Yesterday, I was on very interesting IT event – tech.3camp [http://3camp.pl/]. I was attending these meetings about 10 years ago, when they were organised in the old city in Gdansk. This group of IT passionates, every time presents new IT ideas, approaches, projects.
Yesterday’s meeting was very special about Machine Learning. I used to have lectures of AI and Machine Learning during my studies – but then technology was so poor, that implementing image recognition or speech recognition was extremly hard.
Nowadays technology is so well developed that image recognition is an easy peasy, thanks to very well implemented libraries. Speech recognition, even in Polish language is also easy-thanks to this Polish project:
or this university program:
Three presentations shown during this event:
- nGraph, czyli jak rozpoznawać zdjęcia przy pomocy żyrafy – Tomasz Dolbniak, Intel
It was very interesting presentation about graph used in neural networks. Intel implements its own library called nGraph, which is used to parallel calculations in neural networks. More info:
2. Etykiety to nie wszystko –Jan Chorowski, NavAlgo Ltd
This presentation was like scientific dissertation about teaching neural networks without using labels. Application of this approach was shown in tracking courier packages in one Polish logistic company.
3. Anonimizacja a uczenie maszynowe –Maciej Sołtysiak, Kainos
This presentation was super useful for me, as last November I had to implement data anonimization for one of my application in my company. It turned out, that my approach was too easy. The whole presentation covered topics as:
Why is anonymization useful? Hashing sensitive data. Case studies of successful data deanonymisation and how it could have been prevented. Balancing between customer-useful data and the privacy of system users. Operating on GPS tracks. Their specificity and methods of anonymization. Impact of GPS data representation on the anonymization task. Notes on the efficiency of the anonymization process.
Next time for sure I will improve my implementation;)