Robustness of transfer learning to image degradation

Ren, Sijin; Li, Cheryl Q.

doi:10.1016/j.eswa.2021.115877

Cited by 14 publications

(8 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Transfer learning is commonly employed to solve computer vision problems with a small dataset, and the acquisition of more data is time-consuming or expensive. Ren et al [27] demonstrated accurate image classification employing transfer learning algorithms on small or poor image datasets. The minimum data size requirements for the training transfer learning model were explored.…”

Section: Related Workmentioning

confidence: 99%

An Integrated Approach towards Efficient Image Classification Using Deep CNN with Transfer Learning and PCA

Sharma

Singh

2022

Adv. technol. innov.

View full text Add to dashboard Cite

In image processing, developing efficient, automated, and accurate techniques to classify images with varying intensity level, resolution, aspect ratio, orientation, contrast, sharpness, etc. is a challenging task. This study presents an integrated approach for image classification by employing transfer learning for feature selection and using principal component analysis (PCA) for feature reduction. The PCA algorithm is employed for reducing the dimensionality of the features extracted by the VGG16 model to obtain a handful of features for speeding up image reorganization. For multilayer perceptron classifiers, support vector machine (SVM) and random forest (RF) algorithms are used. The performance of the proposed approach is compared with other classifiers. The experimental results establish the supremacy of the VGG16-PCA-Multilayer perceptron model integrated approach and achieve a reorganization accuracy of 91.145%, 95.0%, 92.33%, and 98.59% on Fashion-MNIST dataset, ORL dataset of faces, corn leaf disease dataset, and rice leaf disease datasets, respectively.

show abstract

Section: Related Workmentioning

confidence: 99%

An Integrated Approach towards Efficient Image Classification Using Deep CNN with Transfer Learning and PCA

Sharma

Singh

2022

Adv. technol. innov.

View full text Add to dashboard Cite

show abstract

“…Second, the Intel Image Classification dataset with 6 classes containing 17.034 labeled images of natural scenes around the world, used in many studies as well [31,33,41]. It provided by Intel corporation to create another benchmark in image classification tasks such as scene recognition [3].…”

Section: Data Collection and Preprocessing: Datasetsmentioning

confidence: 99%

Guiding the retraining of convolutional neural networks againstadversarial inputs: best guidance metrics and configurations

2022

View full text Add to dashboard Cite

Background: When using deep learning models, there are many possible vulnerabilities and some of the most worrying are the adversarial inputs, which can cause wrong decisions (e.g., wrongly classifying an image) with minor perturbations. Therefore, it becomes necessary to retrain these models against adversarial inputs, as part of the software testing process addressing the vulnerability to adversarial inputs. Furthermore, for an energy efficient testing and retraining process, data scientists need support on which are the best guidance metrics for reducing the adversarial inputs to use as well as optimal dataset configurations.Aims: We examined four guidance metrics for retraining deep learning models, specifically with convolutional neural network architecture, and three retraining configurations. Our goal is to improve the convolutional neural networks against adversarial inputs with regard to accuracy, resource utilization and time from the point of view of a data scientist in the context of image classification.Method: We conduced an empirical study in two datasets for image classification. We explore: (a) the accuracy, resource utilization and time of retraining convolutional neural networks by ordering new training set by four different guidance metrics (neuron coverage, likelihood-based surprise adequacy, distance-based surprise adequacy and random), (b) the accuracy and resource utilization of retraining convolutional neural networks with three different configurations (from scratch and augmented dataset, using weights and augmented dataset, and using weights and only adversarial inputs).Results: We reveal that retraining with adversarial inputs from original model weights and by ordering with surprise adequacy metrics gives the best model w.r.t. accuracy, resource utilization and time.Conclusions: Although more studies are necessary, we recommend data scientists to use the above configuration and metrics to deal with the vulnerability to adversarial inputs of deep learning models, as they can improve their models against adversarial inputs without using many inputs. We also show that dataset size has an important impact on the results. CCS Concepts• Computing methodologies → Machine learning; Artificial intelligence; • Software and its engineering → Software creation and management.

show abstract

“…According to [17], TL can be roughly divided into four categories: instance-based TL [18], featurebased TL [19], model-based TL [20], and relational-based TL [21]. Feature-based TL aims to reduce the difference between the source and target domains via using the feature of samples, which includes manifold learning [22], feature extraction [23], and transfer subspace learning [24,25]. For transfer subspace learning, there are two ways to find the common subspace.…”

Section: Introductionmentioning

confidence: 99%

Robust transfer subspace learning based on low-rank and sparse representation for bearing fault diagnosis

Yu,

Xiu,

et al. 2024

Meas. Sci. Technol.

View full text Add to dashboard Cite

With the development of industrial intelligence, data-driven fault diagnosis plays an important role in Prognostics and Health Management (PHM). However, there is usually a large amount of unlabeled data from different working conditions, making cross-domain fault diagnosis unstable and inflexible. To deal with this issue, we propose two novel transfer subspace learning methods based on the Low-Rank Sparse Representation (LRSR), called LRSR-G and LRSR-R. Specifically, LRSR-G integrates an additional matrix with LRSR to characterize the Gaussian noise for robustness, as well as capture global and local structures. Furthermore, LRSR-R adaptively learns the label matrix from samples instead of using the binary labeling matrix in LRSR-G, thus providing the possibility to improve the flexibility. In addition, we develop two efficient algorithms using the Alternating Direction Method of Multipliers (ADMM) to solve the proposed LRSR-G and LRSR-R. Extensive experiments are conducted on the Case Western Reserve University (CWRU) dataset and Jiangnan University (JNU) dataset. The results show that the proposed LRSR-G and LRSR-R perform better than the existing methods, while LRSR-R has more potential in cross-domain fault diagnosis tasks.

show abstract

Robustness of transfer learning to image degradation

Cited by 14 publications

References 35 publications

An Integrated Approach towards Efficient Image Classification Using Deep CNN with Transfer Learning and PCA

An Integrated Approach towards Efficient Image Classification Using Deep CNN with Transfer Learning and PCA

Guiding the retraining of convolutional neural networks againstadversarial inputs: best guidance metrics and configurations

Robust transfer subspace learning based on low-rank and sparse representation for bearing fault diagnosis

Contact Info

Product

Resources

About