Login for PhD students/staff at UCPH      Login for others
International Summer School on Semi-Supervised Learning in Image Analysis
Provider: Faculty of Science

Activity no.: 5165-16-02-31
Enrollment deadline: 01/01/2016
PlaceDepartment of Computer Science
Universitetsparken 1, 2100 København Ø
Date and time01.01.2016, at: 06:00 - 02.01.2016, at: 16:00
Regular seats60
ECTS credits3.00
Contact personSune Darkner    E-mail address: darkner@di.ku.dk
Enrolment Handling/Course OrganiserSune Darkner    E-mail address: darkner@di.ku.dk
Written languageEnglish
Teaching languageEnglish
Semester/BlockSummer
Block noteDuration: 5 days. Scheduled dates: August 2016 (dates yet to be fixed)
Exam formCourse participation
Course workload
Course workload categoryHours
Preparation20.00
Course40.00
Evaluation/reporting15.00

Sum75.00


Content
This will be the eigth in a series of summer school organized together with the Image Analysis and Computer Graphics section at DTU. The concept is to invite 2-4 internationally renowned experts to teach a week-long course together with local teachers from KU and DTU. The participants are a good mix of students from KU and DTU and international participants. The summer schools are held in remote locations to encourage interaction between students and teachers. In addition to bringing international expertise in to the groups, the summer schools also provide an important networking opportunity for the students.

• Subject area: Semi-Supervised Learning in Image analysis

One of the main challenges in medical image analysis, computer vision, and machine learning is to discover relations between low level data representations (pixels and features) and semantic labels. This is handled by the growing field of supervised machine learning applied to image analysis. The central idea is, in a principled way, to choose among hypothesis in a hypothesis space, that best fit the data and the semantic labels at hand. The generation of the hypothesis space, and the algorithms to choose the ”best” fit, is the discipline of machine learning. This is being used increasingly in image analysis, computer vision, and its applications to among others medicine.

However, often, it is very ”expensive” to obtain the semantic labels on the data, whereas the data themselves are ”cheaper” to obtain. This may be in the situation of video streams, where annotations of actions must be performed by hand, in medical imaging, where manual outlining of organs of interest are performed by trained radiologists. It may take one radiologist up to a full work day to manually outline the bone and cartilage in a knee MRI, as an example. Hence, often we are in the situation where much more data is available than what can be obtained semantic labels for.

The idea of semi-supervised learning is to use also the unlabelled data to improve the hypotheses that can be obtained from the very limited amount of labelled data. In the example below, the clustering of the unlabelled data clearly indicate how the red and green labels may be propagated using the clustering of the data.

• Scientific content:

The summer school will consist of 5 days of lectures and exercises. The students will be expected to read a predefined set of scientific articles on machine learning methods prior to the course. Additionally, the students should bring a poster presenting their research field (preferably with an angle towards machine learning in image analysis).


The course will consist of the following parts:

• A crash course on supervised learning and unsupervised learning.
• A theoretical insight in the challenges of semi-supervised learning.
• A practical session with hands-on exercises.
• Applications of semi-supervised learning in image analysis.

Learning outcome
After participating in the summer school, the student should:

• Understand semi-supervised learning and be able to differentiate between the different types of models.
• Have a strong knowledge about the theoretical foundations of semi-supervised learning and its relations to active learning and domain adaptations techniques.
• Be able to implement basic machine learning from scratch and train them using appropriate initialization and optimization techniques.
• Be able to apply semi-supervised learning for his/her own research projects.

Remarks
Department and research area: Department of Computer Science, Image Analysis and Machine Learning (In collaboration with DTU)

Course organisers:

Sune Darkner (darkner@diku.dk)
Aasa Feragen (aasa@diku.dk)

Guest lecturer 1: Serge Belongie, Cornell University.
Guest lecturer 2: Jimbo Shi, University of Pensylvania.

Guest lecturer 3: Marco Loog, Technical University of Delft
Guest lecturer 4: Anders Dahl, DTU Compute.

Teacher 1 from UCPH-SCIENCE (names and e-mail address): Christian Igel, igel@diku.dk
Teacher 2 from UCPH-SCIENCE (names and e-mail address): Mads Nielsen,

Search
Click the search button to search Courses.


Course calendar
See which courses you can attend and when
JanFebMarApr
MayJunJulAug
SepOctNovDec



Publication of new courses
All planned PhD courses at the PhD School are visible in the course catalogue. Courses are published regularly.