Support Vector Machines Flashcards
What are SVM’s based upon?
A strong theoretical foundation, which is understood and provable
What can SVM’s be considered?
An optimisation problem
What were the two key developments that led to SVM’s?
- Computational learning theory
- Efficient separability of non linear functions, with the use of kernel functions
What is an advantage of SVMs over ANNs?
They can guarantee the globally optimal solution
What are the features of SVMs?
- Explicitly based on theoretical model
- Guarantees about performance
- Local minima not a problem
- Don’t suffer from dimensionality
What does an SVM return?
It returns an optimal hyperplane that separates the classes.
What are support vectors?
The data objects that lie closest to the decision surface.
Why are support vectors critical to this model?
As they are the objects in the training set that would change the position of the dividing hyperplane if removed.
They have direct bearing on the optimum location of the decision surface
Simply, what does a SVM attempt to do?
It attempts to maximise the margin (distance) between the separating hyperplane and all training objects.
What is the complexity of the maximisation problem determined by?
The number of objects, not the dimensionality of the feature space
What is the time and space complexity of the maximisation problem algorithm?
With n training objects,
Time: O(n^3)
Space: O(n^2)
