ISSN 2394-5125
 

Research Article 


Hyper-parameter fine tuning in CNNís using data augmentation and regularization to detect direction of learning

Vikas B, Dr. Y Radhika, S. Surya Prakash.

Abstract
Deep neural networks are highly successful in solving numerous real time problems and practical
applications. However, large quantities of high-quality training data play a major role in the effectiveness of deep
learning by avoiding overfitting. There are many cases where a deep learning classifier cannot be trained perfectly
due to insufficient amount of training data. In such scenarios, the most commonly adopted strategy is ďData
augmentationĒ, which enlarges the training data. However, the quality and standards of this augmented data may
be arguable. While training deep neural networks overfitting seems to be a major problem. This problem arises
due to the unequal distribution of classes within a dataset and initialization of the model parameters affecting the
performance of the model. To overcome such problems, many deep neural network problems have been
successfully implemented using the Dropout regularization in their model training. In this work, Data
augmentation is applied to MNIST data set to generate required volume of training data from original training
data. Later Dropout Regularizer is implemented in convolutional neural networks to solve the overfitting problem.
Moreover, a comparative analysis between various standard models and CNN is made in terms of the accuracy.
Data augmentation here has been used to provide crucial data which is unavailable in most general cases. In most
training scenarios the data available is scarce which causes the most common problem where a model might either
overfit or underfit depending on the configuration and the harshness of error balancing. The method used here can
help in these scenarios to provide a hint on the root cause for the model inefficiency. The parallel comparison of
a dropout infused model indicates much harsher error correction. This observation of relative performance helps
to detect if the standard model is suffering due to under-fitting or over-fitting.The findings conclude the same
indicating the increase in accuracy seen in the initial phases but a drop when continued further.

Key words: Hyper-parameter fine tuning in CNNís using data augmentation and regularization to detect direction of learning


 
ARTICLE TOOLS
Abstract
PDF Fulltext
How to cite this articleHow to cite this article
Citation Tools
Related Records
 Articles by Vikas B
Articles by Dr. Y Radhika
Articles by S. Surya Prakash
on Google
on Google Scholar


How to Cite this Article
Pubmed Style

Vikas B, Dr. Y Radhika, S. Surya Prakash. Hyper-parameter fine tuning in CNNís using data augmentation and regularization to detect direction of learning. JCR. 2020; 7(17): 1531-1537. doi:10.31838/jcr.07.17.193


Web Style

Vikas B, Dr. Y Radhika, S. Surya Prakash. Hyper-parameter fine tuning in CNNís using data augmentation and regularization to detect direction of learning. http://www.jcreview.com/?mno=5226 [Access: May 30, 2021]. doi:10.31838/jcr.07.17.193


AMA (American Medical Association) Style

Vikas B, Dr. Y Radhika, S. Surya Prakash. Hyper-parameter fine tuning in CNNís using data augmentation and regularization to detect direction of learning. JCR. 2020; 7(17): 1531-1537. doi:10.31838/jcr.07.17.193



Vancouver/ICMJE Style

Vikas B, Dr. Y Radhika, S. Surya Prakash. Hyper-parameter fine tuning in CNNís using data augmentation and regularization to detect direction of learning. JCR. (2020), [cited May 30, 2021]; 7(17): 1531-1537. doi:10.31838/jcr.07.17.193



Harvard Style

Vikas B, Dr. Y Radhika, S. Surya Prakash (2020) Hyper-parameter fine tuning in CNNís using data augmentation and regularization to detect direction of learning. JCR, 7 (17), 1531-1537. doi:10.31838/jcr.07.17.193



Turabian Style

Vikas B, Dr. Y Radhika, S. Surya Prakash. 2020. Hyper-parameter fine tuning in CNNís using data augmentation and regularization to detect direction of learning. Journal of Critical Reviews, 7 (17), 1531-1537. doi:10.31838/jcr.07.17.193



Chicago Style

Vikas B, Dr. Y Radhika, S. Surya Prakash. "Hyper-parameter fine tuning in CNNís using data augmentation and regularization to detect direction of learning." Journal of Critical Reviews 7 (2020), 1531-1537. doi:10.31838/jcr.07.17.193



MLA (The Modern Language Association) Style

Vikas B, Dr. Y Radhika, S. Surya Prakash. "Hyper-parameter fine tuning in CNNís using data augmentation and regularization to detect direction of learning." Journal of Critical Reviews 7.17 (2020), 1531-1537. Print. doi:10.31838/jcr.07.17.193



APA (American Psychological Association) Style

Vikas B, Dr. Y Radhika, S. Surya Prakash (2020) Hyper-parameter fine tuning in CNNís using data augmentation and regularization to detect direction of learning. Journal of Critical Reviews, 7 (17), 1531-1537. doi:10.31838/jcr.07.17.193