Traineeship at University of Zielona Gora

As a doctorate student at Vilnius Gediminas technical university I have planned minimum of 2 months of internship/traineeship in abroad science institutions. Starting from the beginning second study year I began the search for traineeship place, which would be beneficial for my studies. I contacted with various universities and asked if there is some scientist working in my study direction. Finally, I got contacts of Assoc. Prof. Krzysztof Patan from Institute of Control & Computation engineering, Faculty of Computer, Electrical and Control Engineering at University of Zielona Gora in Poland. Professor kindly agreed to define traineeship program and mentor me during the traineeship. University of Zielona Gora is one of the biggest and best universities in Poland, there is a lot of study programs and faculties.

Poland is neighboring country of Lithuania. It is also European Union country. There are no airports, no direct train or bus routes to city of Zielona Gora, so I choose to ask someone to shuttle me by car and to take me back after traineeship is over. It is about 800+ km and more than 12 hours journey from Vilnius to Zielona Gora. On arrival, I got accommodated in university dormitory. University of Zielona Gora has 3 big dormitories near in university campus and usually provides Erasmus student with accommodation. Accommodation in dormitory room cost was 375 Zloty per month. The city of Zielona Gora is very green and beautiful, there are trees along each road, a lot of big parks, nice old city center and famous palm greenhouse restaurant called Palmernia, which is located on hill in the center of vineyard. The people are happy, there is a lot of young people. There is a lot of students in that city. I believe this city can be called university city.

My place of work was in one of renovated university campus buildings at 2nd floor in doctorate and exchange students room, also I got permission to use Institute of Control & Computation engineering computer laboratory at 4th floor for my research. There were workstations prepared for deep learning research. That workstations have fast central processing units and the most import fast graphical processing units. This were one of the main point for selecting Zielona Gora University for traineeship as I had no access to such workstations at my university. Also, the scientist work on deep learning in this institute. Also, there were local doctorate students and professors, who introduced software used in laboratory to me.

My traineeship title was Obstacle Avoidance, and the tasks of traineeship was made in a way, that I would get skills in preparing workstation for deep learning study, get knowledge about structure of deep neural networks and software frameworks, build an algorithm for self-driving car and make simulation My mentoring professor and I set the meetings schedule and discussed my work progress as I was working on tasks defined in my traineeship agreement. Each meeting was very useful, because professor has a lot of expertise in field of machine learning.

I first did investigation on how different parameters effects the deep neural network accuracy and speed. I used MNIST data set for training and testing of created models. After that, I compared how fast different software framework does training of same deep neural network model. At last, I had enough knowledge to train some algorithm on car driving dataset. My mentor recommended to use Udacity Self-Driving Car Challenge 2016 dataset. I build the network model and training algorithm based on research papers and examples and got interesting results. These results showed that model learned how to drive car, but had some errors also which I analyzed. After that I decided to modify Udacity self-driving car simulator used for self-driving car Nanodegree program in Udacity MOOC. I added various obstacle, such as different size and shape rocks on the road and tires. After that I drove a virtual car in simulator and got training dataset. I used that data to train the deep neural network to avoid obstacles on road and remain on road. After training, my car could drive on road and avoid most of the obstacle. I completed all the tasks defined in traineeship agreement, prepared report and discussed all results with my mentor. Lastly, we discussed about further research directions.

Traineeship was great experience for me and will be very useful in my scientist career. I got contacts, got a lot of knowledge and skills during my stay at University of Zielona Gora at Poland. I am thankful for ERASMUS+ and Partners4Value for this great opportunity.