ComputerAlgebra Systems in Teaching Basics of Statistical Learning

spraytownspeakerΤεχνίτη Νοημοσύνη και Ρομποτική

16 Οκτ 2013 (πριν από 3 χρόνια και 9 μήνες)

86 εμφανίσεις




Тезисы

доклада


1.

НАЗВАНИЕ

ДОКЛАДА
:

Computer

Algebra Systems in Teaching Basics of
Statistical Learning

-


2.

АВТОРЫ:

Aleksandr Mylläri
1)

and Tatiana Mylläri
2)


3.

ОРГАНИЗАЦИЯ

(полное наименование, без аббре
виатур):


1)
University

of

Turku

2)
Å
bo

Akademi

University


4.

ГОРОД:

Turku/
Åbo


5.

ТЕЛЕФОН:

+358
-
40
-
7524003


6.

ФАКС:


7.

E
-
mail
:

alemio@utu.fi


8.

ТЕКСТ ТЕЗИСОВ ДОКЛАДА:

Modern computer algebra systems not only make calculations (analytic and numeric)
easy, but also have ready tools for solving optimization

problems and good facilities for
visualization. Visual demonstrations show visually the work of algorithms, and this facilitates
thei
r being understood by students.
The models considered in our paper are used in the
introductory course on Machine Learning

and Support Vector Machines (SVMs). This course
is based on the classic book by N. Cristianini and J. Shave
-
Taylor (Cristianini and Shave
-
Taylor (2000); see also Schlkopf and Smola (2002)). We consider the problem of binary
classification. SVMs are attra
ctive pedagogically since one can introduce them gradually,
starting with simple perceptrons and moving on to more advanced classifiers. We start with
Rosenblatt’s perceptron, a simple binary classifier for linearly separable cases, and generalize
it to a
maximum margin classifier; then we introduce a kernel trick that allows generalization
to non
-
linearly separable cases; finally we accept misclassifications in the training stage. In
our introductory course on SVMs and Statistical Learning, we use
Mathemat
ica

6 and
Maple

12 to
construct models of Rosenblatt’s perceptron and simple SVMs

Algorithms for solving optim
ization problems together with
good visualization
facilities make
Mathematica

a convenient

environment for introductory courses on SVMs and
Stati
stical Learning. With
Mathematica

it is easy to implement basic SVMs and to experiment
with different parameter values, etc. Codes are transparent and work fast for toy examples.


REFERENCES

Cristianini, N. & Shave
-
Taylor J. (2000) An Introduction to Support Vector Machines,
Cambridge Univ. Press.

Schölkopf, B. and Smola, A. J. (2002)
Learning with Kernels
. MIT Press, Cambridge, MA.