Anda di halaman 1dari 4

Statistical learning theory

See also: Computational learning theory

After learning a function based on the training set data,
This article is about statistical learning in machine learn- that function is validated on a test set of data, data that
ing. For its use in psychology, see Statistical learning in did not appear in the training set.
language acquisition.

2 Formal description

Statistical learning theory is a framework for machine

learning drawing from the elds of statistics and
functional analysis.[1][2] Statistical learning theory deals
with the problem of nding a predictive function based
on data. Statistical learning theory has led to successful applications in elds such as computer vision, speech
recognition, bioinformatics and baseball.[3]

Take X to be the vector space of all possible inputs, and

Y to be the vector space of all possible outputs. Statistical
learning theory takes the perspective that there is some
unknown probability distribution over the product space
Z = X Y , i.e. there exists some unknown p(z) =
p(x, y) . The training set is made up of n samples from
this probability distribution, and is notated

S = {(x1 , y1 ), . . . , (xn , yn )} = {z1 , . . . , zn }

The goals of learning are prediction and understanding. Learning falls into many categories, including
supervised learning, unsupervised learning, online learning, and reinforcement learning. From the perspective
of statistical learning theory, supervised learning is best
understood.[4] Supervised learning involves learning from
a training set of data. Every point in the training is an
input-output pair, where the input maps to an output.
The learning problem consists of inferring the function
that maps between the input and the output, such that the
learned function can be used to predict output from future

Every xi is an input vector from the training data, and yi

is the output that corresponds to it.
In this formalism, the inference problem consists of nding a function f : X 7 Y such that f (x) y . Let H
be a space of functions f : X Y called the hypothesis
space. The hypothesis space is the space of functions the
algorithm will search through. Let V (f (x), y) be the loss
functional, a metric for the dierence between the predicted value f (x) and the actual value y . The expected
risk is dened to be

Depending on the type of output, supervised learning

problems are either problems of regression or problems
of classication. If the output takes a continuous range of
values, it is a regression problem. Using Ohms Law as
an example, a regression could be performed with voltage
as input and current as output. The regression would nd
the functional relationship between voltage and current to
be R1 , such that


I[f ] =

V (f (x), y) p(x, y) dx dy

The target function, the best possible function f that can

be chosen, is given by the f that satises
I[f ] = inf I[h]

Because the probability distribution p(x, y) is unknown,

a proxy measure for the expected risk must be used.
This measure is based on the training set, a sample from
this unknown probability distribution. It is called the
empirical risk


Classication problems are those for which the output

will be an element from a discrete set of labels. Classication is very common for machine learning applications.
In facial recognition, for instance, a picture of a persons
face would be the input, and the output label would be
that persons name. The input would be represented by a
large multidimensional vector whose elements represent
pixels in the picture.

V (f (xi ), yi )
n i=1

IS [f ] =

A learning algorithm that chooses the function fS that

minimizes the empirical risk is called empirical risk minimization.


Loss functions

The choice of loss function is a determining factor on the

function fS that will be chosen by the learning algorithm.
The loss function also aects the convergence rate for
an algorithm. It is important for the loss function to be
Dierent loss functions are used depending on whether
the problem is one of regression or one of classication.



The most common loss function for regression is the

square loss function (also known as the L2-norm). This
familiar loss function is used in ordinary least squares regression. The form is:

V (f (x), y) = (y f (x))2
The absolute value loss (also known as the L1-norm) is
also sometimes used:

V (f (x), y) = |y f (x)|



Main article: Statistical classication

In some sense the 0-1 indicator function is the most natural loss function for classication. It takes the value 0
if the predicted output is the same as the actual output,
and it takes the value 1 if the predicted output is dierent from the actual output. For binary classication with
Y = {1, 1} , this is:

V (f (x), y) = (yf (x))

where is the Heaviside step function.


This image represents an example of overtting in machine learning. The red dots represent training set data. The green line represents the true functional relationship, while the blue line shows
the learned function, which has fallen victim to overtting.

variation in the learned function. It can be shown that

if the stability for the solution can be guaranteed, generalization and consistency are guaranteed as well.[6][7]
Regularization can solve the overtting problem and give
the problem stability.
Regularization can be accomplished by restricting the hypothesis space H . A common example would be restricting H to linear functions: this can be seen as a reduction
to the standard problem of linear regression. H could also
be restricted to polynomial of degree p , exponentials, or
bounded functions on L1. Restriction of the hypothesis
space avoids overtting because the form of the potential
functions are limited, and so does not allow for the choice
of a function that gives empirical risk arbitrarily close to
One example of regularization is Tikhonov regularization. This consists of minimizing
V (f (xi , yi )) + f 2H
n i=1

where is a xed and positive parameter, the regularIn machine learning problems, a major problem that ization parameter. Tikhonov regularization ensures exis[8]
arises is that of overtting. Because learning is a predic- tence, uniqueness, and stability of the solution.
tion problem, the goal is not to nd a function that most
closely ts the (previously observed) data, but to nd one
that will most accurately predict output from future input. 5 See also
Empirical risk minimization runs this risk of overtting:
nding a function that matches the data exactly but does
Reproducing kernel Hilbert spaces are a useful
not predict future output well.
choice for H .
Overtting is symptomatic of unstable solutions; a small
Proximal gradient methods for learning
perturbation in the training set data would cause a large


[1] Trevor Hastie, Robert Tibshirani, Jerome Friedman

(2009) The Elements of Statistical Learning, SpringerVerlag ISBN 978-0-387-84857-0.
[2] Mehryar Mohri, Afshin Rostamizadeh, Ameet Talwalkar
(2012) Foundations of Machine Learning, The MIT Press
ISBN 9780262018258.
[3] Gagan Sidhu, Brian Cao. Exploiting pitcher decisionmaking using Reinforcement Learning. Annals of Applied
[4] Tomaso Poggio, Lorenzo Rosasco, et al. Statistical Learning Theory and Applications, 2012, Class 1
[5] Rosasco, L., Vito, E.D., Caponnetto, A., Fiana, M., and
Verri A. 2004. Neural computation Vol 16, pp 1063-1076
[6] Vapnik, V.N. and Chervonenkis, A.Y. 1971. On the uniform convergence of relative frequencies of events to their
probabilities. Theory of Probability and its Applications
Vol 16, pp 264-280.
[7] Mukherjee, S., Niyogi, P. Poggio, T., and Rifkin, R. 2006.
Learning theory: stability is sucient for generalization
and necessary and sucient for consistency of empirical
risk minimization. Advances in Computational Mathematics. Vol 25, pp 161-193.
[8] Tomaso Poggio, Lorenzo Rosasco, et al. Statistical Learning Theory and Applications, 2012, Class 2


Text and image sources, contributors, and licenses



Statistical learning theory Source: Contributors: Michael

Hardy, Hike395, Jitse Niesen, Bearcat, Tomchiukc, Klemen Kocjancic, Rajah, Qwertyus, Chris the speller, Geach, ClydeC, Olaf, Nyq,
Katharineamy, VolkovBot, Ostrouchov, Melcombe, XLinkBot, Fgnievinski, MichalSylwester, Brightgalrs, RealityApologist, Olexa Riznyk,
OnceAlpha, Callanecc, Chire, Zephyrus Tavvier, Scientic29, ClueBot NG, BG19bot, Marcocapelle, Aisteco, Atastorino, MazinIssa, Francisbach, I3roly, Mgfbinae, Jala Daibajna, Slee325, Aetilley, Blu3r4y at and Anonymous: 24



File:Overfitting_on_Training_Set_Data.pdf Source:
Training_Set_Data.pdf License: Attribution Contributors:{}9.520/spring12/slides/class02/class02.pdf Original
artist: Tomaso Poggio
File:Portal-puzzle.svg Source: License: Public domain Contributors: ?
Original artist: ?


Content license

Creative Commons Attribution-Share Alike 3.0