We sometimes use so$-$called the kernel trick in SVM to classify data that is not linearly separable. What is the main idea of the kernel trick in this process?

We sometimes use so$-$called the kernel trick in SVM to classify data that is not linearly separable. What is the main idea of the kernel trick in this process?

To remove noises associated with the data set.

To map the linearly non$-$separable data to a higher dimension in which it becomes linearly separable.

To address the overfitting problem with SVM$.$

None of above.