi-source transfer learning

To establish the sound quality evaluation model of roller chain transmission system, we collect the running noise under different working conditions. After the noise samples are preprocessed, a group of experienced testers are organized to evaluate them subjectively. Mel frequency cepstral coefficient (MFCC) of each noise sample is calculated, and the MFCC feature map is used as an objective evaluation. Combining with the subjective and objective evaluation results of the roller chain system noise, we can get the original dataset of its sound quality research. However, the number of high-quality noise samples is relatively small. Based on the sound quality research of various chain transmission systems, a novel method called multi-source transfer learning convolutional neural network (MSTL-CNN) is proposed. By transferring knowledge from multiple source tasks to target task, the difficulty of small sample sound quality prediction is solved. Compared with the problem that single source task transfer learning has too much error on some samples, MSTL-CNN can give full play to the advantages of all transfer learning models. The results also show that the MSTL-CNN proposed in this paper is significantly better than the traditional sound quality evaluation methods.

Multi-Source Transfer Learning via Optimal Transport Feature Ranking for EEG Classification

Motor imagery (MI) brain-computer interface (BCI) paradigms have been extensively used in neurological rehabilitation. However, due to the required long calibration time and non-stationary nature of electroencephalogram (EEG) signals, it is challenging to obtain a substantial sample size of EEG data. Transfer learning (TL), which leverages knowledge from existing subjects to enhance the learning performance in new subjects, has been employed to solve this problem. Previous studies often extract sample features directly from the two domains for transfer learning, leading to poor features and negative transfer. To overcome this limitation, an optimal transport feature selection method was developed in this study to select the most suitable features for migration to enhance the generalization capability of the TL. This was achieved by analyzing the diagonal value similarity of the optimal transport coupling matrix. First, the Riemannian tangent space mapping method was used to map the sample covariance matrix of EEG trials from the Riemannian manifold to tangent space. Secondly, the optimal subset of EEG features was selected via the optimal transport feature selection method. Subsequently, the separate distribution alignment method was employed to minimize dissimilarities between the source and target domains. Finally, the weighted voting mechanisms were integrated for decision fusion. The proposed multi-source transfer feature learning (MSTFL) method was validated based on two public BCI Competition IV datasets. Our results demonstrated superior classification accuracy of 85.93% and 79.48%, respectively, on the two public datasets, compared to state-of-the-art models. These findings indicate the effectiveness of our proposed MSTFL model for both feature extraction and classification of MI signals.

Multi-source self-supervised domain adaptation network for VRLA battery anomaly detection of data center under non-ideal conditions

As the key component of uninterruptible power supply, the anomaly detection of valve-regulated lead-acid (VRLA) batteries is the reliable guarantee for the operation of internet data centers. Although deep neural networks (DNNs) have been investigated in battery anomaly detection, the key problem of feature learning under non-ideal conditions, i.e., data scarcity and class-imbalance, is not addressed very well. In this article, a novel DNN called multi-source self-supervised domain adaptation network (MSSSDAN) is proposed for VRLA battery anomaly detection of data center under non-ideal conditions. Firstly, multi-dimensional collaborative self-supervised learning (MDCSSL) is proposed to address the class-imbalanced problem by performing auxiliary feature learning task considering the temporal dependency of different channels. Secondly, a novel transfer learning (TL) framework, multi-source adaptive transfer learning (MSATL) is developed to solve the domain shift problem by adaptively minimizing the distribution discrepancy of different domains. Finally, the outperformance of MSSSDAN is verified on the VRLA battery dataset collected from Tencent data center. The testing results demonstrate that MSSSDAN can effectively solve the problem of battery anomaly detection under non-ideal conditions.

Multi-source domain transfer learning with small sample learning for thermal runaway diagnosis of lithium-ion battery

Data-driven thermal runaway diagnosis based on small amounts of thermal runaway data often struggles to produce satisfactory accuracy. However, in actual application scenarios, obtaining real thermal runaway data has a high cost. To this end, we propose a diagnostic method for multi-source domain transfer learning with few-shot learning (MDTL-FSL), which combines the ideas of small sample learning and adversarial learning. Use multiple different but related thermal runaway case data to obtain common diagnostic knowledge to achieve thermal runaway diagnosis. First of all, in order to avoid negative migration of the algorithm model caused by large differences in data distribution between multi-source domains, a data distribution difference measurement method under segmented pressure differences is proposed. This measurement method is simple and effective from multi-source domains. Select the source domain with a small difference in data distribution from the target domain. Secondly, the adversarial learning idea is integrated into multi-source domain transfer learning to learn temporal invariant features. Then, due to the small number of samples in each source domain in the multi-source domain under the thermal runaway scenario, a source domain reorganization mechanism was designed to find the decision boundary based on meta-learning ideas to achieve small-sample learning in the multi-source domain. Finally, we used cells produced by different battery manufacturers to trigger thermal runaway. By comparison and verification with other data-driven algorithms, the results show that the MDTL-FSL model proposed in this article has higher accuracy. At the same time, we use batteries of different types and capacities to trigger thermal runaway under different working conditions. The MDTL-FSL algorithm can issue early warnings before thermal runaway occurs, thereby effectively ensuring the safe operation of the energy storage system.

Optimization-based multi-source transfer learning for modeling of nonlinear processes

This work develops an optimization-based multi-source transfer learning scheme for modeling of nonlinear chemical processes. Specifically, a transfer learning neural network model for a target process with limited data is developed using the pre-trained model obtained with multiple source processes. Since the performance of transfer learning models depends on the quality of the pre-trained models, we propose a novel Bayesian optimization problem to optimize the selection of multi-source data for the pre-trained models by first deriving a generalization error bound for multi-source domain adaptation using Y-discrepancy distance. Subsequently, the optimization problem is formulated using the theoretical error bound to select the optimal set of multiple sources, which can provide a good initial guess of the weight parameters for transfer learning model. Finally, a simulation study of a chemical reactor process in Aspen Plus Dynamics is conducted to illustrate the effectiveness of the optimization-based multi-source transfer learning scheme.

Enhancing cross-subject EEG emotion recognition through multi-source manifold metric transfer learning

Transfer learning (TL) has demonstrated its efficacy in addressing the cross-subject domain adaptation challenges in affective brain-computer interfaces (aBCI). However, previous TL methods usually use a stationary distance, such as Euclidean distance, to quantify the distribution dissimilarity between two domains, overlooking the inherent links among similar samples, potentially leading to suboptimal feature mapping. In this study, we introduced a novel algorithm called multi-source manifold metric transfer learning (MSMMTL) to enhance the efficacy of conventional TL. Specifically, we first selected the source domain based on Mahalanobis distance to enhance the quality of the source domains and then used manifold feature mapping approach to map the source and target domains on the Grassmann manifold to mitigate data drift between domains. In this newly established shared space, we optimized the Mahalanobis metric by maximizing the inter-class distances while minimizing the intra-class distances in the target domain. Recognizing that significant distribution discrepancies might persist across different domains even on the manifold, to ensure similar distributions between the source and target domains, we further imposed constraints on both domains under the Mahalanobis metric. This approach aims to reduce distributional disparities and enhance the electroencephalogram (EEG) emotion recognition performance. In cross-subject experiments, the MSMMTL model exhibits average classification accuracies of 88.83 % and 65.04 % for SEED and DEAP, respectively, underscoring the superiority of our proposed MSMMTL over other state-of-the-art methods. MSMMTL can effectively solve the problem of individual differences in EEG-based affective computing.

Remaining useful life prediction and cycle life test optimization for multiple-formula battery: A method based on multi-source transfer learning

Achieving highly accurate predictions based on less data with multiple formulations has become a significant challenge. Unlike the traditional prediction model that ignores the similarities and differences between multi-formula batteries, the proposed multi-source transfer RUL prediction is an effective way to jointly transfer several samples and merge degradation information from multi-source domains, thereby taking full advantage of the information. Aiming at the problem of negative transfer and insufficient transfer when multi-source domains are available, dynamic time warping is employed to adaptively select transferable samples, ignoring the local individual differences. The least absolute shrinkage and selection operator are utilized to generate an optimal subset and allocate personalized weights. Then, an optimization strategy for the cycle life tests is proposed by automatically giving a reasonable threshold of stopping the test. The proposed method is verified based on an actual dataset with multiple formulations collected under different test conditions from a battery manufacturer. Experimental results show that the proposed method has high accuracy based on even a small amount of data. The systematic optimization strategy can significantly reduce the time and cost of the test.

H-ensemble: An Information Theoretic Approach to Reliable Few-Shot Multi-Source-Free Transfer

Multi-source transfer learning is an effective solution to data scarcity by utilizing multiple source tasks for the learning of the target task. However, access to source data and model details is limited in the era of commercial models, giving rise to the setting of multi-source-free (MSF) transfer learning that aims to leverage source domain knowledge without such access. As a newly defined problem paradigm, MSF transfer learning remains largely underexplored and not clearly formulated. In this work, we adopt an information theoretic perspective on it and propose a framework named H-ensemble, which dynamically learns the optimal linear combination, or ensemble, of source models for the target task, using a generalization of maximal correlation regression. The ensemble weights are optimized by maximizing an information theoretic metric for transferability. Compared to previous works, H-ensemble is characterized by: 1) its adaptability to a novel and realistic MSF setting for few-shot target tasks, 2) theoretical reliability, 3) a lightweight structure easy to interpret and adapt. Our method is empirically validated by ablation studies, along with extensive comparative analysis with other task ensemble and transfer learning methods. We show that the H-ensemble can successfully learn the optimal task ensemble, as well as outperform prior arts.

A Transfer Approach Using Graph Neural Networks in Deep Reinforcement Learning

Transfer learning (TL) has shown great potential to improve Reinforcement Learning (RL) efficiency by leveraging prior knowledge in new tasks. However, much of the existing TL research focuses on transferring knowledge between tasks that share the same state-action spaces. Further, transfer from multiple source tasks that have different state-action spaces is more challenging and needs to be solved urgently to improve the generalization and practicality of the method in real-world scenarios. This paper proposes TURRET (Transfer Using gRaph neuRal nETworks), to utilize the generalization capabilities of Graph Neural Networks (GNNs) to facilitate efficient and effective multi-source policy transfer learning in the state-action mismatch setting. TURRET learns a semantic representation by accounting for the intrinsic property of the agent through GNNs, which leads to a unified state embedding space for all tasks. As a result, TURRET achieves more efficient transfer with strong generalization ability between different tasks and can be easily combined with existing Deep RL algorithms. Experimental results show that TURRET significantly outperforms other TL methods on multiple continuous action control tasks, successfully transferring across robots with different state-action spaces.

DA-Net: A Disentangled and Adaptive Network for Multi-Source Cross-Lingual Transfer Learning

Multi-Source cross-lingual transfer learning deals with the transfer of task knowledge from multiple labelled source languages to an unlabeled target language under the language shift. Existing methods typically focus on weighting the predictions produced by language-specific classifiers of different sources that follow a shared encoder. However, all source languages share the same encoder, which is updated by all these languages. The extracted representations inevitably contain different source languages' information, which may disturb the learning of the language-specific classifiers. Additionally, due to the language gap, language-specific classifiers trained with source labels are unable to make accurate predictions for the target language. Both facts impair the model's performance. To address these challenges, we propose a Disentangled and Adaptive Network ~(DA-Net). Firstly, we devise a feedback-guided collaborative disentanglement method that seeks to purify input representations of classifiers, thereby mitigating mutual interference from multiple sources. Secondly, we propose a class-aware parallel adaptation method that aligns class-level distributions for each source-target language pair, thereby alleviating the language pairs' language gap. Experimental results on three different tasks involving 38 languages validate the effectiveness of our approach.

Comparison of Domain Selection Methods for Multi-Source Manifold Feature Transfer Learning in Electroencephalogram Classification

Transfer learning (TL) utilizes knowledge from the source domain (SD) to enhance the classification rate in the target domain (TD). It has been widely used to address the challenge of sessional and inter-subject variations in electroencephalogram (EEG)-based brain–computer interfaces (BCIs). However, utilizing knowledge from a combination of both related and non-related sources can significantly deteriorate the classification performance across individual target domains, resulting in a negative transfer (NT). Hence, NT becomes one of the most significant challenges for transfer learning algorithms. Notably, domain selection techniques have been developed to address the challenge of NT emerging from the transfer of knowledge from non-related sources. However, existing domain selection approaches iterate through domains and remove a single low-beneficial domain at a time, which can massively affect the classification performance in each iteration since SDs respond differently to other sources. In this paper, we compare domain selection techniques for a multi-source manifold feature transfer learning (MMFT) framework to address the challenge of NT and then evaluate the effect of beneficial and non-beneficial sources on TL performance. To evaluate the effect of low-beneficial and high beneficial sources on TL performance, some commonly used domain selection methods are compared, namely, domain transferability estimation (DTE), rank of domain (ROD), label similarity analysis, and enhanced multi-class MMFT (EMC-MMFT), using the same multi-class cross-session and cross-subject classification problems. The experimental results demonstrate the superiority of the EMC-MMFT algorithm in terms of minimizing the effect of NT. The highest classification accuracy (CA) of 100% is achieved when optimal combinations of high beneficial sources are selected for two-class SSMVEP problems, while the highest CAs of 98% and 87% are achieved for three- and four-class SSMVEP problems, respectively. The highest CA of 98% is achieved for two-class MI classification problems, while the highest CAs of 90% and 71.5% are obtained for three- and four-class MI problems, respectively.

State of health estimation for lithium battery based on multi-source transfer learning

Aiming at the problem that the state of health estimation accuracy decreases when the lithium battery in training and testing sets has different charging conditions, we proposed a solution based on the transfer learning with the multi-source method. This scheme uses the complete aging test data of 16 batteries under different charging conditions as the source domain and extracts aging characteristics respectively to train the basic model. The similarity coefficient was obtained by the similarity evaluation of the first 200 cycles of each source and target battery, and the learning weight of each source battery model was weighted by the similarity coefficient. The experimental results show that the MAE and RMSE of the transfer learning with a multi-source model decreased by 53.63% and 54.97% compared with the single-source estimation model under new charging conditions.

An instance based multi-source transfer learning strategy for building’s short-term electricity loads prediction under sparse data scenarios

The use of transfer learning for building’s electricity loads prediction has shown great potential when in practice the available electricity consumption data in local energy communities is sparse or of unacceptable quality. However, current model based transfer learning applications largely focus on knowledge transfer from whole data of similar building(s). Domain deviation will probably occur and lead to degraded prediction performance if the selected source domain(s) is difficult to align with the target one in their distributions. To address this issue, this study proposes an instance based multi-source transfer learning framework for short-term electricity loads prediction. A two-stage similarity metric with a boosting style ensemble learning method is provided. Different from commonly used transfer learning, after selecting similar buildings, the most helpful data samples from source domains will be further picked by the Nearest Neighbor Search method, and an improved TrAdaBoost algorithm based on data weighting and reusing is designed for knowledge transfer and prediction. The proposed strategy is applied on two different educational buildings’ hourly energy forecasting with sparse historical data. The influences of source domain numbers and target data quantities on prediction performance are fully investigated. The results show that by using supervised selection of similar buildings’ number, the proposed transfer learning could achieve a 20% error reduction than single-source transfer learning model. Compared with previous multi-source transfer learning model, the proposed strategy also proves its superior prediction accuracy (1.97% vs 2.26%, MAPE).

Semi-supervised multi-source transfer learning for cross-subject EEG motor imagery classification.

Electroencephalogram (EEG) motor imagery (MI) classification refers to the use of EEG signals to identify and classify subjects' motor imagery activities; this task has received increasing attention with the development of brain-computer interfaces (BCIs). However, the collection of EEG data is usually time-consuming and labor-intensive, which makes it difficult to obtain sufficient labeled data from the new subject to train a new model. Moreover, the EEG signals of different individuals exhibit significant differences, leading to a significant drop in the performance of a model trained on the existing subjects when directly classifying EEG signals acquired from new subjects. Therefore, it is crucial to make full use of the EEG data of the existing subjects and the unlabeled EEG data of the new target subject to improve the MI classification performance achieved for the target subject. This research study proposes a semi-supervised multi-source transfer (SSMT) learning model to address the above problems; the model learns informative and domain-invariant representations to address cross-subject MI-EEG classification tasks. In particular, a dynamic transferred weighting schema is presented to obtain the final predictions by integrating the weighted features derived from multi-source domains. The average accuracies achieved on two publicly available EEG datasets reach 83.57 and 85.09 , respectively, validating the effectiveness of the SSMT process. The SSMT process reveals the importance of informative and domain-invariant representations in MI classification tasks, as they make full use of the domain-invariant information acquired from each subject.

Transfer learning with data alignment and optimal transport for EEG based motor imagery classification

Objective. The non-stationarity of electroencephalogram (EEG) signals and the variability among different subjects present significant challenges in current Brain–Computer Interfaces (BCI) research, which requires a time-consuming specific calibration procedure to address. Transfer Learning (TL) offers a potential solution by leveraging data or models from one or more source domains to facilitate learning in the target domain, so as to address these challenges. Approach. In this paper, a novel Multi-source domain Transfer Learning Fusion (MTLF) framework is proposed to address the calibration problem. Firstly, the method transforms the source domain data with the resting state segment data, in order to decrease the differences between the source domain and the target domain. Subsequently, feature extraction is performed using common spatial pattern. Finally, an improved TL classifier is employed to classify the target samples. Notably, this method does not require the label information of target domain samples, while concurrently reducing the calibration workload. Main results. The proposed MTLF is assessed on Datasets 2a and 2b from the BCI Competition IV. Compared with other algorithms, our method performed relatively the best and achieved mean classification accuracy of 73.69% and 70.83% on Datasets 2a and 2b respectively. Significance. Experimental results demonstrate that the MTLF framework effectively reduces the discrepancy between the source and target domains and acquires better classification performance on two motor imagery datasets.

Prediction of medial knee contact force using multisource fusion recurrent neural network and transfer learning.

Estimation of knee contact force (KCF) during gait provides essential information to evaluate knee joint function. Machine learning has been employed to estimate KCF because of the advantages of low computational cost and real-time. However, the existing machine learning models do not adequately consider gait-related data's temporal-dependent, multidimensional, and highly heterogeneous nature. This study is aimed at developing a multisource fusion recurrent neural network to predict the medial condyle KCF. First, a multisource fusion long short-term memory (MF-LSTM) model was established. Then, we developed a transfer learning strategy based on the MF-LSTM model for subject-specific medial KCF prediction. Four subjects with instrumented tibial prostheses were obtained from the literature. The results showed that the MF-LSTM model could predict medial KCF to a certain high level of accuracy (the mean of ρ = 0.970). The transfer learning model improved the prediction accuracy (the mean of ρ = 0.987). This study shows that the MF-LSTM model is a powerful and accurate computational tool for medial KCF prediction. Introducing transfer learning techniques could further improve the prediction performance for the target subject. This coupling strategy can help clinicians accurately estimate and track joint contact forces in real time.

Multi-source transfer learning for facial emotion recognition using multivariate correlation analysis.

Deep learning techniques have proven to be effective in solving the facial emotion recognition (FER) problem. However, it demands a significant amount of supervision data which is often unavailable due to privacy and ethical concerns. In this paper, we present a novel approach for addressing the FER problem using multi-source transfer learning. The proposed method leverages the knowledge from multiple data sources of similar domains to inform the model on a related task. The approach involves the optimization of aggregate multivariate correlation among the source tasks trained on the source dataset, thus controlling the transfer of information to the target task. The hypothesis is validated on benchmark datasets for facial emotion recognition and image classification tasks, and the results demonstrate the effectiveness of the proposed method in capturing the group correlation among features, as well as being robust to negative transfer and performing well in few-shot multi-source adaptation. With respect to the state-of-the-art methods MCW and DECISION, our approach shows an improvement of 7% and [Formula: see text]15% respectively.

Enhancing classification performance through multi-source online transfer learning algorithm with oversampling

Multi-source online transfer learning uses the tagged data from multiple source domains to enhance the classification performance of the target domain. For unbalanced data sets, a multi-source online transfer learning algorithm that can oversample in the feature spaces of the source domain and the target domain is proposed. The algorithm consists of two parts: oversampling multiple source domains and oversampling online target domains. In the oversampling phase of the source domain, oversampling is performed in the feature space of the support vector machine (SVM) to generate minority samples. New samples are obtained by amplifying the original Gram matrix through neighborhood information in the source domain feature space. In the oversampling phase of the online target domain, minority samples from the current batch search for k-nearest neighbors in the feature space from multiple batches that have already arrived, and use the generated new samples and the original samples in the current batch to train the target domain function together. The samples from the source domain and the target domain are mapped to the same feature space through the kernel function for oversampling, and the corresponding decision function is trained using the data from the source domain and the target domain with relatively balanced class distribution, so as to improve the overall performance of the algorithm. Comprehensive experiments were conducted on four real datasets, and compared to other baseline algorithms on the Office Home dataset, the accuracy improved by 0.0311 and the G-mean value improved by 0.0702.

AUDIT: Functional Qualification in Additive Manufacturing via Physical and Digital Twins

Abstract Additive manufacturing (AM) has revolutionized the way we design, prototype, and produce complex parts with unprecedented geometries. However, the lack of understanding of the functional properties of 3D printed parts has hindered their adoption in critical applications where reliability and durability are paramount. This paper proposes a novel approach to the functional qualification of 3D-printed parts via physical and digital twins. Physical twins are parts that are printed under the same process conditions as the functional parts and undergo a wide range of (destructive) tests to determine their mechanical, thermal, and chemical properties. Digital twins are virtual replicas of the physical twins that are generated using finite element analysis (FEA) simulations based on the 3D shape of the part of interest. We propose a novel approach to transfer learning, specifically designed for the fusion of diverse, unstructured 3D shape data and process inputs from multiple sources. The proposed approach has demonstrated remarkable results in predicting the functional properties of 3D-printed lattice structures. From an engineering standpoint, this paper introduces a comprehensive and innovative methodology for the functional qualification of 3D-printed parts. By combining the strengths of physical and digital twins with transfer learning, our approach opens up possibilities for the widespread adoption of 3D printing in safety-critical applications. Methodologically, this work presents a significant advancement in transfer learning techniques, specifically addressing the challenges of multi-source (e.g., digital and physical twins) and multi-input (e.g., 3D shapes and process variables) transfer learning.

Adaptive and Simultaneous Trajectory Prediction for Heterogeneous Agents via Transferable Hierarchical Transformer Network

Simultaneously and accurately predicting trajectories of multiple heterogeneous agents is crucial for intelligent transportation systems (ITS) applications, e.g., connected and autonomous vehicles. Existing model-based and data-driven methods can achieve good prediction accuracy, but most of them neglect the domain shift issue and prevalent imperfect data problems, i.e., few-shot learning and zero-shot learning issues. To address these issues, we propose a multi-source transfer learning (TL) framework, transferable hierarchical Siamese Transformer network (T-HSTN), for trajectory prediction of multiple heterogeneous agents, e.g., vehicles, bicycles, and pedestrians, at urban unsignalized intersections under small data conditions. Specifically, by extending the self-attention mechanism and exploring feature representations of traffic scenes, a Transformer-based network that hierarchically extracts temporal/spatial features and map features is introduced as the basic prediction model. Moreover, a TL framework with adaptive learning and feature alignment modules is built to explore the feature representations of unfixed traffic scenes and align both statistical and deep features to learn domain-invariant knowledge. More challenging trajectory prediction experiments are designed, corresponding to newly-built or badly-instrumented intersections under real-world scenarios. Experimental results verify the proposed method’s high accuracy, transferability, and generability. Our work fills the gap in solutions and benchmarks for TL tasks in trajectory prediction for heterogeneous agents. The conducted TL experiments provide a more practical setting of considering imperfect data problems in trajectory prediction.