Data Science Weekend Review
Hello! On March 3-4, Data Science Weekend took place, which was organized for the third time by our company with the support of GVA . For those who were not at the event, we prepared a brief overview of what was happening.
The event consisted of two thematic days. The first day was devoted to artificial intelligence and deep learning, the second to practical and business issues in big data.
On the first day, 5 speeches were scheduled, but there was a force majeure: the first speaker fell ill and was unable to speak with us. Nevertheless, we managed to prepare an interesting interactivity with the audience and warm them up a bit for the next speakers.
')
1. The interactive was in the spirit of “Truth is or fiction, that ...”. And then followed some wording about the achievement of artificial intelligence. For example, “Is it true that AI learned how to generate photorealistic images using only the user's request?”. The use case is: “Give me a picture of a volcano”. And then the neural network produces a few cool pictures.
There were 17 such questions. And the majority of our participants coped without problems. As it turned out, the audience came quite prepared.
2. After this interactive, the CEO of Intento, Konstantin Savenkov, spoke. The report was devoted to the problem of integrating various services in the current world. Each service has its own API, but it is rather difficult to turn from one to the other. To solve this problem, Intento has prepared its product, allowing different services to be integrated with each other. According to Konstantin, at the moment they are concentrating on API services in the field of artificial intelligence - machine translation, image recognition, text to voice, etc.
3. The next speaker was Anatoly Vostryakov - the head of (!) Dialogue systems and smart assistants in the Segmento company. Anatoly paid considerable attention to the problems that exist with chat bots in the area of ​​customer support. In the course of the dialogue, people tend to change their goal, ask questions nonlinearly, refer to the previous history of solving a particular issue, formulate the same request in hundreds of different ways, etc. The classical approaches to building chat bots are not always able to cope with these problems. At the end of the presentation, a different approach based on neural networks was proposed, the so-called end-to-end, when the developer does not prescribe strict rules, but allows the network to extract knowledge from the available data itself.
4. Next was the speech of Alexander Serbul from 1C-Bitrix. Alexander was true to himself and talked about his experience of learning neural networks in his bright and emotional manner. In 1C-Bitrix, there is a functional “Open Lines”, which allows you to communicate with clients in different social networks and channels from one of your own windows. At some point, it became clear that people often turn to the same questions, and the idea came up to make a bot that would suggest answers to them. This experience was the story of Alexander.
5. The last performance on this day was a teleconference from San Francisco with Nikolay Davydov, the investor of 2016 according to RBC. Prisma became the app of the year in upstor, and MSQRD was sold to Facebook. These are the projects that Nikolai had to do with. His presentation focused on business in this area. He said that what is happening now is truly a revolution, and in a few years artificial intelligence will be used in many industries. At the same time, Nicholas complained that at the moment most projects go into entertainment and consumer topics. In this regard, advice was given - to go and try to implement projects in industries. Nicholas cited several examples from the fields of agriculture, medicine, genetics, etc. He also noted that it is equally difficult to teach a person from the subject area to AI algorithms, and to a person who knows these algorithms, to teach the tricks of the subject area, therefore one way out is to create teams from experts in different areas and competently manage them.
On the second day of the event, 8 speeches were scheduled.
1. The first were the guys from the OSA Hybrid Platform project. The purpose of this project is to increase the presence of a product on the shelf. To do this, they predict such things as: 1. The expected increase in demand for the product (due to weather, for example); 2. The likelihood of the current absence of goods on the shelf (for example, in real time for checks); 3. In the future we plan to develop a product recognition system using the camera.
2. The second were colleagues from MegaFon , the official partner of the event. Andrei Uvarov, Head of Analytical Services, told how MegaFon achieves a tangible business effect using the technologies of big data analytics, machine learning and artificial intelligence. Continuing the theme of the first day of the conference - chatbot, colleagues introduced the participants to Elena, a virtual assistant. This voice command recognition technology, which even without connecting the client with the call center operator, itself transfers the caller to the desired menu item and gives tips on how to use any service of the operator. Alexander Bashmakov, Director of Infrastructure, spoke about the use of data analysis technology for “smart” planning for the development of the MegaFon network. The presentation ended with an overview of the automatic network monitoring and control system, which was first demonstrated to everyone in the MegaFon interactive zone.
3. The next was a data scientist from E-Contenta, Yuri Makarov. He spoke about how to classify texts on data from the search results, having tried many algorithms, including neural networks. The random forest algorithm won, using one tricky feature, being even faster than the network. The company then uses a trained classifier to create personalized content recommendations.
4. The last speech before the dinner was from Artem Pichugin, head of data-related educational programs, New Professions Lab. He told about why it is worth studying data science, how to do it correctly, and briefly about the programs that are currently being recruited. Answering the question “Why?”, Artem spoke about current trends and that the hype curve is not the only curve to pay attention to, but there is, for example, a curve for the spread of innovation in the market. To the question “How?”, He said that adults and children learn differently, therefore they should be taught differently too, after which he cited several examples of the use of andragogy (the science of adult education) in their programs.
5. After lunch, the first speaker was Yevgeny Gapon, director of analytics at Qlean. He spoke about how machine learning in their company looks like, going through all stages of the process: from data collection to model implementation. Eugene paid particular attention to the case of predicting the client’s refusal to clean. Sudden refusals of cleaning are detrimental to the business, so the task of early forecasting this event is urgent. The forecast allows you to prevent this event through a flexible bonus system: the higher the probability of canceling an order with a customer, the higher the bonus for saving the order will be shown.
6. The next speaker of the program was Artem Prosvetov, data scientist from CleverDATA. The topic of the speech sounded intriguing: “Text mining of Beauty Blogs: What are women talking about?” Artem demonstrated the process of identifying the most influential beauty bloggers for the purpose of promoting a product from the field of cosmetics. After identifying the most influential bloggers (by the way, the interesting fact was that the most popular bloggers write posts that are positively colored, that is, a blogger praises a particular product), an analysis was made about which products bloggers usually write about. The result of the analysis is a recommendation of the type: anti-wrinkle cream is better promoted through blogger A, and butter - through blogger B.
7. The next-to-last speaker of the program was Svetlana Krylova, head of Brand Analytics analytical center. She told about a similar story - their project was associated with an analysis of what people write on social networks about allergies. As a result of the analysis, it became clear that in advertising the image of a person suffering from allergies is different from reality. In advertising, this is a girl who cares about her appearance, but in reality, these are mothers of children suffering from allergies. Also from this analysis it became clear why Suprastin is the most popular allergy medication, although it is an antihistamine of early generations, and currently there are more advanced remedies with fewer side effects.
8. The program was completed by Andrey Karmatsky, CEO of Urbica. The report was devoted to the design of cities using data. There were a lot of beautiful visuals in the presentation, the viewer's eyes rejoiced. The most interesting case, which told Andrew, was a project to reorganize public transport routes in Moscow. The project team conducted an analysis of how the Moscow transport system currently exists, having built a simulation model. After that, optimization of routes was proposed, which was tested in practice and showed an increase in passenger traffic, as well as a decrease in waiting times for buses and trolleybuses.
We would like to thank everyone who spoke at our event, as well as all those who have been as spectators for these two days. You created a wonderful and lively atmosphere in the hall, it was very cool!
The next event, a full-fledged Data Science Week, will take place on September 7, 8, 11 and 12.
We will wait for you!
»All presentations are posted here .
»Access to video speeches can be obtained here .
The event consisted of two thematic days. The first day was devoted to artificial intelligence and deep learning, the second to practical and business issues in big data.
On the first day, 5 speeches were scheduled, but there was a force majeure: the first speaker fell ill and was unable to speak with us. Nevertheless, we managed to prepare an interesting interactivity with the audience and warm them up a bit for the next speakers.
')
1. The interactive was in the spirit of “Truth is or fiction, that ...”. And then followed some wording about the achievement of artificial intelligence. For example, “Is it true that AI learned how to generate photorealistic images using only the user's request?”. The use case is: “Give me a picture of a volcano”. And then the neural network produces a few cool pictures.
There were 17 such questions. And the majority of our participants coped without problems. As it turned out, the audience came quite prepared.
2. After this interactive, the CEO of Intento, Konstantin Savenkov, spoke. The report was devoted to the problem of integrating various services in the current world. Each service has its own API, but it is rather difficult to turn from one to the other. To solve this problem, Intento has prepared its product, allowing different services to be integrated with each other. According to Konstantin, at the moment they are concentrating on API services in the field of artificial intelligence - machine translation, image recognition, text to voice, etc.
3. The next speaker was Anatoly Vostryakov - the head of (!) Dialogue systems and smart assistants in the Segmento company. Anatoly paid considerable attention to the problems that exist with chat bots in the area of ​​customer support. In the course of the dialogue, people tend to change their goal, ask questions nonlinearly, refer to the previous history of solving a particular issue, formulate the same request in hundreds of different ways, etc. The classical approaches to building chat bots are not always able to cope with these problems. At the end of the presentation, a different approach based on neural networks was proposed, the so-called end-to-end, when the developer does not prescribe strict rules, but allows the network to extract knowledge from the available data itself.
4. Next was the speech of Alexander Serbul from 1C-Bitrix. Alexander was true to himself and talked about his experience of learning neural networks in his bright and emotional manner. In 1C-Bitrix, there is a functional “Open Lines”, which allows you to communicate with clients in different social networks and channels from one of your own windows. At some point, it became clear that people often turn to the same questions, and the idea came up to make a bot that would suggest answers to them. This experience was the story of Alexander.
5. The last performance on this day was a teleconference from San Francisco with Nikolay Davydov, the investor of 2016 according to RBC. Prisma became the app of the year in upstor, and MSQRD was sold to Facebook. These are the projects that Nikolai had to do with. His presentation focused on business in this area. He said that what is happening now is truly a revolution, and in a few years artificial intelligence will be used in many industries. At the same time, Nicholas complained that at the moment most projects go into entertainment and consumer topics. In this regard, advice was given - to go and try to implement projects in industries. Nicholas cited several examples from the fields of agriculture, medicine, genetics, etc. He also noted that it is equally difficult to teach a person from the subject area to AI algorithms, and to a person who knows these algorithms, to teach the tricks of the subject area, therefore one way out is to create teams from experts in different areas and competently manage them.
On the second day of the event, 8 speeches were scheduled.
1. The first were the guys from the OSA Hybrid Platform project. The purpose of this project is to increase the presence of a product on the shelf. To do this, they predict such things as: 1. The expected increase in demand for the product (due to weather, for example); 2. The likelihood of the current absence of goods on the shelf (for example, in real time for checks); 3. In the future we plan to develop a product recognition system using the camera.
2. The second were colleagues from MegaFon , the official partner of the event. Andrei Uvarov, Head of Analytical Services, told how MegaFon achieves a tangible business effect using the technologies of big data analytics, machine learning and artificial intelligence. Continuing the theme of the first day of the conference - chatbot, colleagues introduced the participants to Elena, a virtual assistant. This voice command recognition technology, which even without connecting the client with the call center operator, itself transfers the caller to the desired menu item and gives tips on how to use any service of the operator. Alexander Bashmakov, Director of Infrastructure, spoke about the use of data analysis technology for “smart” planning for the development of the MegaFon network. The presentation ended with an overview of the automatic network monitoring and control system, which was first demonstrated to everyone in the MegaFon interactive zone.
3. The next was a data scientist from E-Contenta, Yuri Makarov. He spoke about how to classify texts on data from the search results, having tried many algorithms, including neural networks. The random forest algorithm won, using one tricky feature, being even faster than the network. The company then uses a trained classifier to create personalized content recommendations.
4. The last speech before the dinner was from Artem Pichugin, head of data-related educational programs, New Professions Lab. He told about why it is worth studying data science, how to do it correctly, and briefly about the programs that are currently being recruited. Answering the question “Why?”, Artem spoke about current trends and that the hype curve is not the only curve to pay attention to, but there is, for example, a curve for the spread of innovation in the market. To the question “How?”, He said that adults and children learn differently, therefore they should be taught differently too, after which he cited several examples of the use of andragogy (the science of adult education) in their programs.
5. After lunch, the first speaker was Yevgeny Gapon, director of analytics at Qlean. He spoke about how machine learning in their company looks like, going through all stages of the process: from data collection to model implementation. Eugene paid particular attention to the case of predicting the client’s refusal to clean. Sudden refusals of cleaning are detrimental to the business, so the task of early forecasting this event is urgent. The forecast allows you to prevent this event through a flexible bonus system: the higher the probability of canceling an order with a customer, the higher the bonus for saving the order will be shown.
6. The next speaker of the program was Artem Prosvetov, data scientist from CleverDATA. The topic of the speech sounded intriguing: “Text mining of Beauty Blogs: What are women talking about?” Artem demonstrated the process of identifying the most influential beauty bloggers for the purpose of promoting a product from the field of cosmetics. After identifying the most influential bloggers (by the way, the interesting fact was that the most popular bloggers write posts that are positively colored, that is, a blogger praises a particular product), an analysis was made about which products bloggers usually write about. The result of the analysis is a recommendation of the type: anti-wrinkle cream is better promoted through blogger A, and butter - through blogger B.
7. The next-to-last speaker of the program was Svetlana Krylova, head of Brand Analytics analytical center. She told about a similar story - their project was associated with an analysis of what people write on social networks about allergies. As a result of the analysis, it became clear that in advertising the image of a person suffering from allergies is different from reality. In advertising, this is a girl who cares about her appearance, but in reality, these are mothers of children suffering from allergies. Also from this analysis it became clear why Suprastin is the most popular allergy medication, although it is an antihistamine of early generations, and currently there are more advanced remedies with fewer side effects.
8. The program was completed by Andrey Karmatsky, CEO of Urbica. The report was devoted to the design of cities using data. There were a lot of beautiful visuals in the presentation, the viewer's eyes rejoiced. The most interesting case, which told Andrew, was a project to reorganize public transport routes in Moscow. The project team conducted an analysis of how the Moscow transport system currently exists, having built a simulation model. After that, optimization of routes was proposed, which was tested in practice and showed an increase in passenger traffic, as well as a decrease in waiting times for buses and trolleybuses.
We would like to thank everyone who spoke at our event, as well as all those who have been as spectators for these two days. You created a wonderful and lively atmosphere in the hall, it was very cool!
The next event, a full-fledged Data Science Week, will take place on September 7, 8, 11 and 12.
We will wait for you!
»All presentations are posted here .
»Access to video speeches can be obtained here .
Source: https://habr.com/ru/post/324622/
All Articles