Standard learning systems operate on input data after they have been transformed into feature vectors. But in many cases the input data cannot be described easily by explicit feature vectors for example in bio-sequences, images etc. Moreover, the construction of Feature extraction module can be complex and even risk losing relevant information during the process.
A solution is in the form of Kernel methods/functions. A Kernel function maps the data into a high dimension space where classes are linearly separable. The High dimension data (e.g. text) can use a simple dot-product as kernel. The Optimisation algorithm only accesses the training data through the
Kernel Matrix.
A kernel k(x,y) is a similarity measure, defined by implicit mapping f from the original space to a vector space (feature space) : k(x,y)=f(x)•f(y)
This similarity measure and the mapping include:
check out for further information on Kernel methods, SVM amd Machine learning
Here are some pointers which include course materials on Machine Learning done this summer in some Universities.
A solution is in the form of Kernel methods/functions. A Kernel function maps the data into a high dimension space where classes are linearly separable. The High dimension data (e.g. text) can use a simple dot-product as kernel. The Optimisation algorithm only accesses the training data through the
Kernel Matrix.
A kernel k(x,y) is a similarity measure, defined by implicit mapping f from the original space to a vector space (feature space) : k(x,y)=f(x)•f(y)
This similarity measure and the mapping include:
- Invariance or other a priori knowledge
- Simpler structure (linear representation of the data)
check out for further information on Kernel methods, SVM amd Machine learning
Here are some pointers which include course materials on Machine Learning done this summer in some Universities.
- This is my favourite starter page for Statistics.
- Stanford
- MIT
- Carnegie Mellon
- Trento
No comments:
Post a Comment