We are the

TU Graz Data Team

A Group of Students interested in Data Science and AI

About Us

Who are we?

The TU Graz Data Team is a team formed by students and former students which participate in machine learning competitions such as Kaggle, but also in research competition. Our main goal is learning and prepare students for their careers in this field. We meet almost weekly where we learn, present, code and have fun.

Benefits

What is it in for you?

Access to GPU and CPU Clusters

When working with big data, you need big computers. Here you have the tools you need to succeed in competitions.

Workshops and weekly Meetings

In the team we organize weekly meetings, where we discuss progress and issues on current competitions. Furthermore we also organize in-house workshops for members.

Possibility for recieving ECTS

You can get ECTS via Data Team, by participating in KDDM2, or NLP. Furthermore there are possibility to receive ECTS via a Bachelor/Master Thesis.

Real Life Experience

All the knowledge gained via the competitions and training will prepare you for a job in the field.

Unlimited Makava and Beer

It is easier to work on competitions when you have a cold drink by your side. You can cool yourself and the GPU.

Data Team Room

We have our own dedicated Data Team room in Plüddemanngasse 106, 1st floor!

Past Challenges

Challenges we joined so far.


Elo Merchant Category Recommendation

Regression

The goal of the competition was to predict customer loyalty of a customer. The data provided was transactional data between merchants and customer including some metadata. We have created around 50-100 features based on the data and created several models such as lightGBM, Random Forest and Ridge Regression. Our final solution was an ensemble of these models. We have ranked 2026/4221 Worldwide.

iMaterialist Furniture Challenge

Image Recognition - Multi-Class Classification

The goal of the competition was to predict household object (Couch, cup, chair etc) based on the image. The dataset contained ~200k Images with 128 classes. Our solution was based on extracted feature vectors from all the images via pretrained CNN models on Imagenet. From these features we built several classifiers which were then ensemble through weighted average. We ranked 153/428.

Contact

Ready to join the team?

Join today!