This week in AI and Machine Learning: Google’s ambitious translation project, Meta’s protein library, and more.
Investments in computer vision are on the rise, but enterprises still overwhelmingly fail to see the results they expect due to a number of common obstacles. In our latest blog, we discuss how taking a solution-centric approach to computer vision strategies provides for a thorough understanding of business challenges and repeatable innovation for dependable, scalable solutions. Check it out.
Google’s 1,000 Languages Initiative
Google is embarking on an ambitious, multi-year initiative to introduce a single multi-modal translation model capable of comprehending each of the world’s 1,000 most widely-spoken languages. The project is supported by Google’s new Universal Speech Model, which has the broadest reach of any existing translation model with more than 400 included languages to date.
Zoubin Ghahramani, VP of research at Google AI, discusses the decision to create a single, all-encompassing model as opposed to individual models for each language. In an interview with The Verge Ghahramani notes, “By having a single model that is exposed to and trained on many different languages, we get much better performance on our low resource languages.” The approach takes advantage of the similarities between languages, finding shortcuts in their evolutionary history.
This initiative is one of a number of projects and solutions unveiled by Google recently. The tech giant recently unveiled its Vertex Vision AI platform, an AI-powered writing assistant, and its AI Test Kitchen for trying out prototypes. Read more details on the 1,000 Languages Initiative on Google’s blog and in this Verge write-up.
AI for Tracking and Cracking Down on Illegal Fishing
Illegal fishing threatens a number of at-risk and endangered species, with Science reporting that annual catches total around $25 billion. International maritime law requires vessels to use automated beacons to display their locations in real time—known as Automatic identification systems (AIS)—which are intended to prevent collisions and help authorities crack down on unlawful activity. Sometimes, however, beacons go silent as a result of accidents or because ships want to explore promising stretches of ocean without tipping off the competition. Researchers are hopeful that machine learning algorithms can help them learn to recognize the differences between illegal AIS deactivation and more innocent instances.
Training a model to recognize suspicious activity involved manually reviewing datasets to identify examples where AIS issues appeared that were related to technical issues as opposed to deliberate illegal activity. Altogether, researchers identified more than 55,000 examples of likely illegal AIS deactivation between 2017 and 2019. In total, 5269 fishing vessels disappeared for millions of combined hours, with some evading detection for days or weeks at a time. Learn more about the ongoing research funded by the U.S. National Oceanographic and Atmospheric Administration’s Office of Law Enforcement.
Meta’s Contributions to Protein Research
Earlier this year, London-based DeepMind unveiled an AI network known as AlphaFold. The solution enables researchers to assess the shape of nearly every protein in the known universe as easily as they might conduct a Google search. This week, Meta entered the world of protein research by launching its ESM Metagenomic Atlas. The database will help researchers gain an understanding of some of the world’s least understood proteins, including those found in the depths of the ocean and inside the human body.
Meta hopes their efforts will make AI into a tool for deciphering the language of proteins and addressign use cases related to fighting disease, protecting the environment, and more. Check out Meta’s summary of the project, explore the database, and view the full research paper.
Join our Community
See you next week! Until then, keep the conversation going on Plainsight’s AI Slack Channel.
About the Author & Plainsight
Bennett Glace is a B2B technology content writer and cinephile from Philadelphia. He helps Plainsight in its mission to make vision AI success repeatable, scalable and traceable for enterprises across industries.
Plainsight provides the unique combination of AI strategy, a visual data science platform, and deep learning expertise to develop, implement, and oversee transformative computer vision solutions for enterprises. From solution-centric strategy, through model deployment, and ongoing monitoring and oversight, Plainsight empowers enterprises to create and operationalize responsible vision AI applications for solving high business challenges.