Post-doc position (18 months) at CIRAD (Montpellier, France) on Machine Learning for Food Security

98 views
Skip to first unread message

Roberto Interdonato

unread,
May 26, 2023, 10:26:16 AM5/26/23
to ml-...@googlegroups.com

Post-doc (18 mois) au CIRAD - UMR TETIS, Montpellier, France

 

 

Titre

Development of a machine learning framework for the prediction of food security indicators at country scale from heterogeneous data 

 

Keywords 

 

Science de données, Apprentissage Automatique, Sécurité alimentaire, Données Hétérogènes

 

Context

 

La sécurité alimentaire (SA) est un problème central dans de nombreuses régions du monde, comme en témoigne sa présence parmi les 17 objectifs de développement durable (ODD 2 - Faim zéro).

Pour surveiller les situations d'insécurité alimentaire, plusieurs systèmes d'alerte précoce sont actifs aujourd'hui, tels que GIEWS (FAO) et FEWSNET (USAID).

Ces systèmes utilisent un ensemble limité de types de données, i.e., données agroclimatiques provenant d'images satellites et indicateurs extraits d'enquêtes auprès des ménages sur les facteurs nutritionnels, économiques et liés à la production. De plus, une intervention humaine est souvent nécessaire pour combiner et résumer toutes les sources d'information.

 

Les recherches menées précédemment par TETIS ont montré que des données ouvertes hétérogènes, liées à différents niveaux à la sécurité alimentaire, peuvent être utilisées pour développer des méthodes d'apprentissage automatique capables de produire des indicateurs de SA. L'idée est d'utiliser des données qui représentent des proxys pour les raisons multiples et interdépendantes à l'origine de ce phénomène. Quelques exemples peuvent être des rasters d'informations spatiales, information géographique bénévole, données météorologiques, indicateurs économiques et données textuelles issus des médias locaux.

 

L'objectif de ce post-doc est de consolider et d'étendre les travaux récents sur ce sujet [1], en exploitant des stratégies d'apprentissage automatique et profond de dernière génération qui peuvent faire face à des aspects critiques, tels qu'une grande hétérogénéité des données et des données de vérité de terrain limitées. Plus précisément, Le travail est structuré au tour des trois questions de recherche :

 

- Quels types de données en accès libre peuvent être ciblés pour produire des indicateurs de Sécurité Alimentaire ?

- Comment des données hétérogènes en termes de thématique, de structure et de résolution spatio-temporelle peuvent-elles être prétraitées pour obtenir des prédictions cohérentes de la Sécurité Alimentaire pour un site d'étude donné ?

 

- Comment les techniques plus récentes d'apprentissage automatique et profond peuvent-elles être exploitées et combinées pour traiter ces données hétérogènes ?

 

 

En plus de la publication d'articles de recherche originaux dans des revues scientifiques de premier plan, l'objectif de ce travail sera également de développer une suite de méthodes (p.ex., une bibliothèque Python) capables de récupérer et d'exploiter des données hétérogènes pour produire efficacement des indicateurs de SA à l'échelle nationale.

L'utilisation de données de référence fournies par le CIRAD et le CGIAR permettra de tester les méthodologies proposées dans différents pays d'Afrique et d'Asie du Sud-Est.

 

Le Post-Doc est financé dans le contexte de l’Institut Convergences Agriculture Numérique #DigitAg ( www.hdigitag.fr ).

 

Salaire brut

 

2400 to 2800 based on previous professional experience.

 

Profil souhaité

 

Doctorat en informatique.

 

La préférence sera donnée aux candidats très motivés ayant une expérience de recherche en apprentissage automatique et profond, en science des données hétérogènes et en science des données appliquée à des sujets liés à la sécurité alimentaire et à l'agriculture.

 

De solides compétences en programmation (de préférence en Python) sont nécessaires.


Modalité de candidature :

 

Qualified applicants are invited to send their application to Roberto Interdonato (roberto.i...@cirad.fr) as a single pdf file containing a cover letter describing their research background and motivation, a detailed CV and the contact details of up to three referees.

 

Date limite pour l’envoi du dossier : 30 Juin, 2023

 

Bibliographie

 

[1] Hugo Deléglise, Roberto Interdonato, Agnès Bégué, Elodie Maître d'Hôtel, Maguelonne Teisseire, Mathieu Roche. Food security prediction from heterogeneous data combining machine and deep learning methods. Expert Syst. Appl. 190: 116189 (2022)

[2] Cheick Tidiane Ba, Chloé Choquet, Roberto Interdonato, Mathieu Roche. Explaining food security warning signals with YouTube transcriptions and local news articles. GoodIT 2022: 315-322

[3] Hugo Deléglise, Agnès Bégué, Roberto Interdonato, Elodie Maître d'Hôtel, Mathieu Roche, Maguelonne Teisseire. Mining News Articles Dealing with Food Security. ISMIS 2022: 63-73

Roberto Interdonato

unread,
May 26, 2023, 10:28:33 AM5/26/23
to ml-...@googlegroups.com

Post-doctoral position (18 months) at CIRAD - UMR TETIS, Montpellier, France

 

 

Topic

Development of a machine learning framework for the prediction of food security indicators at country scale from heterogeneous data

 

Keywords 

 

Data Science, Machine Learning, Food Security, Heterogeneous Data

 

Context

 

Food Security (FS) is a central problem in many areas of the world, as also testified by its presence as one of the 17 Sustainable Development Goals (SDG 2 - Zero Hunger).

To monitor food insecurity situations, several early warning systems are active today, such as GIEWS (Global Information and Early Warning System, FAO), and FEWSNET (Famine Early Warning Systems Network, USAID).

These systems use a limited set of data types, i.e., agroclimatic data from satellite images and indicators extracted from household surveys about nutritional, economical and production-related factors. Also, human intervention is often needed to combine and summarize all the sources of information.

Previous research carried on at TETIS proved that heterogeneous open data, related at different levels with food security, can be used to provide a machine learning based framework able to automatically produce FS indicators. The idea is to use data that represent proxies for the multiple and interrelated reasons behind this phenomenon. Some examples may be raster data containing spatial information, volunteered geographical information, meteorological data, quantitative economic indicators, and textual data from local news media.

 

The aim of this Post-Doc is to consolidate and extend the recent works on this topic [1], by exploiting state-of-the-art machine and deep learning strategies that can cope with challenging aspects, such as high data heterogeneity and sparse ground truth data. More specifically, the work is structured around three research questions:

 

- What types of open-access data can be targeted to produce Food Security indicators?

 

- How can data that are heterogeneous in terms of scientific domain, structure and spatio-temporal resolution be preprocessed to obtain consistent predictions of Food Security for a given study site?

 

- How can more recent machine and deep learning techniques be exploited and combined to process this heterogeneous data?

 

Besides the publication of original research articles in top scientific journals, the aim of this work will also be to develop a suite of user friendly data science methods (e.g., a Python library) able to retrieve and exploit heterogeneous data to effectively produce food security indicators at national scale.  The use of ground truth data provided by CIRAD and CGIAR will allow to test the proposed methodologies on different countries in Africa and South-Eastern Asia.

The Post-Doc is funded in the context of the #DigitAg Convergences Institute for Digital Agriculture ( www.hdigitag.fr ).

 

Gross Salary

 

2400 to 2800 based on previous professional experience.

 

Candidate profile

 

PhD in computer science.

 

Preference will be given to highly motivated candidates with research experience in machine and deep learning, heterogeneous data science and data science applied to topics related to food security and agriculture.

 

Strong programming skills (preferably in Python) are needed.


Application instructions:

 

Qualified applicants are invited to send their application to Roberto Interdonato (roberto.i...@cirad.fr) as a single pdf file containing a cover letter describing their research background and motivation, a detailed CV and the contact details of up to three referees.

 

Application deadline: June 30, 2023

 

Bibliography

Reply all
Reply to author
Forward
0 new messages