This paper focus on multiple CNN-based (Convolutional Neural Network) models for COVID-19 forecast developed by our research team during the first French lockdown. In an effort to understand and predict both the epidemic evolution and the impacts of this disease, we conceived models for multiple indicators: daily or cumulative confirmed cases, hospitalizations, hospitalizations with artificial ventilation, recoveries, and deaths. In spite of the limited data available when the lockdown was declared, we achieved good short-term performances at the national level with a classical CNN for hospitalizations, leading to its integration into a hospitalizations surveillance tool after the lockdown ended. Also, A Temporal Convolutional Network with quantile regression successfully predicted multiple COVID-19 indicators at the national level by using data available at different scales (worldwide, national, regional). The accuracy of the regional predictions was improved by using a hierarchical pre-training scheme, and an efficient parallel implementation allows for quick training of multiple regional models. The resulting set of models represent a powerful tool for short-term COVID-19 forecasting at different geographical scales, complementing the toolboxes used by health organizations in France.

This paper describes a method of predicting wheat yields based on machine learning, which accurately determines the value of wheat yield losses in France. Obtaining reliable value from yield losses is difficult because we are tackling a highly unbalanced classification problem. As part of this study, we propose applying the Synthetic Minor Oversampling technique (SMOTE) as a pretreatment step before applying machine learning methods. The approach proposed here improves the accuracy of learning and allows better results on the set of tests by measuring the operating characteristic of the ROC receiver. The comparative study shows that the best result obtained is 90.07% on the set of tests, obtained by hybridizing the SMOTE algorithm with the Random Forest algorithm. The results obtained in this study for wheat yield can be extended to many other crops such as maize, barley, ...

During the last decades, researchers have developed novel computing methods to help viticulturists solve their problems, primarily those linked to yield estimation of their crops. This article aims to summarize the existing research associated with computer vision and viticulture. It focuses on approaches that use RGB images directly obtained from parcels, ranging from classic image analysis methods to Machine Learning, including novel Deep Learning techniques. We intend to produce a complete analysis accessible to everyone, including non-specialized readers, to discuss the recent progress of artificial intelligence (AI) in viticulture. To this purpose, we present work focusing on detecting grapevine flowers, grapes, and berries in the first sections of this article. In the last sections, we present different methods for yield estimation and the problems that arise with this task.

Over the last few years, machine learning and data mining methods (MLDM) are constantly evolving, in order to accelerate the process of knowledge discovery from data (KDD). Today's challenge is to select only the most relevant knowledge from those extracted. The present paper is directed to these purposes, by developing a new concept of knowledge mining for meta-knowledge extraction, and extending the most popular machine learning methods to extract meta-models. This new concept of knowledge classification is integrated on the cognitive agent architecture, so as to speed-up its inference process. With this new architecture, the agent will be able to select only the actionable rule class, instead of trying to infer its whole rule base exhaustively.