Researchers from Yandex Research have created a new artificial intelligence system called TabM, designed for high-precision processing and analysis of tabular data. The development allows solving a wide range of tasks — from optimizing logistics and predicting energy consumption to medical analytics.
The system is built as an ensemble of neural networks that are trained simultaneously and share some parameters. This approach allows combining high forecast accuracy with economical use of computing resources.
In tests on 46 datasets, TabM showed better stability and accuracy compared to classical algorithms, including gradient boosting.
Yandex stated that the development of Russian scientists surpassed other solutions not only in terms of the average position occupied but also in terms of operational stability.
The algorithm has already been used in international competitions, including a competition on the Kaggle platform, where it helped assess the chances of survival of patients after bone marrow transplantation.
Read more on the topic:
Student from Chelyabinsk Created SheetsGPT, a Neural Network for Working with Excel Spreadsheets
Yandex Launches "Documents" with Built-in Neural Network
Miro Replacement: MTS Launches Beta Version of True Tabs No-Code Platform for Teamwork