How Your Instagram Photos and Hashtags Are Training Facebook's AI
#Sunset, #Avocado, #Cat
Your Instagram #Dogs and #Cats are training Facebook’s AI and Wired has the full story.
Last week, Facebook disclosed an experiment they’ve been working on that uses public Instagram photos and their associated hashtags to train algorithms to categorize images for themselves.
The training set includes over 3.5 billion public Instagram photos labeled with 17,000 hashtags.
The result? Facebook set a new record for the number of correctly categorized images by a machine on the ImageNet test. Trained on 1 billion Instagram photos, their algorithms correctly identified 85.4 percent of images on the test. The record was previously held by Google whose algorithms identified 83.1 percent of images on the test earlier this year.
Read on for the full story of how Facebook had to use 336 high-powered graphics processors spread across 42 servers for more than three solid weeks to set this new record.
Gmail Can Practically Draft Emails For You
Google’s annual product keynote, Google I/O, was held yesterday and AI was a common theme throughout.
According to The Verge, one of the top new AI features being released by Google is the ability for Gmail to practically write emails for you.
Expanding on its Smart Reply feature, Smart Compose will soon be able to “help you draft emails from scratch, faster.” As you’re typing, Smart Compose will make suggestions for complete sentences.
Google also announced a new feature that will combine AI and Google Street View for augmented reality directions—a huge help for those navigating new cities.
Lastly, Google is overhauling its news app with artificial intelligence to bring you a curated news timeline with articles from all over the web.
Who To Follow in AI
Email subject writing AI tool Phrasee has a series featuring individuals involved in AI who are worth following on social media. Their featured follow this week is none other than Paul Roetzer, founder of the Marketing AI Institute. The full Q&A can be found here, but we’ve highlighted a few of the key takeaways below.
When asked the current landscape of AI and marketing, Roetzer replied, “The vast majority of marketing AI technology today is focused on using machine learning—a subset of AI—to make predictions based on historical data. The algorithms look at what has happened before and predict what will happen in the future.”
Taking it a step further, Roetzer’s predictions for the future of AI and marketing include, “The future of AI will be about prescribing strategies and tactics based on goals. The marketer’s primary role will be to curate and enhance algorithm-based recommendations rather than to devise them.”
The full profile addresses other big questions surrounding AI such as first steps for brands to get started with AI, areas where AI will have the biggest impact, and more.
At the Marketing AI Institute, we read dozens of articles on artificial intelligence every week to uncover the most valuable ones for our subscribers (become a subscriber today), and we curate them for you here. We call it 3 Links in 3 Minutes. Enjoy!