Alliance in AI: System for evaluating machine learning models in the agricultural sector

OS: Artificial Intelligence in Agriculture (APK)

2023: Introduction of Industry Machine Learning Model Benchmarking Metrics System

On November 24, 2023, the Alliance in Artificial Intelligence introduced a benchmarking metric system for industry machine learning models in the agricultural sector. The launch^[1] Cystemf online^[2] been announced^[3]which will allow businesses to evaluate their AI solutions. The resource was created within the framework of the Industry Club in the field of AI (agro-industrial complex), which operates on the basis of the Alliance.

{{quote 'author = said Artem Petrov, CEO of Rusagro Tech, Director of ITRusagro. | The agro-industrial complex is a strategically important industry for the state. And its digitalization is gaining impressive momentum, new solutions using AI are emerging. Rusagro officially entered the Alliance in the field of artificial intelligence in October 2023, and in a short time, together with the members of the Industry Club, it was possible to achieve results in creating a methodology for assessing the effectiveness of ML models, which formed the basis of the site. I am sure that the launch of the metric system will realize the good potential of agribusiness in the field of AI,}}

Thanks to the active collective work of professionals, we initiated the creation of industry benchmarks on the Club's platform. This will allow us to develop the best solutions in the industry, and developers to determine the level of accuracy of their models and make the necessary improvements, noted Tatyana Surdina, Marketing Director of OKHK Uralchem JSC.

The Alliance, together with partners, posted on the resource industry tasks based on data analysis, as well as test datacets and leaderboards - ratings of the best machine learning models created by site users to solve published problems.

As early as November 2023, the site contains information on metrics and benchmarks in the field of crop production, animal husbandry and the food industry, as well as tasks for predicting crops, classifying crops, monitoring the condition of animals and determining the quality of food raw materials.

The project is being implemented as part of the execution of the order of the President of Russia on the introduction by business entities of a system for evaluating domestic decisions that use artificial intelligence technologies. Using the example of the agro-industrial complex, a pilot methodology was developed for subsequent replication to other industries.

The project was implemented in several stages: as part of in-depth interviews with companies in the agro-industrial sector, the Alliance collected the most frequently used and reference metrics to create a basic version of the methodology. Then, during the collection of industry benchmarks, metrics were verified by the wider composition of participants and the scientific community. And, finally, according to the results of collecting benchmarks and analyzing the formed leaderboards, the experts made changes to the methodology aimed at increasing the effectiveness of its applied application.

On the site, created in technological partnership with, Cloud.ru the Alliance posted a pilot benchmark, in the launch of which the team helped Geoworld"": the task of detecting sunflower culture from pictures of s. The drones first benchmark has already been attended by the Sberbank Business Soft Forest Analytical Center and the agriculture Albatros company, as well as the university team. ITMO

The resource contains information about the project methodology and about existing sites with relevant benchmarks - both Russian and foreign.

Notes