PhD School in Advanced Machine Learning: Partially Supervised Learning

11th VIPS Advanced School on Computer Vision, Pattern Recognition, and
Image Processing, Dipartimento di Informatica, Universit? di Verona,
Verona (Italy) 19-21 May 2014 ‘);” target=”_blank”>

This school follows the series of intensive courses, aimed at PhD
students and researchers in the areas of Computer Vision, Image
Processing, and Pattern Recognition. It is organized by the VIPS
(Vision, Image Processing, and Sound) lab of the Computer Science
Department of the University of Verona. The course is residential,
spanning 3 days, so that attendees can potentially achieve a more
productive interaction with the lecturers.

Partially Supervised Learning
Prof. Marco Loog
Delft University of Technology, University of Copenhagen

Standard supervised machine learning and pattern recognition methods
often do not fit the real-world requirements exactly and, hence, cannot
be applied directly to solve the decision, prediction, or classification
task at hand. A large class of approaches, which we collectively
referred to here as partially supervised, meet some of the demands
encountered in practice. The course will cover topics within the field
of classification and focuses on multiple instance learning,
semi-supervised learning, and, potentially, transfer learning, domain
adaptation, and active learning.

The lectures will not only focus on the main lines of research within
these particular areas. They also aim to provide some insight, some
theoretical background, and a critical assessment of the main concepts
and ideas underlying the methods. Possible directions for further
research may also be sketched.

The course is completed with a one-day computer exercise session in
which the participants can get some hands-on experience with concepts
and methods from the lectures.


DAY 1 (Theory):
– General introduction to classification, representation,
generalization, and evaluation. The aim of research in pattern
recognition and machine learning.
– Semi-supervised learning. Label propagation, harmonic functions,
expectation maximization, self-learning, transductive SVMs, and learning
under parameter constraints.

DAY 2 (Theory):
– Multiple instance learning. Classical approaches, MIL through
combining, MILES, and kernel and dissimilarity-based techniques.
– Basics of active learning (e.g. uncertainty sampling,
query-by-committee, information vs representation) and transfer
learning/domain adaptation (e.g. learning under covariate shift).

DAY 3 (Lab):
The idea of the lab session is to study two or more methods, related to
the contents of the course, and perform a critically comparison of these
methods, taking, for instance, into account their general applicability,
how efficient they are, how they perform on various data sets, etc.
Convincing arguments and, where necessary, experimental results
supporting the pros and cons put forward should be provided. (Depending
on the choice made, there might of course be the need to implement the
methods yourself.)
If the participant needs an evaluation, the results of the lab session
should be summarized in a to-the-point report of about four pages, which
will be reviewed and scored by the lecturer ? this counting as a final
exam. The report, of about four pages, should clearly describe the
research question, the research setup, the results from the analysis of
the underlying theory, methodology, and/or concepts, and the
experimental results. A conclusion and discussion should wrap up the

150 euro for PhD and undergraduate students.
200 euro for postdocs, researchers, and other people working at a
250 euro for everybody else.

If you want to participate, please send an email to or The mail should
state your full name and your status (undergraduate, PhD student,
other). The ultimate deadline for applications is April 1, 2015. You
will receive a confirmation email from us with a registration form to be
printed, filled out, and returned, together with a proof of payment
(detailed instructions will be in the confirmation email).

The accommodation costs are not covered by the Course Fee. A list of
hotels will be posted on the school’s website.

Directors: Manuele Bicego, Umberto Castellani (University of Verona, Italy)
Steering Director: Vittorio Murino (IIT Istituto Italiano di Tecnologia,

For any other information, please send an email to or;
For latest news check the school’s web page:

Source: SIG-IR List

NLP News