Photo of Computers recognize memorable images

Computers recognize memorable images

News Learning in Machines & Brains 14.06.2016

Why do some images stay fixed in our memories, while others quickly fade away? Researchers have developed a deep learning network that can provide an answer.

The program developed by a team led by Antonio Torralba (MIT), an associate fellow in CIFAR’s Learning in Machines & Brains program, examined tens of thousands of images that had already been ranked by human viewers in terms of memorability. Using this data, the network can look at a new image and predict how memorable it will be on a scale of zero to 10. And its predictions are as good as predictions made by humans.

“The network figures out what the elements are within a picture that make it memorable or not,” says Torralba.

He says he has been able to draw some general conclusions about what makes certain pictures stand out. For starters, faces: A picture that contains a human face is much more likely to be remembered than one that doesn’t.

“We never taught the network to do face detection,” Torralba says, “but the network realized that faces are memorable.”

Human bodies also help make a picture memorable, though not as effectively as faces. At the dull end of the spectrum are landscapes, and, especially, pictures with lots of water or sky.

“Landscapes are very boring,” Torralba says. “You could show people a picture of a sunset – people will say it’s beautiful – but no one will remember it later on.”

It’s perhaps not surprising that humans react to images of other humans – after all, Homo sapiens have lived in social groups since the dawn of our species. But Torralba believes that there’s another principle at work. People are constantly inventing narratives, he says, so a picture that suggests a story will stick in our memory more than a picture with seemingly-random elements. “If you have a story you can tell about a picture, it’s more likely that you’ll remember it,” he says. And it’s probably easier to construct stories about people than about trees and lakes.

The research was presented recently at the International Conference on Computer Vision.

There could be a number of applications for this research, Torralba says. Students might be better able to remember the material in a textbook, for example, if it contains memorable images. Similarly, authors and publishers can choose memorable images for book covers, while advertisers can design memorable images for magazine and billboard ads. And what if you’re just choosing a picture for your Facebook or Tinder profile? If you want people to remember you, best to go with a selfie rather than that Grand Canyon sunset.

 

 

Leave a Comment

Your email address will not be published.

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>

Related Ideas

Announcement | News

CIFAR Research Workshops: Call for Proposals

For more than three decades, CIFAR’s global research programs have connected many of the world’s best minds – across borders...

Announcement | Learning in Machines & Brains

Facebook announces major AI commitment to CIFAR

Facebook announced a major investment with CIFAR today, a result of the Institute’s leadership in the field of artificial intelligence...

News | Learning in Machines & Brains

Machines learn new ways of learning

Intelligent machines have learned to read and write, recognize images, and predict dangerous mutations. But how does a machine learn...

Video | Genetic Networks | Learning in Machines & Brains

Going Deep Into the Human Genome

CIFAR Fellow Brendan Frey’s start-up Deep Genomics is creating an AI system that allows us to look at our DNA...

News | Institutions, Organizations & Growth

Review: Daron Acemoglu on Robotics, AI and the Future of Work

As society looks to the future and the potential of AI and robotics, CIFAR Senior Fellow and MIT economist Daron...