Machine Learning with Indefinite Kernels
Speaker: Mr Cheng Soon Ong
National ICT Australia
Time: 4:00p.m. Wednesday 9th March 2005
Venue: Red Center Room RC-4082
near Barker Street Gate 14
In many instances of machine learning problems, the solution can be found by
only using measures of similarity between objects in the domain. For example,kernel machines utilize only the kernel matrix to access the training data.
One of the conditions for kernel methods is that the kernel has to be positive
This talk will be about our recent work on this area. It covers two
- What is the functional framework for kernels which are not positive semidefinite?
- How can an algorithm perform fast and efficient regularization in this framework?