Primal-dual algorithms for distributed optimization

Lecturer : 
André Schumacher
Event type: 
HIIT seminar
Event time: 
2011-02-18 10:15 to 11:00
Place: 
Kumpula Exactum C222
Description: 
Talk announcement:
HIIT Seminar Kumpula, Friday Feb 18 10:15, Exactum C222

SPEAKER:
André Schumacher
Aalto University

TITLE:
Primal-dual algorithms for distributed optimization

ABSTRACT:
Recently, it was discovered that some protocols for computer networks
can be seen as algorithms that implicitly solve an optimization problem,
which characterizes optimal states of the distributed system.
Many of these protocols can be modeled as algorithms that simultaneously
solve a pair of primal and dual problems. The underlying "locality by
duality" principle was already exploited to design distributed algorithms
for various optimization problems.

In this talk I will highlight the algorithmic ideas that form the basis
for distributed primal-dual algorithms. I will then give an example in
the form of an algorithm for the minimum-weight dominating-set
problem that was proposed as part of an algorithmic scheme for
lifetime maximization in wireless sensor networks. This second part
of the talk is based on joint research carried out at the Department of
Information and Computer Science of Aalto University School of Science.

BIO:
André Schumacher received his doctoral degree from Aalto University
School of Science and Technology in 2010 and is currently a researcher
with the Distributed Computation group at the Department of
Information and Computer Science of Aalto University. His research
interests include distributed algorithms, approximation and online
algorithms, network optimization, as well as adhoc and sensor networks.



Welcome!
--Matti Järvisalo


HIIT SEMINAR TENTATIVE SCHEDULE Spring 2011
-------------------------------------------
Feb 18: André Schumacher
Feb 25: Jose A. Fernandes
Mar  4: Petteri Kaski
Mar 11: Esther Galbrun
Mar 18: Valentin Polishchuk
Mar 25: Esa Junttila
Apr  1: *** free ***
Apr  8: *** free ***
Apr 15: *** free ***

Local Algorithms: Past, Present, Future

Lecturer : 
Jukka Suomela
Event type: 
HIIT seminar
Event time: 
2011-02-14 13:15 to 14:00
Place: 
Computer Science Building, Hall T2
Description: 
Our next speaker for HIIT Otaniemi seminar series is Jukka Suomela from the "New Paradigms in Computing" group of the Helsinki Institute for Information Technology HIIT.
 
All ICS@Aalto researchers are also warmly welcome to attend the seminar!
 
HIIT Otaniemi Seminar, Monday February 14, 13:15
Location: Computer Science Building, Hall T2
 
Jukka Suomela
New Paradigms in Computing Group
Helsinki Institute for Information Technology HIIT
 
Title:
Local Algorithms: Past, Present, Future
 
Abstract:
A local algorithm is a distributed algorithm that runs in constant time, independently of the size of the network. Being highly scalable and fault-tolerant, such algorithms are ideal in the operation of large-scale distributed systems such as computer networks.
 
Even though the model of local algorithms is very limited, in recent years we have seen many positive results for non-trivial problems. In this talk, I will give an overview of the state-of-the-art in the field of local algorithms. I will show how the work in our group has advanced the field, and I will explore the current frontiers and fundamental open questions.
 
Welcome!
 
UPCOMING TALKS OF HIIT OTANIEMI SEMINAR SERIES
----
Feb. 21, Patrik Hoyer
Mar. 07, Petteri Kaski
Mar. 14, Simon Rogers
 
 
-- 
Mehmet Gönen
Helsinki Institute for Information Technology HIIT
Department of Information and Computer Science
Aalto University School of Science

Protein-protein network inference with regularized output and input kernel methods

Lecturer : 
Florence d'Alché Buc
Event type: 
HIIT seminar
Event time: 
2011-02-11 10:15 to 11:00
Place: 
Kumpula Exactum C222
Description: 
Talk announcement:
Combined Guest Lecture / HIIT Seminar Kumpula, Friday Feb 11 10:15, Exactum C222

SPEAKER:
Florence d'Alché Buc
IBISC, Université d'Evry-Val d'Essonne, Evry, France

TITLE:
Protein-protein network inference with regularized output and input  
kernel methods

Prediction of a physical interaction between two proteins has been  
addressed in the context of supervised learning, unsupervised  
learning and more recently, semi-supervised learning using various  
sources of information (genomic, phylogenetic, protein localization  
and function). The problem can be seen as a kernel matrix completion  
task if one defines a kernel that encodes similarity between  
proteins as nodes in a graph or alternatively, as a binary  
supervised classification task where inputs are pairs of proteins.
In this talk, we first make a review of existing works (matrix  
completion, SVM for pairs, metric learning, training set expansion),  
identifying the relevant features of each approach. Then we define  
the framework of output kernel regression (OKR) that uses the kernel  
trick in the output feature space and we develop a new family of  
methods based on Kernel Ridge Regression that benefit from the use  
of kernels both in the input feature space and the output feature  
space. The main interest of such methods is that imposing various  
regularization constraints still leads to closed form solutions. We  
show especially how such an approach allows to handle unlabeled data  
in a transductive setting of the network inference problem.
New results on simulated data and yeast data illustrate the talk.

Joint work with Céline Brouard and Marie Szafranski.


Welcome!
--Matti Järvisalo

Supervised classification applications in fisheries management

Lecturer : 
Jose A. Fernandes
Event type: 
HIIT seminar
Event time: 
2011-02-25 10:15 to 11:00
Place: 
Kumpula Exactum C222
Description: 
Talk announcement:
HIIT Seminar Kumpula, Friday Feb 25 10:15, Exactum C222

SPEAKER:
Jose A. Fernandes
University of Helsinki and
University of Basque Country (ISG group), AZTI-Tecnalia

TITLE:
Supervised classification applications in fisheries management

ABSTRACT:
The impact of how fisheries are managed is of great importance on
biological, economic, social and political levels. However, there is
still a high uncertainty about the relationships between climate, fish
and management decisions. Many activities are performed in marine
science in order to reduce this uncertainty. Classification methods
have  high potential to aid in this activities.

In this presentation three classification applications are presented:
1) Wrapper method to acomplish a trade-off between number of
Zooplakton taxa and performance.
2)  A methodological pipeline of machine learning state-of-the-art
methods is proposed and
its proper application is verified for fish recruitment forecasting.
3) The new machine learning paradigm of multi-dimensional classifiers
is applied to simultaneous multi-species recruitment forecasting in
the context of an ecosystem-based approach to fisheries management.

BIO:
Computer Engineer from the University of Deusto (Spain), MSc in
Computer Systems Security from the University of Glamorgan (Wales). He
will defend his PhD in March 2011, result of a collaboration between
AZTI-Tecnalia Food and Marine Research fundation (Spain)  and the
Department of Computer Science and AI at University of Basque Country
in the ISG group (Spain). Ph.D. in the field of Supervised
Classification problems in Marine Science, mainly in problems with
sparse data and high uncertainty. Publications in refereed journals
and international symposiums(
www.sc.ehu.es/ccwbayes/members/jafernandes/), focusing on develop
machine learning methodologies to applied problems in journals such as
Ecological Modelling and Journal of Plankton Research, leading to
reviewing for these journals and other refereed journals as well as
for the NSF and NOAA (EEUU).


Welcome!
--Matti Järvisalo

Local Algorithms: Past, Present, Future

Lecturer : 
Jukka Suomela
Event type: 
HIIT seminar
Event time: 
2011-02-04 10:15 to 11:00
Place: 
Kumpula Exactum C222
Description: 
Talk announcement:
HIIT Seminar Kumpula, Friday Feb 4 10:15, Exactum C222

SPEAKER:
Jukka Suomela
University of Helsinki

TITLE:
Local Algorithms: Past, Present, Future

ABSTRACT:
A local algorithm is a distributed algorithm that runs in constant time,
independently of the size of the network. Being highly scalable and
fault-tolerant, such algorithms are ideal in the operation of large-scale
distributed systems such as computer networks.

Even though the model of local algorithms is very limited, in recent years
we have seen many positive results for non-trivial problems. In this talk,
I will give an overview of the state-of-the-art in the field of local
algorithms. I will show how the work in our group has advanced the field,
and I will explore the current frontiers and fundamental open questions.


Welcome!
--Matti Järvisalo


HIIT SEMINAR TENTATIVE SCHEDULE Spring 2011
-------------------------------------------
Feb  4: Jukka Suomela
Feb 11: Florence d'Alche-Buc
Feb 18: André Schumacher
Feb 25: Petteri Kaski
Mar  4: *** free ***
Mar 11: Esther Galbrun
Mar 18: Valentin Polishchuk
Mar 25: Esa Junttila
Apr  1: *** free ***

Pages