Focused Multi-task Learning Using Gaussian Processes

Lecturer : 
Jaakko Peltonen
Event type: 
HIIT seminar
Event time: 
2011-09-19 13:15 to 14:00
Place: 
Computer Science Building, Hall T2
Description: 

Our next speaker for HIIT Otaniemi seminar series is Jaakko Peltonen from the "Statistical Machine Learning and Bioinformatics" 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 September 19, 13:15
Location: Computer Science Building, Hall T2

Jaakko Peltonen
Statistical Machine Learning and Bioinformatics Group
Helsinki Institute for Information Technology HIIT
Department of Information and Computer Science
Aalto University School of Science

Title:
Focused Multi-task Learning Using Gaussian Processes

*** This work by Gayle Leen, Jaakko Peltonen, and Samuel Kaski won the Award for Best Paper in Machine Learning at ECML PKDD 2011, the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases. ***

Abstract:
Given a learning task for a data set, learning it together with related tasks (data sets) can improve performance. Gaussian process models have been applied to such multi-task learning scenarios, based on joint priors for functions underlying the tasks. In previous Gaussian process approaches, all tasks have been assumed to be of equal importance, whereas in transfer learning the goal is asymmetric: to enhance performance on a target task given all other tasks. In both settings, transfer learning and joint modeling, negative transfer is a key problem: performance may actually decrease if the tasks are not related closely enough. In this paper, we propose a Gaussian process model for the asymmetric setting, which learns to “explain away” non-related variation in the additional tasks, in order to focus on improving performance on the target task. In experiments, our model improves performance compared to single-task learning, symmetric multi-task learning using hierarchical Dirichlet processes, and transfer learning based on predictive structure learning.

Welcome! 

-- 
Mehmet Gönen
Helsinki Institute for Information Technology HIIT
Department of Information and Computer Science
Aalto University School of Science

Deconstructing Approximate Offsets

Lecturer : 
Eric Berberich
Event type: 
HIIT seminar
Event time: 
2011-09-09 10:15 to 11:00
Place: 
Kumpula Exactum B222
Description: 
Talk announcement
HIIT Seminar Kumpula, Friday September 9 10:15, Exactum B222

SPEAKER:
Eric Berberich
Max-Planck-Institut Informatik MPII

TITLE:
Deconstructing Approximate Offsets

ABSTRACT:
Today's problem arised in a wood-cutting company that got
a new cutter. The new tools needs a smaller safety-margin
to the desired cutout. However, the only available legacy
data for the old cutter is an approximated offset, that is,
it is desired to regain the original shape before
offsetting.

Formally we study the following problem:
Can a polygonal shape Q with n vertices be expressed, up
to a tolerance eps in Hausdorff distance, as the Minkowski
sum of another polygonal shape P with a disk of fixed
radius? If it does, we also seek a preferably simple
solution shape P; its offset constitutes an accurate,
vertex-reduced and smoothened approximation of Q. We give
a decision algorithm for fixed radius in O(n log n) time
that handles any (bounded or unbounded, connected or
disconnected) polygonal shape. For convex shapes, the
complexity drops to O(n), and within the same bound, we
compute a solution shape P with at most one more vertex
than a vertex-minimal one. The talk also discusses
recent achievements as to use only rational arithmetic
for the decision and the search of good parameters
eps and r.

Sifting through Images with Multinomial Relevance Feedback

Lecturer : 
Dorota Glowacka
Event type: 
HIIT seminar
Event time: 
2011-06-17 10:15 to 11:00
Place: 
Kumpula Exactum C222
Description: 

Talk announcement:
HIIT Seminar Kumpula, Friday June 17 10:15, Exactum C222

Speaker:
Dr Dorota Glowacka
University College London

Title:
Sifting through Images with Multinomial Relevance Feedback

Abstract:

This talk presents the theory, design principles, implementation and
performance results of a content-based image retrieval system based on
multinomial relevance feedback. The system relies on an interactive
search paradigm in which at each round a user is presented with
a set of k images and is required to select one that is closest to
their target. Performance is measured by the number of rounds needed
to identify a specific target image  as well as the the average
distance from the target of the set of k images presented to the
user at each iteration. Results of experiments involving simulations
as well as real users are presented. The conjugate prior Dirichlet
distribution is used to model the problem motivating an algorithm that
trades exploration and exploitation in presenting the images in each
round. A sparse data representation makes the algorithm scalable.
Experimental results show that the new approach compares favourably with
previous work.


Welcome!

HIIT and new Academy of Finland CoE's

Sun, 12.06.2011

Academy of Finland has selected its new 15 Centres of Excellence in Research 2012‒2017. HIIT contributes to three:

Samuel Kaski, Petri Myllymäki, Ilkka Niemelä: Finnish Centre of Excellence in Computational Inference Research

Aapo Hyvärinen: Finnish Centre of Excellence in Inverse Problems Research 

Veli Mäkinen: Finnish Centre of Excellence in Cancer Genetics Research

Congratulations!

 

Negotiated interaction with computers: sensing, inference and multimodal feedback

Lecturer : 
Professor Roderick Murray-Smith
Event type: 
Guest lecture
Event time: 
2011-06-09 13:00 to 14:00
Place: 
E332, Innopoli 2
Description: 

 Abstract: This talk will focus on the 'negotiated interaction' approach to the design of human-computer interaction. This approach incorporates sensing, signal processing, probabilistic inference and multimodal feedback to shape the human-computer interaction loop. I will present examples from the mobile interaction, brain-computer interaction, and music recommendation areas.


Bio: Roderick Murray-Smith is a Professor of Computing Science in the Department of Computing Science at Glasgow University, where he runs the Dynamics and Interaction research group. He works in the overlap between machine learning, interaction design and control theory. Prior to this he has held positions at the Hamilton Institute, NUIM, Technical University of Denmark, M.I.T., and Daimler-Benz Research, Berlin. He works closely with the mobile phone industry, Syntonetic A/s and is a member of Nokia's Scientific Advisory Board.

Pages