Rademacher dropout: An adaptive dropout for deep neural network via optimizing generalization gap

Wang, Haotian; Yang, Wenjing; Zhao, Zhenyu; Luo, Tingjin; Ji, Wang; Tang, Yuhua

doi:10.1016/j.neucom.2019.05.008

Cited by 23 publications

(9 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Modified linear unit (ReLU) is used as the activation function, and L2 regularization is introduced to avoid network over-fitting. A dropout layer (Wang et al, 2019) (with random inactivation rate 0.2) is added between the two stacked convolutional layers in the residual module to randomly discard some neurons during training, in order to prevent over-fitting and enhance generalization performance of the network.…”

Section: Residual Modulementioning

confidence: 99%

Retinal Vessel Segmentation Algorithm Based on Residual Convolution Neural Network

Chen

Cao

et al. 2021

Front. Bioeng. Biotechnol.

View full text Add to dashboard Cite

Retinal vessels are the only deep micro vessels that can be observed in human body, the accurate identification of which has great significance on the diagnosis of hypertension, diabetes and other diseases. To this end, a retinal vessel segmentation algorithm based on residual convolution neural network is proposed according to the characteristics of the retinal vessels on fundus images. Improved residual attention module and deep supervision module are utilized, in which the low-level and high-level feature graphs are joined to construct the encoder-decoder network structure, and atrous convolution is introduced to the pyramid pooling. The experiments result on the fundus image data set DRIVE and STARE show that this algorithm can obtain complete retinal vessel segmentation as well as connected vessel stems and terminals. The average accuracy on DRIVE and STARE reaches 95.90 and 96.88%, and the average specificity is 98.85 and 97.85%, which shows superior performance compared to other methods. This algorithm is verified feasible and effective for retinal vessel segmentation of fundus images and has the ability to detect more capillaries.

show abstract

Section: Residual Modulementioning

confidence: 99%

Retinal Vessel Segmentation Algorithm Based on Residual Convolution Neural Network

Chen

Cao

et al. 2021

Front. Bioeng. Biotechnol.

View full text Add to dashboard Cite

show abstract

“…It is understood that as the brain learns stimuli over a period of time, it shows stronger functional connections while having a decrease in randomness within the networks [76]. In contrast, it has been seen in artificial neural networks that random dropouts can benefit learning, and improve the robustness of the networks [77]- [79]. The amount of randomness can have a major impact on the learning ability of the networks.…”

Section: B Randomnessmentioning

confidence: 99%

The Why, What and How of Artificial General Intelligence Chip Development

James¹

2020

Preprint

View full text Add to dashboard Cite

“…In feature learning and cognition of large-scale multi-state mixed iris data, the statistical learning under the ideal steady state iris training set can ensure intuitive and reliable feature summary. However, over-fitting phenomenon will occur [15]. The cognitive learning of set labels cannot fully evaluate the iris acquisition state in advance, but the initial concept is dependent on the original tags.…”

Section: Cognitive Learning [3]mentioning

confidence: 99%

Statistical Cognitive Learning and Security Output Protocol for Multi-State Iris Recognition

Liu

Zhu

et al. 2019

IEEE Access

View full text Add to dashboard Cite

In this paper, the research focuses on the constrained iris under different acquisition states, and trains the mixed data without manual label division. A multi-state iris multi-classification recognition method based on convolutional neural network fusion statistical cognitive learning is proposed. The recognition process is divided into an image processing module, a classification module and a result output module. The image processing module converts iris images into recognition tags through convolutional neural network. By combining the characteristics of cognitive learning and statistical learning, the classification module converts iris features of the same category to the convolutional neural network label parameters by data statistical, and forms a single iris cognitive concept, thereby designing a single-category recognizer. Multiple single recognizers are combined in parallel to perform multi-classification recognition by the codeless fusion recognition mode, the results are further encrypted by external encryption. The final classification result is exported through the result output module, which can improve the security of data transmission. After confirming stealing attacks, the result output security protocol is activated, and legitimate users who pass the reliable third-party authentication can still get correct results after changing the decryption key and the result output mapping path. In the multi-state iris recognition, the experimental results of JLU iris library demonstrate that the proposed method can ensure the correct rate of the single classifier certification. In multi-classification recognition, it can effectively cluster iris of the same category and distinguish different categories of irises. In addition, the method can also effectively respond to the stealing attack behavior.

show abstract

Rademacher dropout: An adaptive dropout for deep neural network via optimizing generalization gap

Cited by 23 publications

References 10 publications

Retinal Vessel Segmentation Algorithm Based on Residual Convolution Neural Network

Retinal Vessel Segmentation Algorithm Based on Residual Convolution Neural Network

The Why, What and How of Artificial General Intelligence Chip Development

Statistical Cognitive Learning and Security Output Protocol for Multi-State Iris Recognition

Contact Info

Product

Resources

About