0

Expirience from Lisbon Machine Learning Summer School

I have participated on Lisbon Machine Learning Summer School (LxMLS), which took place on July 16-23  at Instituto Superior Técnico, a leading Engineering and Science school in Portugal. It is organized jointly by IST, the Instituto de Telecomunicações and the Spoken Language Systems Lab – L2F of INESC-ID. It was quite a great experience, on the one side to see Lisbon, while on the other learn a bit from the best people in Machine Learning and Natural Language Processing and meet fellow PhD students who work in the same or similar area as I do around the Europe. I will just briefly tell the experience.

I arrived to Lisbon day before, on 15th July. That evening we already could register, so I went there to check the place and register. I met there couple of people and we went to the city center with some of the locals and lab administrators for a dinner. After a dinner we ended up in a district full of bars for couple of beers.

Tomorrow morning the summer school started. In a first day MARIO FIGUEIREDO did a probablity refresher lecture. It was quite a good lecture to refresh a knowledge from probability and most of the things were already familiar, just the lecture were quite intense. It was 137 slides with formulas almost on each slide and all that in less then 1,5 hours. Quite intense start. I realized that I forgot lunch tickets in my hotel during the coffie break and asked some of the organizers what to do. They told me that someone can borrow me. Second lecture of the day was Introduction to python by LUIS PEDRO COELHO. This was quite basic lecture, with “Hello World” and similar things you do when you start learning programming language. I decided that, since I worked in python, I will go and pick up my lunch tickets. After the lunch we have every day labs, where we were supposed to implement some of the algorithms from the morning lectures. On the first day, there were no evening talk.

IMG_20150722_231043

Second day we had a talk about linear classifiers. The talk was by SHAY COHEN from the University of Edinbourgh. He was explaining how linear classifiers work and mathematically proving it. However, he was quite unenthusiastic and dull in his presentation and it was a lot of maths with some technical problems with a computer, so I cannot say that I enjoyed it. However, from the third day presentations got better and better. On the third day NOAH SMITH from the University of Washington presented sequential models and he was very very good. Enthusiastic, clear in explenation, just great lecture. Next day XAVIER CARRERAS from Xerox presented about structured predictors and this lecture was also very good and clear. SLAV PETROV from Google did a lecture on Parsing and told us some details of research done in Google on parsing and current state of the art. Also, since he is bulgarian, he was able to relate to some highly flectual slavic languages, where some things that work for English, just don’t work. CHRIS DYER day after did a presentation about learning from big data. I believe he was quite good, but I did something already with map-reduce, and day before I stayed out drinking until about 3am with a group of German and UK PhD students from the Summer School, so lecture at 9am, was quite hard to follow. On the last day Yoshua Bengio, one of the initiators of deep learning did a lecture on deep learning. It was a great, however a bit less technical and more popular lecture. I enjoyed it.

Evening lectures were also good. I think I enjoyed most the lecture by FERNANDO PEREIRA, a research director at Google, who was talking about efforts google is doing on information extraction and linking data semantically. However, there was also good lectures on statistical translation and lecture from PHIL BLUNSOM from Deep Mind about machine understanding and predicting next word using deep learning.

We also had a demo day, where mainly Portugese companies and research institutions were presenting demos of their research and products. There was even one venture capital fund, I managed to pitch an idea and schedule a meeting for the next day.

IMG_20150719_135625

Last but not least – Lisbon. It is a very nice city, as everybody was saying. Still my top is Florence, but Lisbon is very nice. The unfortunate thing is that I have not managed to explore it, since the lectures were usually starting at 9am and finishing about 6:30pm. But what I have seen, I liked.

Lisbon1

Born in Bratislava, Slovakia, but he lived in Belgrade, Serbia. Now he is doing a PhD in natural language processing at the University of Manchester. Great enthusiast of AI, natural language processing, machine learning, web application security (founder of OWASP local chapter in Serbia, currently one of the leaders of OWASP Manchester chapter and OWASP Seraphimdroid project), open source, mobile and web technologies. Looking forward to create future. Nikola wants to teach machines to feel and understand. Always finding way for dreams to come true.

Twitter LinkedIn Google+ YouTube Xing  

email

Leave a Reply

Your email address will not be published. Required fields are marked *