The VK Predict analytical product center, part of the VK group, has created a platform for automatically creating ML models and working with AI solutions without deep knowledge of information technology. According to Forbes, the AutoML platform will help build development strategies, help marketers and allow you to train AI models with partner companies without exchanging proprietary data – based on federated learning. In this case, the emphasis is placed on non-specialists.
The platform launched by VK, for example, allows business analysts to evaluate a company’s market position, build a business development strategy, and solve marketing problems from audience segmentation to identifying the most profitable clients. The platform is suitable for forecasting indicators, sorting data into groups, including generating a revenue forecast for a particular outlet, or sorting users according to various criteria. The platform can be used in real projects, additional training of the model based on new sources and information. Further opportunities will appear, including fine-tuning of AI models.
Image source: Possessed Photography / Unsplash
An important feature of the new platform is the ability for companies to jointly train common ML models with vertical federated learning. Companies will be able to use their own data sets for training, but the exchange is carried out only with encrypted intermediate data – this scheme is suitable for companies of various profiles, including businesses operating in the field of online commerce and fintech. For example, this is how you can calculate the effect of affiliate marketing with a common audience (manufacturer-retailer).
Automation is facilitated by support for the Low-Code principle. The target audience for the platform is any project that needs to automate work with data sets, and companies can employ specialists without knowledge in the field of Data Science. However, such a platform is also well suited for IT experts who want to save time.
Image source: Wes Hicks / Unsplash
VK claims that the use of AutoML will make it possible to implement ML models many times faster than before. It is planned to monetize it both by subscription in the Self-Service format and to sell it as a core for the IT departments of large companies. According to some reports, in the first case we can talk about an amount of 250 thousand rubles/month. A customized solution in the on-premise version can cost from 10 million rubles per project, excluding support and maintenance fees.
Experts agree that such platforms are unlikely to completely replace data science specialists. At the same time, they will be able to provide them with considerable assistance and speed up the introduction of ML models to the market. Experts interviewed by Forbes noted that AutoML platforms have been actively used in large companies for a long time. Domestic tools of this class are offered in Yandex DataSphere, Sber has the LightAutoML framework. Open solutions AutoGluon or FEDOT are also used.