Bayesian integration of multi-way, multi-species, and time-series metabolomic datasets

Lecturer : 
Ilkka Huopaniemi
Event type: 
HIIT seminar
Event time: 
2010-11-26 10:15 to 11:00
Place: 
Kumpula Exactum C222
Description: 
Talk announcement:
HIIT Seminar Kumpula, Friday Nov 26, 10:15 a.m., Exactum C222

SPEAKER:
Ilkka Huopaniemi
Aalto University

TITLE:
Bayesian integration of multi-way, multi-species, and time-series
metabolomic datasets


ABSTRACT:
Multi-way analysis of variance (ANOVA) - type methods are the default
tool for modelling the effects of multiple covariates (disease,
treatments, gender, time-series) in populations of (biomedical)
continues-valued measurements. I present a multivariate Bayesian
modelling framework for multi-way modelling, that can deal with the main
restriction of modern biomedical data: small sample-size and high
dimensionality. I then describe how we've extended this framework to
analyze data from novel biomedical multi-way experiment types: (i)
integrating multiple data sources, (ii) integrating data from multiple
species, and (iii) time-series experiment with mixed aging- and disease
progression effects.


BIO:
Ilkka Huopaniemi is a PhD student at the Department of Information and
Computer Science at Aalto University, in the Statistical Machine
Learning and Bioinformatics Group lead by Samuel Kaski. He received his
M.Sc. degree in 2006 from the Department of Technical Physics and
Mathematics of TKK, and did his Master's thesis in the Statistical
Physics group. He's research interests are multi-way experimental
designs, Bayesian methods, metabolomics, data integration, translational
modelling.

 

Enabling technologies for decentralized interpersonal communication

Lecturer : 
Jani Hautakorpi
Event type: 
Doctoral dissertation
Event time: 
2010-11-12 12:15 to 14:15
Place: 
Auditorium T2 at the Aalto University School of Science and Technology, Espoo, Finland
Description: 

Dissertation for the degree of Doctor of Science in Technology to be presented with due permission of the Faculty of Information and Natural Sciences for public examination and debate in Auditorium T2 at the Aalto University School of Science and Technology (Espoo, Finland) on the 12th of November 2010 at 12 noon.

Opponent: Professor Thomas Plageman, University of Oslo

Custos: Professor Antti Ylä-Jääski, Aalto University
 

Securing the Internet with digital signatures

Lecturer : 
Dmitrij Lagutin
Event type: 
Doctoral dissertation
Event time: 
2010-12-10 12:00 to 15:00
Place: 
Computer science building, hall T2, Konemiehentie 2, Otaniemi, Espoo
Description: 

Opponent: Associate Professor Panagiotis Papadimitratos, KTH School of Electrical Engineering, Sweden
Supervisor: Professor Antti Ylä-Jääski

The dissertation will be available at:
http://lib.tkk.fi/Diss/2010/isbn9789526034652/

Abstract:
The security and reliability of the Internet are essential for many functions of a modern society. Currently, the Internet lacks efficient network level security solutions and is vulnerable to various attacks, especially to distributed denial-of-service attacks. Traditional end-to-end security solutions such as IPSec only protect the communication end-points and are not effective if the underlying network infrastructure is attacked and paralyzed.

This thesis describes and evaluates Packet Level Authentication (PLA), which is a novel method to secure the network infrastructure and provide availability with public key digital signatures. PLA allows any node in the network to verify independently the authenticity and integrity of every received packet, without previously established relationships with
the sender or intermediate nodes that have handled the packet. As a result, various attacks against the network and its users can be more easily detected and mitigated, before they can cause significant damage or disturbance. PLA is compatible with the existing Internet infrastructure, and can be used with complementary end-to-end security solutions, such as IPSec and HIP. While PLA was originally designed for securing current IP networks, it is also suitable for securing future data-oriented networking approaches.

PLA has been designed to scale from lightweight wireless devices to Internet core network, which is a challenge since public key cryptography operations are very resource intensive. Nevertheless, this work shows that digital signature algorithms and their hardware implementations developed for PLA are scalable to fast core network routers. Furthermore, the additional energy consumption of cryptographic operations is significantly lower than the energy cost of wireless transmission, making PLA feasible for lightweight wireless devices. Digital signature algorithms used by PLA also offer small key and signature sizes and therefore PLA's bandwidth overhead is relatively low.

Strong security mechanisms offered by PLA can also be utilized for various other tasks. This work investigates how PLA can be utilized for controlling incoming connections, secure user authentication and billing, and for providing a strong accountability without an extensive data retention by network service providers.

Cross-layer Assisted TCP Algorithms for Vertical Handoff

Event type: 
Doctoral dissertation
Doctoral dissertation
Respondent: 
Laila Daniel
Opponent: 
Professor Xiaoming Fu
Custos: 
Professor Jussi Kangasharju
Event time: 
2010-11-20 10:00 to 12:00
Place: 
Auditorium XII in the Main Building
Description: 

Laila Daniel will defend her thesis "Cross-layer Assisted TCP Algorithms for Vertical Handoff" on Saturday,

20 November 2010 at 10.00. in Auditorium XII in the Main Building. Her opponent is Professor Xiaoming Fu,

Georg-August-University of Göttingen, Germany, and her custos Professor Jussi Kangasharju.

 

 Popular abstract:

Increasingly we use our mobile devices (e.g., mobile phones, laptops) for diverse applications such as
reading e-mail, browsing the web, downloading and listening to music, playing games and for making payments
for products and services irrespective of our location and mobility by connecting the mobile device to
the Internet anytime anywhere.

Access networks that enable a mobile device to connect to the Internet use diverse technologies and differ
widely in their characteristics. Mobile devices inherently use wireless access networks by means of radios
for Internet connectivity. For example, Wireless Local Area Network (WLAN) is a fast network that can be used
inside a building whereas General Packet Radio Service (GPRS) is comparatively slower and can span a city or
a country and even beyond. A mobile device with multiple radio interfaces can changeover to any of the several
access networks depending on its location or the application needs. Handoff refers to this changeover and it
is known as vertical handoff when the underlying technologies of the two access networks are different.

TCP, a data communication software, which resides both at the sender and the receiver of data, delivers the
application data reliably and also adjusts its sending rate depending on the availability of the resources
in the Internet. TCP behaviour depends on the characteristics of the end-to-end path and in particular the
bottleneck link, the link with minimum capacity in the path. A wireless link that connects the mobile device
to the Internet is often the bottleneck link and the abrupt change in its characteristics due to a vertical
handoff significantly affects TCP performance and consequently the application quality.

The focus of this thesis is to study TCP behaviour in a vertical handoff and to devise algorithms to improve
its handoff performance using cross-layer notification to convey information about the characteristics of the
access links involved in the handoff. Our first study is in the WLAN-GPRS environment with minimum information to
TCP about the handoff and the results show that TCP performance can be significantly improved. Subsequently we
enlarge the scope of the study to cover a more general case using rough estimates of the access link parameters.
The evaluation of the algorithms is based on simulation experiments with a wide range of access networks showing
that TCP performance in handoff can be significantly improved using this approach. Our algorithms can be useful
to devise solutions for the real world scenarios.

Statistical dependencies in analysis of naturalistic brain stimulation

Lecturer : 
Arto Klami
Event type: 
HIIT seminar
Event time: 
2010-11-19 10:15 to 11:00
Place: 
Exactum C222
Description: 

Talk announcement:
HIIT Seminar Kumpula, Friday Nov 19, 10:15 a.m., Exactum C222

On Nov 19 HIIT Seminar Kumpula features a talk by Arto Klami from the HIIT Statistical Machine Learning and Bioinformatics group. The presentation will include a short and accessible overview of the group.

Welcome!
--Matti Järvisalo

---------------------------------------------------------------------


SPEAKER:
Arto Klami
Helsinki Institute for Information Technology HIIT / Aalto University

TITLE:
Statistical dependencies in analysis of naturalistic brain stimulation

ABSTRACT:
Use of more naturalistic experimental conditions is currently one of the major trends in neuroscience. Instead of controlled experiments the brains are being scanned for example when the subject is watching a movie. Naturalistic stimulation opens up new possibilities for understanding the brain, but the classical analysis tools are not sufficient for these new setups. I will introduce a new approach based on extracting statistical dependencies between brain activity and rich feature representations of the stimulus and discuss the necessary modeling tools.

BIO:
Arto Klami is a postdoc researcher at the Department of Information and Computer Science at Aalto University. He received his PhD in computer science at Helsinki University of Technology in 2008, and is currently working on an Academy of Finland postdoctoral researcher's project. His research interests include Bayesian modeling, proactive interfaces, and analysis of neuroimaging data.

Pages