|
|
| Line 1: |
Line 1: |
| In [[computational learning theory]], '''probably approximately correct learning''' ('''PAC learning''') is a framework for mathematical analysis of [[machine learning]]. It was proposed in 1984 by [[Leslie Valiant]].<ref name="valiant">L. Valiant. ''[http://web.mit.edu/6.435/www/Valiant84.pdf A theory of the learnable.]'' Communications of the ACM, 27, 1984.</ref>
| |
|
| |
|
| In this framework, the learner receives samples and must select a generalization function (called the ''hypothesis'') from a certain class of possible functions. The goal is that, with high probability (the "probably" part), the selected function will have low [[generalization error]] (the "approximately correct" part). The learner must be able to learn the concept given any arbitrary approximation ratio, probability of success, or [[Empirical distribution function|distribution of the samples]].
| |
|
| |
|
| The model was later extended to treat noise (misclassified samples).
| | Aleta is what's written [http://Www.Alexa.com/search?q=relating&r=topsites_index&p=bigtop relating] to her birth certificate life style she doesn't really like being called like that. Managing people is considered what she does on the other hand she plans on substituting it. She's always loved living into South Carolina. To steer is something her hubby doesn't really like though she does. She is running and looking after a blog here: http://prometeu.net<br><br>Here is my web-site [http://prometeu.net clash of clans hack tool] |
| | |
| An important innovation of the PAC framework is the introduction of [[computational complexity theory]] concepts to machine learning. In particular, the learner is expected to find efficient functions (time and space requirements bounded to a [[polynomial]] of the example size), and the learner itself must implement an efficient procedure (requiring an example count bounded to a polynomial of the concept size, modified by the approximation and [[likelihood]] bounds).
| |
| | |
| == Definitions and terminology ==
| |
| | |
| In order to give the definition for something that is PAC-learnable, we first have to introduce some terminology.<ref>Kearns and Vazirani, pg. 1-12,</ref> <ref>Balas Kausik Natarajan, Machine Learning , A Theoretical Approach, Morgan Kaufmann Publishers, 1991</ref>
| |
| | |
| For the following definitions, two examples will be used. The first is the problem of [[character recognition]] given an array of <math>n</math> bits encoding a binary-valued image. The other example is the problem of finding an interval that will correctly classify points within the interval as positive and the points outside of the range as negative.
| |
| | |
| Let <math>X</math> be a set called the ''instance space'' or the encoding of all the samples, and each ''instance'' have length assigned. In the character recognition problem, the instance space is <math>X=\{0,1\}^n</math>. In the interval problem the instance space is <math>X=\mathbb{R}</math>, where <math>\mathbb{R}</math> denotes the set of all real numbers.
| |
| | |
| A ''concept'' is a subset <math>c \subset X</math>. One concept is the set of all patterns of bits in <math>X=\{0,1\}^n</math> that encode a picture of the letter "P". An example concept from the second example is the set of all of the numbers between <math>\pi/2</math> and <math>\sqrt{10}</math>. A ''[[concept class]]'' <math>C</math> is a set of concepts over <math>X</math>. This could be the set of all subsets of the array of bits that are skeletonized 4-connected (width of the font is 1).
| |
| | |
| Let <math>EX(c,D)</math> be a procedure that draws an example, <math>x</math>, using a probability distribution <math>D</math> and gives the correct label <math>c(x)</math>, that is 1 if <math>x \in c</math> and 0 otherwise.
| |
| | |
| Say that there is an algorithm <math>A</math> that given access to <math>EX(c,D)</math> and inputs <math>\epsilon</math> and <math>\delta</math> that, with probability of at least <math>1-\delta</math>, <math>A</math> outputs a hypothesis <math>h \in C</math> that has error less than or equal to <math>\epsilon</math> with examples drawn from <math>X</math> with the distribution <math>D</math>. If there is such an algorithm for
| |
| every concept <math>c \in C</math>, for every distribution <math>D</math> over <math>X</math>, and for all <math>0<\epsilon<1/2</math> and <math>0<\delta<1/2</math> then <math>C</math> is '''PAC learnable''' (or ''distribution-free PAC learnable''). We can also say that <math>A</math> is a '''PAC learning algorithm''' for <math>C</math>.
| |
| | |
| An algorithm runs in time <math>t</math> if it draws at most <math>t</math> examples and requires at most <math>t</math> time steps. A concept class is '''efficiently PAC learnable''' if it is ''PAC learnable'' by an algorithm that runs in time polynomial in <math>1/\epsilon</math>, <math>1/\delta</math> and ''instance'' length.
| |
| | |
| == Equivalence ==
| |
| | |
| Under some regularity conditions these three conditions are equivalent:
| |
| | |
| # The concept class ''C'' is '''PAC learnable'''.
| |
| # The [[VC dimension]] of ''C'' is finite.
| |
| # ''C'' is a uniform [[Glivenko-Cantelli class]].
| |
| | |
| == References ==
| |
| <references/>
| |
| | |
| == Further reading ==
| |
| * M. Kearns, U. Vazirani. ''An Introduction to Computational Learning Theory.'' MIT Press, 1994. A textbook.
| |
| * D. Haussler. [http://www.cs.iastate.edu/~honavar/pac.pdf Overview of the Probably Approximately Correct (PAC) Learning Framework]. An introduction to the topic.
| |
| * L. Valiant. [http://www.probablyapproximatelycorrect.com/ ''Probably Approximately Correct.''] Basic Books, 2013. In which Valiant argues that PAC learning describes how organisms evolve and learn.
| |
| | |
| [[Category:Computational learning theory]]
| |
Aleta is what's written relating to her birth certificate life style she doesn't really like being called like that. Managing people is considered what she does on the other hand she plans on substituting it. She's always loved living into South Carolina. To steer is something her hubby doesn't really like though she does. She is running and looking after a blog here: http://prometeu.net
Here is my web-site clash of clans hack tool