CODEWITHSUNDEEP
pythonoptimizationsvm
Mohammed Farahmand
Mohammed Farahmand

Reputation: 498

Optimizer for an SVM-like Model

I've been studying the paper, "Support Vector Data Description". The authors describe a model similar to One-Class SVMs. However, the solution finds a sphere that encapsulates the samples. The objective function is formulated as:

enter image description here

K denotes the kernel function, and alpha is the adjustable parameter. I was thinking of solving this problem using PyTorch autograd, but the package has no method for enforcing the constraint. I've also checked the SciPy optimization package but I'm not really familiar with any of these methods.

Upvotes: 0

Views: 43

Answers (1)

Karl
Karl

Reputation: 5303

It sounds like that's just weight regularization. You can use the weight decay parameter in your optimizer to implement L2 regularization.

If you want custom weight regularization, you can implement that in your loss function.

Upvotes: 1

Related Questions