Which algorithms can be used for online learning?

[sgiri]

[Separated from another question asked by ss7dec]
What I grasped from the Video recording, let me recall below —

For Online Learning ML Algorithmic techniques one can use either of the following methods namely
a) SGD (Stochastic Gradient Descent) or
b) ANN (Artificial Neural Networks)

SVM (Support Vector Machine) Algorithm cannot be used for Online Learning technique.

However, usage of Kernelized SVM can be used for Online learning.

Kindly let me know what I have understood and written as part of Question 2 is correct or incorrect?

Need to understand if any mistakes/errors have been made so that my mistakes can be rectified. This is necessary and important for better understanding of the theoretical concepts.

Looking forward to your response pertaining to both the above 2 queries.

– by ss7dec

[sgiri]

@sgiri

For Online Learning ML Algorithmic techniques one can use either of the following methods namely
a) SGD (Stochastic Gradient Descent) or
b) ANN (Artificial Neural Networks)

SVM (Support Vector Machine) Algorithm cannot be used for Online Learning technique.

You understanding is right but there are more algorithms such as decision trees which require the re-processing of entire data set in order to prepare the model.

However, usage of Kernelized SVM can be used for Online learning.

No. The only way to make SVM online learning is that instead of using the SVM algorithm, we SGD with the SVM loss (also called hinge loss).

If you want to have Kernels with online learning, you can generate the non-linear features and the use SGD with SVM loss.

[ss7dec]

Very important points regarding SVM have been clarified with your answers.

For this, I must appreciate Sandeep for his valuable inputs. It has helped in giving more clarity to SVM ML Algorithmic concept. Thanks once again for the gyaan imparted.