Scaling species distribution models with deep learning
Event details
| Date | 06.07.2023 |
| Hour | 14:00 › 16:00 |
| Speaker | Robin Zbinden |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Maria Brbic
Thesis advisor: Prof. Devis Tuia
Co-examiner: Prof. Stéphane Joost
Abstract
Species distribution models (SDMs) relate environmental conditions to the presence of species and play an important role in conservation ecology. Nonetheless, the development of reliable SDMs encounters obstacles such as limited data availability and selection bias, which hinder their predictive capability and transferability. Recently, advanced machine learning and deep learning approaches have emerged to address similar challenges, presenting exciting prospects to enhance the performance and generalizability of SDMs. This proposal aims to investigate the viability of employing these methods in SDMs and explore their adaptability within this particular domain.
Specifically, we discuss three existing works and their relevance to our research. We highlight first the inherent difficulties in SDMs and explore standard approaches. We then delve into recent machine learning techniques that can be leveraged to address these challenges effectively. Finally, we outline our ongoing efforts and future research directions that aim to integrate deep learning methods into SDMs.
Background papers
Exam president: Prof. Maria Brbic
Thesis advisor: Prof. Devis Tuia
Co-examiner: Prof. Stéphane Joost
Abstract
Species distribution models (SDMs) relate environmental conditions to the presence of species and play an important role in conservation ecology. Nonetheless, the development of reliable SDMs encounters obstacles such as limited data availability and selection bias, which hinder their predictive capability and transferability. Recently, advanced machine learning and deep learning approaches have emerged to address similar challenges, presenting exciting prospects to enhance the performance and generalizability of SDMs. This proposal aims to investigate the viability of employing these methods in SDMs and explore their adaptability within this particular domain.
Specifically, we discuss three existing works and their relevance to our research. We highlight first the inherent difficulties in SDMs and explore standard approaches. We then delve into recent machine learning techniques that can be leveraged to address these challenges effectively. Finally, we outline our ongoing efforts and future research directions that aim to integrate deep learning methods into SDMs.
Background papers
- A maximum entropy approach to species distribution modeling By Steven J. Phillips, Miroslav Dudík, Robert E. Schapire
https://dl.acm.org/doi/10.1145/1015330.1015412 - Presence-Only Geographical Priors for Fine-Grained Image Classification, By Oisin Mac Aodha, Elijah Cole, Pietro Perona
https://ieeexplore.ieee.org/document/9008116 - A Simple Framework for Contrastive Learning of Visual Representations, By Ting Chen, Simon Kornblith, Mohammad Norouzi, Geoffrey Hinton
https://proceedings.mlr.press/v119/chen20j.html
Practical information
- General public
- Free
Contact
- edic@epfl.ch