---

Due     March 26, 2025 by 11:59pm     Points     100     Submitting     a Google doc shared with Louisa (lconwill@nd.edu)

---

Today in class we have discussed two popular approaches to extend vanilla linear SVM to a multi-class linear classifier. Both one-vs-one and one-vs-rest (or one vs-all) approaches may end up with an arrangement of separating hyperplanes that will create a triangle middle region you see below:

If an unknown sample to be classified (black dot) falls into that region, assigning the class selected by the largest number of SVMs (as their "positive" class) won't end up with a conclusive decision. Morever, a sample that ends up in that region sometimes doesn't belong to any of the known classes.

And here's the homework question: provide an idea for a solution to this problem making use of an alternative linear SVM formulation and a fusion strategy considering multiple SVM outputs that is able to correctly determine that a sample doesn't belong to a known class. Hint: there is a common form of SVM classifier that applies to this scenario.