Both can be used as activation functions. You can use SVM as activation function of a CNN model not in a direct way but through an indirect one. The process is you have to collect the features of the final layer of CNN model then perform SVM classification on that feature matrix. Dimensionality reduction techinques such as PCA,LDA are sometimes used also for reducing dimension of the matrix.

You can also implement SVM with CNN in keras by using "Squared Hinge" loss function in Keras(for binary classification though)

Dear Divedari Nagamani , from a theoretical standpoint I would not understand why one would want to combine them and have a system implement both of them because both functions work rather interchangeably. Here is for a nice tutorial explaining why

https://cs231n.github.io/linear-classify/#svm-vs-softmax

And here are two more papers discussing the advantages of SVMs in CNNs over softmax:

https://arxiv.org/pdf/1712.03541.pdf

Article Convolutional support vector machines for speech recognition

Good luck with your research and happy modelling!

https://www.analyticsvidhya.com/blog/2021/06/build-an-image-classifier-with-svm/

Hi Divedari Nagamani ,

From theory both softmax and SVM can be used as final layer classification purpose. I do not see any use of SVM after softmax, instead we can use SVM in place of softmax.

Yes!!

Yes you can.

Cnn -svm-softmax

Here cnn for feature extraction

Svm for simple simple classification model

Softmax for probability generation

Article An Architecture Combining Convolutional Neural Network (CNN)...