¡Hola! I'm an applied mathematician passionate about artificial intelligence for social impact and sustainability. My long-term goal is to learn, understand, and construct computational models against the global ecological crisis.
I’m Juan Sebastián, a PhD student in the People and Nature Lab and the CBER at University College London (UCL). My research interests span the fields of artificial intelligence, biodiversity monitoring, sustainability, and social impact, with a particular focus on utilizing bioacoustics and deep learning. Currently, I’m working on understanding the drivers of population change in different taxonomic groups such as amphibians, insects, and bats through creating machine learning systems that work on real-world monitoring networks in Europe and the Neotropics.
Previously, I was a researcher at the Humboldt Institute in Colombia and a collaborating researcher at the Mila-Quebec AI Institute in Montréal, living between Audiomoths and Moths. My undergraduate studies were at the National University of Colombia in the Mathematics Department where I worked on interpretability in healthcare and diverse problems in policy.
I am also part of the Ecoacoustics Colombian Network (Red Ecoacústica Colombiana - REC) where I work on different projects around education, environmental policy, and artificial intelligence for the Neotropics.
GitHub / LinkedIn / Twitter / Google Scholar
Jain, A., Cunha, F., Bunsen, M., Cañas, J.S., …, Rolnick, D. Insect Identification in the Wild: The AMI Dataset. ECCV 2024 (2024). https://arxiv.org/abs/2406.12452[Webpage]
Cañas, J.S., Toro-Gomez, M.P., Sugai, L.S.M., …, Ulloa, J.S. A dataset for benchmarking Neotropical anuran calls identification in passive acoustic monitoring. Scientific Data 10, 771 (2023). https://doi.org/10. 1038/s41597-023-02666-2 [Webpage]
Cañas, J.S., Gomez, F., Costilla-Reyes, O. Counterfactual Explanations and Predictive Models to Enhance Clinical Decision-Making in Schizophrenia using Digital Phenotyping. arXiv:2306.03980 (Preprint) (2023). https://arxiv.org/abs/2306.03980
Ecology and biodiversity monitoring have a symbiotic relationship with machine learning. On one hand, machine learning offers a set of novel methods to process, extract, and analyze ecological data. On the other hand, ecology could be seen as a playground for exploring the scientific foundations of machine learning. Ecological systems are unique in scale, complexity, and richness. These problems often involve multiple parties for combinatorial decisions in highly dynamic and uncertain environments. Hence, machine learning challenges arise such as domain shift, spatiotemporal correlations, fine-grained categories, learning across data modalities, self-supervised representation learning, human-machine collaboration, and long-tailed distributions. Finally, I’m also interested in how to work together with local communities to co-create conservation tools that have an impact on biodiversity conservation.
If you’d like to discuss anything above, feel free to email me at juan.canas (at) ucl.ac.uk