Instagram is a social media platform that is full of photos and short videos. It is one of the biggest image databases in the past 10 years, and Facebook, which owns Instagram, is using this platform to train AI to identify what is in a picture.
Facebook trains A.I to identify Instagram pictures
On Feb. 4, Facebook announced that it had created an artificial intelligence program to identify what it is looking at. The AI could do this after Facebook fed it with more than 1 billion public images from Instagram.
Facebook's computer vision program named SEER was able to outperform existing AI models in an object recognition test. The AI got a classification accuracy score of 84.2% when it attempted a test provided by ImageNet, a massive visual database created to be used in visual objection recognition software research.
ImageNet can test whether an AI program can identify what is in the image fed to it, according to The Verge.
Facebook's new approach
Numerous AI models in the past were trained by using labeled datasets, but SEER learned how to identify images by analyzing random, uncurated, and unlabeled images from Instagram, according to Facebook. This technique is known as self-supervised learning.
Facebook's researchers wrote in a blog post that the future of AI is in creating systems that can learn directly from whatever information they are given, whether it is text, images, or another type of data, without relying on curated and labeled data that sets to notify them how to recognize objects in an image, interpret a text or perform other tasks that they ask it to.
The researchers added that SEER's performance demonstrates that self-supervised learning can excel at computer vision tasks in real-world settings and that it is a breakthrough that clears the path for more flexible, accurate, and adaptable computer vision models in the future.
Even though this is only a project and is not ready to be launched yet, a spokesperson from Facebook stated that the potential uses were broad.
They include improved automatically generated text for describing images to people who have visual impairments, automatic categorization of items that were sold on the Marketplace on Facebook, and better systems to keep harmful images from being seen on the platform.
Despite the impressive performance of SEER, numerous Instagram users were surprised to know that their images were used to train AI systems, according to TechCrunch.
Priya Goyal, a software engineer at Facebook AI Research, told CNBC in an interview that they told Instagram users that in their data policy, they can use the information placed on the platform to support their research their innovation, including AI projects.
Meanwhile, Facebook stated that the will open source some of the software so that other researchers can study the AI too.
Goyal added that they are currently sharing the details of their research, and they are creating an open-sourced library that will enable other researchers to use self-supervised learning to train models on uncurated images, but they will not share SEER mode or the images.
Related Article: Facebook Reportedly Developing an AI to Summarize News Articles
This article is owned by Tech Times
Written by Sieeka Khan