I hold a B.Sc. in Informatics and a M.Sc. in Information Systems from the School of Informatics of the Aristotle University of Thessaloniki (AUTH). Since April 2010 I am a PhD Student in the area of Machine Learning and Data Mining at the same school and member of the Machine Learning and Knowledge Discovery (MLKD) group. Since January 2012 I also work as a research associate at the Multimedia Group of the Information Technologies Institute (CERTH-ITI) where I develop algorithms multimedia indexing and search.
- multi-target prediction (classification/regression)
- image-text annotation/retrieval
- learning from data streams
- recommender systems
A full list of my publications along with their accompanying documents and media can be found here.
I have participated in several Data Mining contests with some very good results:
I participated for a second consequtive year in the Retrieving Diverse Social Images task of MediaEval 2014, representing the SocialSensor team. This year we built upon previous year's "visual" method and used it to generate all types of runs. As a result, we developed the best-performing visual-only run, the 3rd text-only run and the best (by a large margin) visual+textual run. Our visual+textual run was also ranked 2nd overall, marginally beaten by a run that used external data. Here is a working notes paper and a poster describing our approach. Also check out this article in SocialSensor's site!
I got a top 10 finish (7th) in a Kaggle competition (WISE 2014) and was awarded the kaggle master badge!!! Here are the slides of a talk that I gave at WISE 2014 conference, describing my approach. And also the talk itself (synced with slides) is available here!
I was responsible for the best-performing visual-only run (and 3rd overall!) in the "Retrieving Diverse Social Images" task. Our team also received an idea-originality award. Here is a working notes paper and a poster describing our approach.
The participation in Data Mining Cup 2013 could have been a great success (1st place and 2000$ prize) if there was not this tiny but catastrophic bug in the implementation. Here is a short report with more details about the worst(=best) performing approach and here is a funny graphic.
In JRS 2012 data mining competition on topical classification of biomedical research documents we crossed the line 9th out 396 teams. Given the relatively short time that we invested and the tough competition this is considered a very good placement. Here is a short description of our approach.
We submitted the best-performing run in the "Concept-based Retrieval" subtask of the "Photo Annotation" task. Here is the paper and the poster describing our approach and finally a photo from the ImageCLEF lab.
We took the 2nd place in the "Cold Start Recommendations" track of the ECML/PKDD 2011 Discovery Challenge, receiving a 700 euro prize and an invitation for a paper describing our approach in the workshop proceedings. Here is the award and some photos from the workshop.
We took the 7th place in the Data Mining Cup 2011! Here are the slides of a short presentation describing our approach and a photo taken at the award ceremony.
I took the 1st place in the "Music Instruments" track of the ISMIS 2011 contest on Music Information Retrieval. The 1st place award included a prize of 1000 USD and an invitation for a paper describing the approach in the conference proceedings. I also wrote a non-technical blog post about my experience in this contest. There is also a footage of my presentation.
- 15-20/9/2014: I attended a great conference (ECML/PKDD 2014) and had the chance to meet great people working on Machine Learning. Also gave a talk on "Drawing Parallels between Multi-label Classification and Multi-target Regression" at the Multi-target Prediction (MTP) workshop of ECML/PKDD 2014!
- 15/7/2014: I got a top 10 finish (7th) in a Kaggle competition (WISE 2014) and was awarded the kaggle master badge!!!
- 1/7/2014: My first IEEE Transactions article was published in IEEE Explore!
- 9/6/2014: Our paper "Multi-Target Regression via Random Linear Target Combinations" (see Publications page) got accepted in ECML PKDD 2014!