Special Program Modern Statisti

randombroadΤεχνίτη Νοημοσύνη και Ρομποτική

15 Οκτ 2013 (πριν από 3 χρόνια και 5 μήνες)

79 εμφανίσεις

1


Special Program


Modern Statisti
cal Methods in Machine Learning”


18
-
22/6/2012
:
Lectures on multivariate statistics in machine learning

30/7
-
3/8/2012
:
Lectures on Bayesian nonparametric methods and probabilistic
graphical modes

6
-
17/8/2012
.
Discussion on s
tatistical machine learning and lectures of Professor
John Lafferty

1.

O
bjective

Machine learning is an interdisciplinary field which seeks to develop both the
mathematical foundations and practical applications of systems that learn, reason
and act. Machine
learning

is one of the most exciting areas in
computer science

that have witnessed tremendous developments in the past few decades.
Increasingly much of mo
d
ern technologies and sciences have come to rely on our
ability to analyze large
-
scale and complex da
ta, to draw inference and make
decisions in real
-
time, resource
-
constrained situations. M
a
chine learning plays a
central role at the forefront of this broad effort.

Recently,

this field has seen
enormous growth
of

researchers in machine learning gradually
recognized the
field’s broader, unique, and deeper intellectual roots coming from mathematical
statistics and probability theory, optimization, functional analysis, linear algebra,
approximation theory, etc.
Concurrent to the machine learning development,
mathematical statistics including the area of
multivariate data analysis

have also
witnessed substantial changes in the last decades, motivated by the progress of
computational sciences

and real
-
world engineering and scientific applications
involving large
r and more complex datasets. Statisticians and probabilists have
come to appreciate and embrace the roles of data structures, algorithms and
computational complexity, all of which are fundamental concepts of computer
science. The integration of machine lea
rning, modern st
a
tistics and probability
enables each of these fields to become more powerful.

The objective of the project is two
-
fold
:

(1)

To introduce to the machine learning community in Vietnam the statistical
machine learning methods developed in recent

years.

(2)

To offer an opportunity for researchers in machine learning in Vietnam to
work together at VIASM in contact with invited scientists in statistical
machine learning.


2


The project is organized by

-

Prof. Ho Tu Bao

(JAIST
,
http://www.jaist.ac.jp/~bao
)



-

Prof. Ngo Quang Hung

(SUNY at Buffalo,
http://www.cse.buffalo.edu/~hungngo
)

-

Prof. Nguyen Xuan Long

(Michigan Univ.,

http://www.stat.lsa.umich.edu/~xuanlong
)


Especially, the project has the participation of Professor John Lafferty from
Chicago
University
, a leading expert in statistical machine learning
(
http://newfaculty.uchicago.edu/psd/lafferty.shtml
).


2.

Program

The project activities consist of
:

a)

Lectues on a number of selected topics in statistical machine learning
,

b)

Research on the problems proposed by the project researchers,

c)

Discu
ssion on statistical learning topics with experts.

The project is carried out in the period from 18th June to 17 August, 2012 with the
following plan:

18
-
22
/6/2012
:
Lectures on multivariate statistics in machine learning



Intructor
:
Prof. Ho Tu Bao




The l
ectures aim to introduce to the development, recent directions and some
challenges in machine learning as well as some basis f
or statistical machine
learning
:

-

Machine learning: its roles in other sciences,
recent directions

and
some
challenges

-

Model assess
ment and selection

in mult
iple and multivariate regresion

-

Kernel methods and support vector machines

-

D
imensionality

reduction và

manifold learning

-

Topic models in text analysis.

25/6
-
27/7/2012
.
Project researchers work at VIASM



30/7
-
3/8/2012
:

Lectures o
n Bayesian nonparametric methods an
d probabilistic
graphical modes





Instructor
:
Prof. Nguyen Xuan

Long

3




These lectures introduce to the two above fields of statistical machine
learning
:

-

Infinite mixture models based on stick
-
breaking processes

-

Dirichlet p
rocesses, stick
-
breaking processes, Chinese restaurant processes

-

Markov Chain Monte Carlo
algorithms for infinite mixtur

-

Hierarchical nonparametric Bayes

-

Nonparametric Bayes for learning latent network structures

-

Asymptotic theory for statistical inference

in infinite mixtures

-

Variational inference and message
-
passing algorithms
.


6
-
17/8/2012
.
Discussion on statistical machine learning and lectures of Professor
John Lafferty on
:

-

Sparsity in regression

-

Graphical

model structure
learning

-

Nonparametric inferen
ce

-

Topic models