Expo Talk Panel

Wild Me is a not-for-profit based in Portland, OR, USA that works directly with ecologists around the world to automate wildlife conservation. This talk covers the concepts of wildlife conservation and how it uses statistics to monitor an animal population, presents a motivating use-case for how machine learning can be used for social good, and details some of the specific machine learning algorithms and approaches that are used in the field. One of our premier ID platforms, Flukebook (flukebook.org), has helped to power state-of-the-art mark-recapture research and publications; Flukebook now supports 9 cetacean species (humpback whale, 2 right whale species, sperm whale, orca, and 4 dolphin species), has over 220,000 reported sightings, serves hundreds of collaborators, and exposes 7 unique computer vision ID algorithms (HotSpotter, CurvRank for flukes and dorsal fins, the winning 2016 Kaggle competition ID algorithm for right whales by Deepsense.ai, two learned triplet-loss embedding ID algorithms, and more). We will do a deep dive into our deep learning stack, detection pipeline, and ID algorithms, including: image classification, bounding box regression, instance classification, class segmentation, object of interest (AoI) classification, triplet-loss embedding computations, and more.  Our deep learning stack utilizes Theano and PyTorch, the NVIDIA's CUDA, CNMeM, and CuDNN deep learning stack, and employs NVIDIA GPU hardware. The Flukebook platform also integrates "citizen science" input into conservation research through the help of an intelligent agent. Our agent automatically ingest video data from YouTube using NLP and OCR, plus image sightings reported via Twitter, and feeds them into our machine learning pipeline. Join our session and listen about the social good that machine learning can achieve as Wild Me helps to modernize wildlife conservation as a data-driven science. All code is available and open-source at github.com/wildbookorg.

Chat is not available.