Deadline: 31-Aug-21
The Agence Nationale De La Recherche (ANR) is pleased to announce the Challenge IA-Biodiv: Research in Artificial Intelligence in Field of Biodiversity to develop research and knowledge about biodiversity, with goals including defining integrative indicators to quantify and map the state of biodiversity.
The IA-Biodiv challenge is intended for scientific communities in AI and biodiversity in France and AFD’s partner countries in Africa. International consortia will be financed by Expertise France (mandated by AFD) according to the participation modalities for international consortia.
The challenge aims to develop AI research for better exploration and exploitation of biodiversity data.
Nowadays, biodiversity monitoring and management is based on indicators constructed mainly on field data without always being able to transcribe the complexity of ecosystems, interactions and feedback that play out, non-linear trajectories, threshold effects or even magnitude of the effects.
Objectives
- Objective 1: designing AI methods optimised for biodiversity research – The AI research objective is to design:
- explainable methods (e.g. by considering approaches based on reasoning about structured data and knowledge representation) or at least interpretable methods by experts in the field (ecologists, biologists etc.), for example by visualising the input data that led to the final decision (LIME, Grad-CAM, Deconvolution etc.),
- methods requiring as few costly annotations as possible, possibly considering (inter)active learning approaches, transfer learning, domain adaptation and optimal transport, budget learning and all the variants of zero-shot/one-shot learning, self-learning etc.
- incremental methods enabling the expert in the field to be kept at the heart of the decisionmaking process. To achieve this, the incremental methods must be able to adapt to dynamic changes in the input data (spatial, temporal or functional evolution or changes through interactions with the user), accounting for simultaneously different levels of analysis (for instance using game theory methods).
- Objective 2: designing models and predictive indicators for biodiversity research – The goal is to help advance knowledge in the field of biodiversity, including the predictive or functional modelling of systems and ecosystems and the possibility of producing predictive indicators by exploring AI solutions
- Objective 3: designing hybrid AI models – The aim here is to concentrate on reasoning-based approaches using structured data (a knowledge representation) and machine learning approaches using unstructured data. The AI research questions in this context go beyond meta-learning or relational machine learning. All three dimensions of AI research should be considered: symbolic approaches (knowledge driven), subsymbolic approaches (data driven) and combined methods.
The IA-Biodiv challenge is based on three pillars: scientific coordination, evaluation of AI systems and the IA-BiodivNet virtual research environment. As the central player, the Operational Consortium (COpé) manages the organisation and process of the challenge.
Eligibility Criteria
- Project proposals are submitted by the lead scientific coordinator by agreement with the project partners who constitute the consortium.
- However, when a project is selected for funding, the grant is allocated to the institution (legal person), which will be subject to a number of obligations.
- The scientific coordinators of each partner must therefore approve the project proposal that commits their institution (research organisation) before it is submitted for potential funding.
For more information, visit https://anr.fr/en/call-for-proposals-details/call/challenge-ia-biodiv-research-in-artificial-intelligence-in-the-field-of-biodiversity/
