Jet Infosystems, Rosreestr, NLMK and Utkonos launch AI hackathon
Friends, we launched a competition among AI / ML-developers - RAIF Hackathon . We invite you to participate! The event is held within the framework of the business business forum RAIF 2018 (The Russian Artificial Intelligence Forum). A year ago, we did a similar hackathon , but this time the format will be different. You are waiting for 2 stages: qualifying online and final offline battle in front of the guests of RAIF. Three tasks - from Rosreestr, Utkonos and the Novolipetsk Metallurgical Combine (NLMK) - and, accordingly, three nominations to choose from. The prize fund is more than 1 million rubles. And yes, like last year, it will be necessary to solve the tasks with the help of machine learning technologies;).
You can participate alone or in teams of up to 3 people. You can choose one, two or even all three presented tasks at your discretion. In all three cases, you will have access to the real data of our partners (of course, impersonal). That is, you will have a great opportunity to look behind the scenes of their business, which, as it seems to us, is much more interesting than abstract imaginary tasks that are offered on most hackathons.
One task is for an absolute result:
- Accelerate rolled steel production (NLMK)
Participants of this stream will be able to see their results in the leaderboard on the competition website.
Two other tasks are creative:
- Analyze the demand for goods ("Utkonos").
- Predict the cadastral value of real estate (Rosreestr)
Here we will only provide data from partners, and you yourself will have to figure out what can be done with this data.
But first things first.
Online and offline
RAIF Hackathon will be held in two stages: online and offline. Till October 19 inclusive there passes the qualifying online tour of the task from NLMK. For tasks from Utkonos and Rosreestr, this stage is 1 day less - until October 18. After registering and filling out the profile in your account, you can download data. Upon completion of the work, it will be necessary to upload the created mathematical models there, in the personal account.
October 11th is the date of the “reconciliation of hours”: on this day you can send intermediate (or already final) solutions and get feedback from the hackathon curators. Timely expert advice will increase your chances of reaching the final. This option is especially relevant for those who participate in the “Utkos” and Rosreestr nominations, each of which involves several solutions.
On October 20, the selection stage will be summarized, and on October 21, the results will come to the participants in the mailing list and will be published on the RAIF Hackathon website .
In each nomination, the top 10 teams that reached the final will meet on October 23 in Moscow as part of the RAIF business forum at the final competition. Finalists are waiting for the additional data provided and 4 hours of coding to finalize their project. All this is in front of recognized experts in the field of AI / ML and top management of large Russian companies. In conclusion, the results will be summarized and the winners will be awarded.
In the nomination from NLMK winners will be identified by the absolute result. In nominations from Utkonos and Rosreestr, the best decisions will be determined by the jury on the basis of the protection of the submitted works. The prize fund will be divided by 3 teams - each will receive 350 thousand rubles.
Tasks, they are also nominations
AI for NLMK
The task is to predict the time of passage of the steel strip in the area of the hot rolling mill.
NLMK hot rolling mill produces rolled products up to 1,850 mm wide and from 1.45 mm to 16 mm thick. Brand grade - from low carbon to high strength, including carbon grades, as well as electrical steel. The main consumers of hot rolled coils and sheet products are construction industry enterprises, shipbuilding, automotive, pipe manufacturers, as well as their own cold-rolled steel production.
Hot rolled as follows. Heated slabs — steel plates serving as a rolling stock — are unloaded from the methodical furnaces to the mill line. In the process of rolling along the mill line, the steel strip is pressed in the stands of the rough and finishing group, becoming more and more thin and long, and at the end is wound into rolls on special coilers. The thinner and longer the strip becomes, the faster it should move along the mill.
Anonymous data is used as input data of bands (width, thickness, etc.) and impersonal data on the operation of the mill before rolling the next band (speeds of roll tables, power of stands, etc. without reference to the scheme). their physical meaning is indicated.
Andrzej Arshavsky, Data Analysis Director, NLMK:
In the hackathon format, we want to try to solve the problem of predicting the time of rolled steel on one of our key units. Hackathon gives you the opportunity to look at the usual production process from different angles, to observe how different, sometimes unexpected approaches are used to optimize it. And for the participants of the RAIF Hackathon is a chance to prove themselves, to try to solve another practical task and understand their level among their colleagues.
AI for Rosreestra
Determine the parameters that affect the value of real estate, and build a mathematical model that evaluates the market value of these objects.
New technologies, in particular machine learning, can significantly improve the efficiency of real estate valuation. Analytical conclusions can be gradually replaced by conclusions generated by machine learning algorithms based on an analysis of the situation on the real estate market and the degree of influence of various factors on the value of real estate objects.
In this competition, participants are invited to build a predictive model based on the provided download and any other data from open sources, which will determine the market value of the object. In this case, the data that will be the basis for determining the market value, and the sample itself for building the model, participants must independently find in open sources. As an estimated result of the hackathon, we will consider a bunch of the proposed mathematical model and presentation.
The presentation should reflect:
- external data that was used in building the model
- methods for assessing the correctness of the model and their results
- description of the model itself
- A description of the most important parameters and conclusions that can be made on this basis.
Evaluation options creative solutions
- Practical applicability
- Good : analytical work. When building the model, various external factors affecting the value of real estate objects were taken into account. The model can predict the value of real estate objects given the lack of information on a number of external factors.
- Bad : the conclusions that all factors affect the same, or the model works only for a small part of the objects
- Method for assessing the accuracy of the solution
- Good : finding the correct test sample, the ability to demonstrate the operation of the model
- Bad : calculated the cadastral value of the known formula
- Using external data
- Good : you analyzed and assessed the influence of various external factors (proximity to key infrastructure, transport accessibility, condition of the house, parks / forest parks, water reservoirs, lack of landfills, etc.)
- Bad : did not add any parameters or used them incorrectly (allowed the target variable to leak)
- Good : conclusions and decisions are different from those commonly known and accessible.
- Bad : applied the standard formula to calculate
Timofey Alekseev, Deputy Head of the IT Department of the Federal Registration Service:
We would be interested to evaluate the practical benefits of the presented solutions and the possibility of their further application in the work of the service. We expect non-standard solutions and attention to details from the participants.
AI for "Platypus"
Analyze the demand for goods online hypermarket, using historical data on the redemption of goods from warehouses over the past few years.
The decision will help the company to provide the necessary amount of goods in warehouses, given the changing demand.
In the framework of this task are interesting:
- Algorithms and solutions that could take into account the effect of price changes and the availability of some products on the demand for other products (Halo effect, “cannibalization”).
- Definition of goods that are substitute goods and related products.
- Identification of patterns in customer behavior, forecasting orders for goods, taking into account these patterns.
Vladimir Alabin, Forecasting Automation Manager, Utkonos:
We want to have a more complete picture of the demand and the factors affecting it in order, on the one hand, to maximize customer satisfaction, and on the other, to optimize warehouse operations.
Evaluation options creative solutions
- Understanding the subject area
- Good : the solution is based on an understanding of business needs.
- Bad : according to the participant, all the parameters are equally useful, predicted more or less - there is no difference.
- Cost effectiveness
- Good : you calculated indicators that may be of interest to the retail business (for example, the profit from the implementation of the system).
- Bad : Calculated abstract AUC or accuracy. What is the use of the store - it is not clear.
- Using external data
- Good : you appreciate the influence of holidays, weather and other external factors.
- Bad : added parameters, far-fetched (like weather effects on Mars).
- Good : you brought something of your own and showed how it differs from ready-made solutions.
- Bad : they opened Stackoverflow, found a similar question and the answer to it, did it by analogy.
>> Become a member of RAIF Hackathon <<
Attention! Playing 10 tickets to the technical section
At the end of the RAIF Hackathon, a technical section will be held where you can listen to recognized experts from Data Science - representatives of famous Russian and foreign companies (including start-ups). Among them: Konstantin Vorontsov , Professor of the Department of Intelligent Systems, FUPM MIPT; Dmitry Bugaychenko , Odnoklassniki Software Engineer; Emily Dral , Chief Data Scientist Mechanica.AI; Nikolai Knyazev , Jet Science Infosystems team leader; Alexey Dral CEO BigData Team, and others.
All hackathon finalists will be able to visit the technical section for free. For those who are not yet confident in their abilities or do not plan to participate in the hackathon, but really want to attend this event, we announce the drawing of 10 tickets! Until October 9, inclusive, make a repost on Facebook and / or Vkontakte and send a link here - in a personal message. Winners will be determined on October 10 by a random number generator. Each report through the LAN.
UPD: friends, for operative interaction with the participants of RAIF Hackathon, we created a telegram-chat t.me/RAIFHACK - questions on the hackathon can be asked there.