Wildbook: Shared AI Infrastructure for Wildlife Monitoring

Jason Allan Holmberg

Funding Area: Critical shared infrastructure / Infraestructura compartida critica

Problem Statement: According to a 2017 study in the Proceedings of the National Academy of Sciences, a “sixth mass extinction” is underway, a trend signaled by widespread vertebrate losses that “will have negative cascading consequences on ecosystem functioning and services vital to sustaining civilization.” This study represents a growing awareness that more rapid animal population assessment and response are needed to counter this decline. Unfortunately, the collection and management of wildlife data remains a largely ad hoc and academic exercise focused on moving small data sets (often in Excel and Access) into local population studies for “one-off” analyses without long-term data curation or collaboration across regions. Arriving at a critical mass of data can take years, and manual data processing can create long delays for scientific results. Simply put: wildlife research needs access to advanced and affordable data science tools to collect more data, analyze it faster, collaborate across regions and disciplines, engage the public in focused conservation, and use data to continuously optimize solutions to prevent extinction. Wild Me (wildme.org) supports 1400+ wildlife researchers across the globe with shared data management and machine learning infrastructure to promote collaboration and support local efforts to fight extinction.

Proposed Activities: Wild Me (wildme.org) is a 501c3 nonprofit organization uniquely composed of software and machine learning engineers supporting marine and terrestrial wildlife researchers across the globe. We build and support collaborative research platforms that put cutting-edge AI, data sharing, and analytics in the hands of local wildlife experts, allowing them to rapidly assess population sizes using computer vision to track individual animals across scientific, tourist, and even social media images and videos. Wild Me's technology gives researchers the ability to build animal profiles (sightings, behavior, social groups, and more), analyze this growing data faster, and use the resulting population estimates to advocate for the local changes needed to truly stop extinction. You can read more about how Wild Me provides shared computer vision infrastructure and runs a computer vision pipeline for the wildlife research community here: https://docs.wildme.org/product-docs/en/wildbook/introduction/image-analysis-pipeline/ You can see a complete list of our infrastructure platforms: https://www.wildme.org/platforms.html Request: IT Infrastructure Support From African wild dogs to whale sharks, and from jaguars to leafy seadragons, Wild Me technology is supporting efforts to protect a wide range of species. We operate our IT infrastructure on Azure cloud virtual machines (20+) and colocated deep learning servers (3). Storage and compute costs to provide AI, good data management, and new collaboration opportunities to our growing community of users represent a monthly expense of approximately $12,000 and growing. Accounting for growth in storage costs as our total data volume increases monthly, we anticipate IT costs of approximately $148,000 in 2023. From 2018 to the end of 2022, Wild Me received Azure compute credits to largely cover our needs. With the natural end of the Microsoft AI for Earth five-year program, we need to raise funding to help cover this critical expense, which grows every month as we add new data, users, and species to our online platforms. We are requesting support to help us empower 1400+ biologists, management authorities, and volunteers with advanced technology for population monitoring. From custom data schemas to modern computer vision, we provide tools that conservationists generally lack the technical skills and funding to develop, consume, and support. As a nonprofit, we provide it for free, bringing browser-base AI to local efforts to fight extinction.

Openness: Wild Me as a nonprofit organization is open source, open AI model, and open infrastructure, hosting wildlife data and providing AI model inference (computer vision) for 1400+ researchers across the globe. Our code repositories are here: https://github.com/wildmeorg You can see how we transparently support the wildlife community here: https://community.wildme.org/ We provide access to any wildlife researcher who requests it, supporting local and collaborative research into species on all seven continents. Wild Me servers and infrastructure are open for data storage, analysis, collaboration, and AI inference. However, we support this infrastructure with a staffed support position and on-staff software and ML engineers, providing professional support and responsiveness for community infrastructure.

Challenges: Wild Me is request support to keep its shared infrastructure for wildlife monitporing up and running. As such, we do not anticipate new challenges but rather the standard challenges of managing software and server infrastructure for a very skilled and demanding community. We manage these challenges openly and transparently through communication on https://community.wildme.org.

Neglectedness: We have applied for similar funding from the Patrick J. McGovern Foundation. To date, this funding request has not been approved but is under consideration.

Success: Success would be conintuing support and 100% uptime for two claendar months of all of the Wild Me infrastructure and all 1400+ users on our platforms listed here: https://www.wildme.org/platforms.html. Success would also be measured in user growth (adding new researchers for new species) as well as data growth and growth in requested AI inference.

Total Budget: 25,000

Affiliations: Wild Me (wildme.org) is a U.S. 501c3 nonprofit organization.

LMIE Carveout: No, though we serve many users (wildlife researchers) in LMIE locations.

Team Skills: Jason Holmberg (USA) - Executive Director and Founder of Wild Me - Jason is a 2022 The Tech for Global Good Laureate, a current NSF Principal Investigator, and has 20+ years of experience integrating computer vision and collaborative data management through various Wildbook platforms serving marine and terrestrial researchers across the globe. Jason’s original experiments in Intelligent Agents for whale sharks (Rhincodon typus) are the foundation of this proposal. Lasha Otarashvili (Republic of Georgia) - Machine Learning Engineer - Lasha joined Wild Me in 2022 and has already advanced the state of the art in computer vision for individual animal re-ID using AI. Lasha developed the new embeddings-based MiewId algorithm that can match individual African lion faces, dolphin fins, and more across large catalogs using AI. Wildbook Community (global) - Wild Me directly supports a global community of 1400+ researchers and volunteers who will benefit from our project, gaining access to shared, AI-backed infrastructure. You can see how Wild Me directly interacts with and supports this community on a daily basis at: https://community.wildme.org. You can see a list of the science generated by Wild Me and our community at: https://www.wildme.org/publications.html

Submission Number: 190