Our overarching research goal spans across multiple directions with a common theme of Ecological Conservation. Following are a few of our research directions:
Animal Re-identification. Animal re-identification is fundamental for the study of population dynamics, community, and behavioural ecology. It is traditionally performed using tagging or DNA sampling, which is laborious, invasive to an animal, and expensive. For example, Maksim Kholiavchenko and Charles V Stewart have worked on a comprehensive deep learning pipeline for whale shark recognition. The goal is to further develop and apply these techniques to solve complex problems in animal re-identification. As another representative work, the HotSpotter algorithm for animal re-identification was developed by a team of researchers including Jonathan P. Crall, Charles V. Stewart, Tanya Y. Berger-Wolf, Daniel I. Rubenstein, and Siva R. Sundaresan. HotSpotter is a fast, accurate algorithm for identifying individual animals against a labeled database. It is not species-specific and has been applied to various animals such as Grevy’s and plains zebras, giraffes, leopards, and lionfish. HotSpotter offers a non-invasive, scalable, and cost-effective alternative that leverages the widespread availability of digital cameras and the wealth of photographic data for animal population analysis. This approach allows anyone with a camera — scientists, ecotourists, and even ordinary citizens — to contribute to the collection of data on animal populations. The algorithm is based on extracting and matching keypoints or “hotspots” from images, and it can match each query image in just a few seconds. This makes it a powerful tool for ecological studies and conservation efforts.
AI for Biology. AI helps to automate and simplify image analysis, predict protein structures, and aid drug discovery. Researchers are increasingly turning to artificial intelligence in biology to address complex problems. The Imageomics Institute is also working on exciting problems in AI for Biology. The goal is to develop AI models and tools that can help biologists analyze and interpret complex biological data. One of the notable contributions is the development of BioCLIP, a foundation model for the tree of life. Built using the CLIP architecture, BioCLIP serves as a vision model for general organismal biology. BioCLIP is trained on TreeOfLife-10M, a specially-created dataset that covers over 450K taxa, making it the most biologically diverse machine learning-ready dataset available to date. The model has been rigorously benchmarked on a diverse set of fine-grained biological classification tasks, consistently outperforming existing baselines by 17% to 20%.
Computer Vision for Ecological Conservation. Charles V Stewart has made significant contributions to the field of computer vision with applications in ecology and environmental conservation. He has worked on the Wildbook project, where he led much of the research in computer vision and machine learning. His work aims to develop and apply computer vision techniques to solve complex problems in ecological conservation. A Framework for Autonomic Computing for In Situ Imageomics is another representative work in this direction by members of the lab and other partners from different universities.
Intelligent Cyberinfrastructure with Computational Learning. The ICICLE institute aims to build the next generation of Cyberinfrastructure to render Artificial Intelligence (AI) more accessible to everyone and drive its further democratization in the larger society. The goal is to develop a national Cyberinfrastructure for AI that can seamlessly adapt to the changing data and infrastructure needs of the end user.