Go to goto
Hi, Habr! My name is Grigory Kuzovnikov. I work for Senior Backend Engineer at FunCorp. I recently started learning machine learning. There is no one directly to ask something, you have to search for everything on the Internet. Therefore, I wanted to go to any specialized conference and hear how to use machine learning in combat at all. I didn’t find anything relevant at once, however, GOTO Berlin had a lot of machine learning topics, so I decided to visit it. Under the cat a small review of this conference and a lot of photos
About the conference
GOTO is held not the first year and not only in Berlin. The next one, for example, will take place in Copenhagen this November. There is no narrow direction for the conference: here you can listen to reports about the server and mobile development, as well as strange, but now popular speeches about self-motivation.
The conference was held in the very center of Berlin, on Alexanderplatz, in a modern two-story building. In total there are three small audiences and one large hall. Before the entrance were set beautiful branded flags.
Registration of participants took place without queues and very quickly. Immediately presented a backpack with a logo and offered to stick on the badge technology icons, which I would like to talk with other listeners or speakers. In the halls on the first and second floors were the stands of the sponsors of the conference: Amazon Web Services, eBay Tech, and others. Standard promo kits were offered here: pens, stickers, socks, and T-shirts. At many stands it was possible to take a small curtain for a laptop camera. In addition, you could register and win some prize. Headphones, Lego kits and a Google Home speaker were played. True, the prize draw was held on Friday, and I stayed there until Thursday, so I did not know if I won or not.
GOTO has its own application in Google Play and the AppStore with a schedule of all performances. Through it, you can also ask questions to the speakers (they are read out after the speech), and you can also leave feedback about the report. There are practically no questions from the audience.
There is always food on the court: some small snacks, ice cream, all kinds of sweets, drinks. At any time, you can take and eat. For lunch, full meals are served, such as rice with meat and salad.
In Russia, except on Highload, I have not been anywhere. In my opinion, Highload, of course, is larger, and the program is higher. The guys from "Ontiko" - Respect!
I went to the conference for practical knowledge, which, unfortunately, I could not actually get (the most useful and interesting that I managed to learn, I will describe in the following paragraph ). Most of the reports on machine learning turned out to be quite simple (basic level and a little higher), the students are also not experts. When a speaker on a Java chat report asked a speaker to raise the hands of those who use machine learning in production, only me and another person raised their hands.
However, there was one very complex report with a large number of graphs and diagrams from a professor of Computer Science at the Humboldt University of Berlin. He talked about the autogeneration of tests using machine learning, as well as about the inverse problem - the autogeneration of code that satisfies the tests.
An interesting report was Tsshidriha Olaf (Olaf Zschiedrich), Technical Director of OLX Group, which was called the From big to mess data to data as with the innovation enabler . He talked about how data is collected in OLX and how access is given to them for later use in ML.
Data from various sources is collected in one common repository, and then it is not just accessible, but depending on the current task, special repositories are collected with data obtained from various sources with varying degrees of detail. This is done to ensure safety, as well as compliance with all kinds of European laws.
The most useful report, in my opinion, is that of Christoph Windheuser.Artificial Intelligence Reloaded - AI Applications in the Industry . As I understand it, his company is doing a lot of ML application development as a contractor. In the report, he showed a few examples of their approach to the development and deployment of ML applications.
Interestingly, different people are involved in developing models and the application itself, so adapting the model for a real application is not a trivial task. It is also curious that for testing a ready-made ML application they feed him the same Validation Set data that is used in training. If the application has processed the required percentage of tasks correctly, it is considered that it passes the tests.
5 interesting facts about ML:
- For ML in Java, use Deeplearning4j, but it’s not as powerful as TensorFlow.
- Google has interesting tools: Cloud Dataflow, BigQuery, Cloud AutoML, which can be used to experiment with ML.
- There is LoRa technology - a long-range radio protocol for the Internet of things. There is also the provider The Things Network (thethingsnetwork.org), which integrates LoRa access points and allows you to use ready-made infrastructure for your devices.
- Data intended for use in ML should be collected and stored centrally. It is worthwhile to think in advance how this data will be extracted.
- To test the ML application, you can simply feed him the Validation Set.
It seemed to me that foreign conferences should be good in that you can personally communicate and exchange experiences with developers from large international companies who rarely come to Russia. But at the GOTO of some cool network did not work, including because the organizers themselves did not pay enough attention to this. Therefore, it was possible to communicate only on the stands and ask the speakers questions through the applications, and once even into the microphone. By the way, I was nervous because of this, since it is unusual for me to speak English in front of a full audience.
In general, the reports are not bad, if you want to learn in general about some technologies. If you need more details, you can also find something, but not as much as we would like.