A

Seminar
On

APPLICATION OF DATA AUGMENTATION ON DEEP LEARNING

Submitted by
Priya Sharma
2116602
Supervised by
Mrs. Debasmita Ghosh Roy

SCHOOL OF AUTOMATION
BANASTHALI VIDYAPITH RAJASTHAN
October 2022
Presentation Outlines
 Introduction

 Problem Description

 Advantages

 disadvantages

 Results & Discussions

 Conclusion

 Future scope

2
 INTRODUCTION
Data Augmentation has been a highly explored aspect in Machine Learning. One of the most important differences
between earlier works and the present is the advancement of the prior knowledge encoded in augmentations. The success
of Data Augmentation in Computer Vision has been aided by the simplicity of designing label-preserving transformations,
as we will go into greater detail about later in the survey. For example, a dog image is still a dog after rotating it, translating
it on the x or y axis, increasing the intensity of the red channel, and so on. It is easy to brainstorm these semantically-
preserving augmentations for photos, but it is considerably harder to accomplish this in the text domain.

3
 PROBLEM DESCRIPTION
The task of training a DNN under robust generalization can be cast as modified version of Equation 1, where, in
addition to optimizing for parameters θ we also need to select, for each training data input x, a suitable variant x ′
= x + δ, where δ ∈ S. Following [21] this can be cast as the following saddle-point optimization problem:
min θ E(x,y)∼D [max δ ∈S L(θ, x + δ,y)]

4
 ADVANTAGES
It reduces the cost of collection of data.

It reduces the cost of labelling data.

It improves the model prediction accuracy.

It prevents data scarcity.

It frames better data models.

It reduces data overfitting.

It creates variability and flexibility in data models.

It increases the generalization ability of the data models.

5
 DISADVANTAGES
It requires very large amount of data in order to perform better than other techniques.

It is extremely expensive to train due to complex data models. Moreover deep learning requires
expensive GPUs and hundreds of machines. This increases cost to the users.

There is no standard theory to guide you in selecting right deep learning tools as it requires
knowledge of topology, training method and other parameters. As a result it is difficult to be
adopted by less skilled people.

It is not easy to comprehend output based on mere learning and requires classifiers to do so.
Convolutional neural network based algorithms perform such tasks.

6
 Result and discussion
A summary of techniques for data preparation and augmentations looked at in this essay. Data pre-
main processing's objective aims to offer the highest quality data possible for data mining. cleaning
techniques are employed to eliminate all the noise from data and eliminate unneeded data. To the
use of data integration is to centralize all available data. Strategies for data reduction and
transformation. Thus, it follows that the successful use of data pre-processing in machine learning
artificial intelligence to improve the precision of our models.

7
 CONCLUSION
In conclusion, this survey has presented several strategies for applying Data Augmentation in Text
data. These augmentations provide an interface to allow developers to inject priors about their task
and data domain into the model.

We have additionally presented how Data Augmentation can help simulate distribution shift
and test generalization. As Data Augmentation for NLP is relatively immature compared to
Computer Vision, we highlight some of the key similarities and differences.

8
 FUTURE SCOPE
Our future work related to this paper will contain but will not be limited to testing the
neural network efficiency after pre-training it with the usage of
synthetic images generated with the style transfer methods .

9
 References

1.Shorten C, Khoshgoftaar T, Furth B. Deep learning applications for covid-19. J Big Data. 2021. https://doi.org/10.
1186/s40537-020-00392-9.
2. Tang R, Nogueira R, Zhang E, Gupta N, Cam P, Cho K, Lin J. Rapidly bootstrapping a question answering dataset for
covid-19. 2020. arXiv:2004.11339. Accessed Jul 2021.
3. Cachola I, Lo K, Cohan A, Weld DS. TLDR: extreme summarization of scientific documents. 2020. arXiv:2004.15011.
Accessed Jul 2021.
4. Srivastava N, Hinton G, Krizhevsky A, Sutskever I, Salahuddin R. Dropout: a simple way to prevent neural networks
from over fitting. J Mach Learn Res. 2014;15(1):1929–58.
5. Kukačka J, Golkov V, Cremer’s D. Regularization for deep learning: a taxonomy 2017. arXiv:1710.10686. Accessed
Jul 2021 .
6. Shorten C, Khoshgoftaar T. A survey on image data augmentation for deep learning. J Big Data. 2019; 6:1–48.
7. Wang R, Lehman J, Clune J, Stanley KO. Poet: Open-ended coevolution of environments and their optimized
solutions. In: Proceedings of the Genetic and Evolutionary Computation Conference. GECCO ’19, pp. 142–151.
Association for Computing Machinery, New York, NY, USA 2019. https://doi.org/10.1145/3321707.3321799.
10
8. Weiss K, Khoshgoftaar T, Wang D. A survey of transfer learning. J Big Data. 2016. https://doi.org/10.1186/
s40537-016-0043-6.

9. O’Gara S., and McGuinness K., "Comparing data augmentation strategies for deep image classification", IMVIP
2019: Irish Machine Vision and Image Processing, Technological University Dublin, Dublin, Ireland, August 28-30.
doi:10.21427/148b-ar75

10. J. Zhang, K. Yu, Z. Wen, X. Qi, A.K. Paul, in: 3D Reconstruction for Motion Blurred Images Using Deep Learning-
Based Intelligent Systems, 66, CMC-Computers Materials & Continua, 2021, pp.2087-2104.

11. Nemade, D. Shah, IoT based water parameter testing in linear topology, in: Proceedings of the 10th
International Conference on Cloud Computing, Data Science & Engineering (Confluence), 2020, pp. 546–551,
doi:10.1109/Confluence47617.2020.90582.

12. Devlin J, Chang M-W, Lee K, Toutanova K. BERT: Pre-training of deep bidirectional transformers for language
understanding. In: Proceedings of the 2019 Conference of the North American Chapter of the Association for
Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pp. 4171–4186.
11
Association for Computational Linguistics, Minneapolis, Minnesota 2019. https://doi.org/10.18653/v1/N19-1423.
13. OpenAI: DALL.E: Creating Images from Text. OpenAI 2021. https://openai.com/blog/dall-e/. Accessed Jul 2021

14. Feng SY, Gangal V, Wei J, Chandra S, Vosoughi S, Mitamura T, Hovey E. A Survey of Data Augmentation Approaches
for NLP. 2021;2105:03075.

15. Deng J, Dong W, Socher R, Li L-J, Li K, Fei-Fei L. ImageNet: A large-scale hierarchical image database. In: 2009 IEEE
Conference on Computer Vision and Pattern Recognition, pp. 248–255 2009. Ieee

12
T han k ….
… … ……
yo u… … 13