who: Grant Van Horn

when: 10/10 (Thursday) 11:45a – 1:15p

where: Computer Science Building Rm 150

food: Athena’s Pizza
generous sponsor:  ORACLE LABS

Visipedia + iNaturalist:
integrating machine learning into a naturalist community

Abstract:
iNaturalist is a community of over 2 million nature enthusiasts that share observations of the natural world. These observations consist of at least one photo, a GPS coordinate, and a taxonomic label from the tree of life. The iNaturalist community has recorded over 30 million observations of organisms from around the world. These recordings are exported to GBIF for scientists and conservationists to use in research and reports. As the iNaturalist community grows and the expertise becomes more diffuse, the need for automation becomes crucial for maintaining an engaging user experience and for ensuring the accuracy of data exported to GBIF. In this talk I will describe our work on learning the taxonomic identification skills of iNaturalist users, using these skills to assign taxonomic labels to images, and subsequently training computer vision systems on the resulting dataset. These computer vision systems can be accessed in the Apple App Store or Google Play Store in both the “iNaturalist” and “Seek by iNaturalist” apps as well as the iNaturalist website.

Bio:
Grant is a postdoc at Cornell Lab of Ornithology. He earned his PhD at Caltech advised by Prof. Pietro Perona, and BS and MS degrees at UCSD advised by Prof. Serge Belongie. His research focuses on efficient dataset collection and fine-grained visual categorization, particularly for the natural world. He works closely with iNaturalist and the Cornell Lab of Ornithology where he puts his research into production.