Students Build Machine Learning Algorithm More Powerful than Google’s
At the Marketing AI Institute, we read dozens of articles on artificial intelligence every week to uncover the most valuable ones for our subscribers and we curate them for you here. We call it 3 Links in 3 Minutes. Enjoy!
Advanced AI Isn’t Just For Tech Giants
Elite programmers aren’t the only people capable of advancing artificial intelligence. This week, a small group of students created an AI algorithm that outperforms code from Google’s researchers.
According to Technology Review, the part-time students at Fast.ai, a small organization that runs free machine learning courses online, were able to compete with Google because they did a lot of simple things really well. Cropping images correctly and introducing progressive resizing (training the system on smaller images at the beginning so it can make rapid progression when the model is most inaccurate) were a few of these tricks.
The result is an algorithm that was trained on the ImageNet database in 18 minutes, which is about 40 percent better than Google’s effort. This feat demonstrates that resources and hardware are just part of the equation for advancing artificial intelligence—a solid understanding of the technology and some creativity are essential, as well.
Facebook Research Releases Machine Learning Video Series
Whether you’ve already mastered all the online AI courses we’ve recommended or you’re just getting started, you should check out Facebook’s new machine learning video series “The Facebook FIeld Guide to Machine Learning.”
The six-part series shares best practices and provides practical tips about how to apply machine-learning capabilities to real-world problems.
The videos are broken down into six steps for understanding machine learning including problem definition, data, evaluation, features, model, and experimentation. Examples are used in each section to highlight how the decisions you make along the way can help you successfully apply machine learning to your product or use case.
Each video is less than 10 minutes, making it possible to complete in under an hour. The full series can be found here.
How AI is Closing the Gender Gap in Science
Meet Quicksilver, the AI tool helping overcome gender bias in science on Wikipedia by covering overlooked scientists, many of which are women.
Primer is to thank for training Quicksilver’s AI. Their method started by feeding the system 30,000 scientist Wikipedia entries including the Wikipedia articles themselves, scientists’ Wikidata entries, and more than three million sentences from news coverage of the scientists.
The next step included feeding Quicksilver 200,000 names and affiliations who have written scientific papers. In less than 24 hours, the tool had determined that 40,000 scientists didn’t have Wikipedia pages even though they had been covered in the news just as much as those scientists with pages.