Thursday, July 24, 2014

Reproducing Kernel Hilbert Space (RKHS)



A very good tutorial of RKHS:


Terminology:
  • Vector space - A set endowed with two special operations: addition and scalar multiplication.
  • Normed vector space - A vector space with length of vectors defined.
  • Metric space - A set (nor necessarily a vector space) where the distance between two points in the set is defined. A normed vector space is a metric space. But the converse is not true.
  • Banach space - A complete normed vector space.
  • Inner product space - An inner product space, (V,.,.) is a vector space V over a field F with a binary operator .,.:V×VF, known as the inner product, that satisfies the following axioms for all vectors x,y,zV and all scalars aF.

    1. Conjugate symmetry: x,y=y,x⟩*
    2. Linearity in the first argument: ax,y=ax,y and x+z,y=x,y+z,y
    3. Positive definiteness:  x,x0 where the equality holds only when x=0.

  • An inner product space is also known as a pre-Hilbert space
Hilbert space:
  • Hilbert space - An inner product space which is complete with respect to the norm induced by its inner product. 
Reproducing kernel:
  • Let X be a nonempty set and H⊆CX be a Hilbert space of complex-valued functions defined on X. A kernel k:X×X→C is called a reproducing kernel of H if it has the two properties:
    1.     For every x0∈X, k(y,x0) as a function of y belongs to H.

    2.     The reproducing property: for each x0∈X and f∈H, f(x0)=⟨f(.),k(.,x0)⟩H
Reproducing Kernel Hilbert Space (RKHS):
  • Reproducing Kernel Hilbert Space (RKHS) - A Hilbert space H of complex-valued functions on a nonempty set X is called a reproducing kernel Hilbert space if the point evaluation functional Lx is a bounded (equivalently, continuous) linear operator for all xX
  • Point evaluation functional - The point evaluation functional Lx:HC at a point xX is defined as Lx(f):=f(x).
  • Linear operator (Linear map) - Let f:VW be a function between two vector spaces V,W over the same field F. f is called a linear operator if the following two conditions are satisfied for all x,yV and aF.
1. f(u+v)=f(u)+f(v)
2. f(au)=af(u)





No comments:

Post a Comment