# Why are certain scientists working with deep learning instead of traditional machine learning in some case studies? [closed]

Are there scientific papers or projects that explain their choice of working with deep learning instead of machine learning in their case studies?

Specifically, I am looking for references to papers or website urls of existing projects which have explanations of why the authors used deep learning on their data over traditional machine learning.

For example, why did they use deep neural networks (or CNN) instead of SVM (or any other machine learning technique)

## closed as primarily opinion-based by Stephan Kolassa, kjetil b halvorsen, mdewey, Tim♦Mar 1 '18 at 17:13

Many good questions generate some degree of opinion based on expert experience, but answers to this question will tend to be almost entirely based on opinions, rather than facts, references, or specific expertise. If this question can be reworded to fit the rules in the help center, please edit the question.

• Well, deep learning is still machine learning, just sounds "cooler" nowadays. Can you be more specific what are you actually asking for? Are you looking for references to papers which have a paragraph explaining why the authors used deep learning on their data over, say, random forests? – Jan Kukacka Mar 1 '18 at 11:35
• Yes, I am looking for references to papers or existing projects' web site urls which have a paragraph explaining why the authors used deep learning on their data over. for example, why did they use deep neural networks (or CNN) instead of SVM (or any other machine leaning existing technique) – miss mimi Mar 1 '18 at 12:07
• You should edit your question to include the explanatory comment. – Peter Flom Mar 1 '18 at 14:06
• Because for some tasks deep neural networks are much better than other machine learning approaches. Try getting $> 99.3\%$ accuracy on MNIST (a very simple dataset) with an SVM, multinomial logistic regression, LDA or even GPC. If you manage to, try getting $>80\%$ accuracy on ImageNet, and compare the results with a standard CNN architecture such as ResNet-50. Of course it’s not the case for all applications, but for some of them DNNs just wipe the floor with other ML algorithms. – DeltaIV Mar 1 '18 at 15:22

Deep Learning is a subset of machine learning, not an alternative/cooler name for the same thing. There are many specific reasons as to why and when anyone would choose deep learning over shallow learning. I believe any good book on the subject would answer all your questions (I recommend Deep learning with Python).

You need to ALSO understand representation learning before differentiating between machine learning and deep learning

Machine learning requires you to chose features by experimenting and applying experience. It is shallow in the sense that it does not have multiple layers unlike deep learning.

ML is a superset of Representation learning which learns the features automatically. Deep learning is a subset of representation learning.

Here is a tweet with a picture that might help. https://twitter.com/dickeysingh/status/967944753550471168?s=21

If you are technically inclined and favor core fundamentals I recommend THE book by Ian Goodfellow deeplearningbook.org

I recommend the kindle version instead of the free web version.

Hope this helps somewhat.