Google to Integrate Biodiversity Research AI into Tensorflow Hub

2019-10-28

Machine learning algorithms have been widely used in biodiversity research, but most of them lack proper attribution or supervision. In order to improve the academic level, Google announced that it will release an institutional artificial intelligence workflow developed in collaboration with the Global Biodiversity Information Facility (GBIF), naturalists, and Visipedia. The researchers at this tech giant stated that this workflow will support cross-team data aggregation and collaboration, while ensuring compliance with standardized licensing terms, using compatible file formats, and providing fair and comprehensive data coverage for the tasks at hand. "The prospect of using machine learning for species identification is about to be realized, revealing its transformative potential in biodiversity research," wrote visiting professor Serge Belongie and Google Research Engineering Director Hartwig Adam in an academic article to be published at the next Leiden Biodiversity Conference in the Netherlands. "The highlight competitions at international workshops often involve developing state-of-the-art classification algorithms based on images from wildlife camera traps, herbarium specimens, and plant specimen tables. The inspiring results from these competitions will motivate us to expand the availability of biodiversity datasets and ML models from workshop scale to a global scale." The workflow will consist of two parts: datasets packaged by GBIF and models trained and released by Google in collaboration with Visipedia. The former will be reviewed to ensure they meet baseline licensing and citation requirements, and they will be published with digital object identifiers (persistent identifiers or handles used to uniquely identify objects) and linked through DOI references from the International Organization for Standardization's DOI referencing scheme. Meanwhile, the latter will be provided with documentation from TensorFlow Hub (Google's public repository of machine learning models), which will include information on sources, architectures, license information, and interactive model demos running on user-provided images. Belongie and Adam stated, "The core of academic research tradition lies in the conventions of citation and attribution, so when ML extends its tentacles into the life sciences, it should bring appropriate counterparts of these conventions. More broadly, people are increasingly aware of the importance of ethics, fairness, and transparency in the ML community... We look forward to collaborating with institutions around the world to achieve innovative applications of machine learning in biodiversity." Machine learning has a long history and has made significant contributions in various fields, including the field of biology. However, research on diversity is still in its early stages, and Google's exploration, regardless of the outcome, will accumulate important experience in this field.

RECENT AI NEWS

RECENT AI TOOLS