There is a misconception that Support vector machine are only utilized in straight forward Machine learning.Can SVM is deep learning classifier?.Needs your noble opinions/views?
I agree with Graham's answer. Deep learning by definition involves fully-connected and convolutional neural networks and not traditional classification SVMs or regressive SVMs. it is however possible to use SVMs as a technique to enhance deep learning models. This can be achieved for instance by replacing the softmax
layer of the neural network with a linear support vector machine. Results are explained in paper below:
If I understand right what you want to do i suppose that you can do that. There are some descriptions, e.g. "Deep Network with Support Vector Machines" by Kim.
Hello. I think more detail is needed to solicit meaningful answers about this question. The terms "straightforward" and "deep" are not very meaningful, at least to me. Do you wish to add extra layers of complexity or extra terms in the decision function of the SVM? What do you expect to gain? The SVM is already a very flexible framework with a solid mathematical basis.
I agree with Graham's answer. Deep learning by definition involves fully-connected and convolutional neural networks and not traditional classification SVMs or regressive SVMs. it is however possible to use SVMs as a technique to enhance deep learning models. This can be achieved for instance by replacing the softmax
layer of the neural network with a linear support vector machine. Results are explained in paper below:
That misconception is as told wrong. Deep learning and SVM are different techniques.But thinking SVM as deep learning has misconceptions too. They can not be same but can be used together. Deep learning is more powerfull classifier than SVM. However there are many difficulties to use DL. So if you can use SVM and have good performance,then use SVM. Otherwise DL will be required. By the way DL is not like SVM. SVM has different kernels and other details are the same. However DL is not a single structure. It means ANN with many layers. But in these layers you can use different kind of sturctures like RNN,CNN,LSTM etc..
Combining apparently different ideas is usually fruitful, but sometimes not straightforward.
In this case, note that SVM can be generalized by kernel methods. And there is a very interesting line of research on hierarchical kernel methods. I suggest to review the work by Prof. Johan Suykens (KU Leuven) and coauthors. You will find fancy names as "Deep Restricted Kernel Machines".