Our paper on pseudo-absences in Species Distribution Modeling with Deep Learning is published!.. With @nina_vTiel @b_kellenb @LloydHughesZA @devistuia, we address geographic biases, class imbalance, and presence-only samples in citizen science datasets. 🧵

RT @zdeborova: I would like to think about this as an invitation that physics extends towards AI/ML to think like physicists: -"Let's figur….

Join millions who have switched to Grok.

RT @josauder: Interested in underwater SfM/SLAM?🪸📸🌊🤿. I will present our approach to remove detrimental caustics and backscatter from singl….

RT @devistuia: 🚨 job alert!.We just opened a PhD position in my lab for a crazy interesting and important project with @ETH_en and the @ICR….

RT @b_kellenb: Today, I have started my new position as Lecturer/researcher in the People & Nature Lab at University College London (@ucl,….

What is the best time to go to lunch to avoid queues at EPFL? 🕧🥗.Are more energy drinks and coffees bought during exam sessions? 🥤☕.These are the types of questions we explored in this study, to which I contributed during my master's!.

7/7. Additionally, take a look at our other paper (, presented at the CCAI workshop at ICLR 2024; we tested our loss on other citizen science datasets, achieving great performance on GeoLifeCLEF 2023. Code: Feel free to reach out!.

6/7. Through qualitative analyses, we also demonstrate the role of pseudo-absence samples in mitigating geographic biases.

5/7. Our loss function achieves superior results on average compared to previously defined losses (we test many alternative losses in the Appendix). We show that our loss function performs particularly well on species with few observations, which are often rare and endangered.

4/7. We introduce a novel loss function that aims to address the aforementioned challenges. The different terms within the function serve various purposes and are weighted to account for dataset biases. Check out the paper for more detailed mathematical explanations.

3/7. In addition, strong taxonomic biases (favoring iconic or common species over neglected or rare ones) and geographic biases (mainly due to accessibility) lead to significant class imbalances and spatially biased predictions.

2/7. However, datasets of species observations often contain only presence data, lacking information about species absences. This is generally the case in citizen science datasets, such as iNaturalist. To obtain a contrast with presences, pseudo-absences need to be added.

1/7. Species distribution models aim to uncover the complex relationships between environmental conditions at a given location and the likelihood of species occurring there. Using deep learning, the distributions of multiple species can be mapped simultaneously.

RT @devistuia: For the first paper of @nina_vTiel ‘s PhD, we embarked in the adventure of mapping 10’000 species of trees 🌲 globally and st….

RT @diego_marcos_g: PhD position on NLP+Earth observation at Inria Montpellier: In the context of the GEO-ReSeT pro….

RT @hager_radi: If you are @iclr_conf, come catch us at the poster session of “Tackling Climate Change with ML” at 2:30 where we present ou….

Interested in addressing geographic biases, class imbalances, and presence-only data in species distribution modeling? Don't miss our poster at the @ClimateChangeAI workshop at @iclr_conf. Collaborative work with @nina_vTiel @MarcCoru @devistuia

RT @MarcCoru: Heavy poster teamwork action today with @ZbindenRobin @kklmmr at #ICLR2024

RT @MarcCoru: Come join us at our spotlight 🤩 poster today at #ICLR2024. @kklmmr @ZbindenRobin, and I will present it in Hall B #266 Wed….

RT @josauder: Towards real-time 3D mapping of coral reefs with deep learning using only videos from cheap cameras, now published in Methods….