An Algorithm Can Tell How Forgettable Your Selfies Are

BuzzfeedVideo on Facebook
BuzzfeedVideo on Facebook / BuzzfeedVideo on Facebook

According to a recent survey, the average millennial will take over 25,000 selfies in his/her lifetime. Science Alert reports that researchers at the Computer Science and Artificial Intelligence Laboratory (CSAIL) at the Massachusetts Institute of Technology (MIT) have developed a new algorithm that could help shutterbugs better distinguish between the forgettable and the memorable photos before posting them to social media accounts.

The algorithm is called "MemNet," and was actually designed to work for all photos, not just self-portraits. According to a CSAIL report [PDF] authored by graduate student Aditya Khosla and his MIT colleagues, MemNet uses a form of artificial intelligence called deep learning (also used in Google's Smart Reply automatic email system and in Apple's Siri) to develop ways to process data and find patterns on its own. "While deep-learning has propelled much progress in object recognition and scene understanding, predicting human memory has often been viewed as a higher-level cognitive process that computer scientists will never be able to tackle," principal research scientist Aude Oliva said. "Well we can, and we did!" 

To teach the algorithm what makes photos memorable, the researchers used the La Mem (large-scale image memorability) database of over 60,000 images that had been previously given "memorability scores" based on how well humans remembered them in trial experiments conducted online. After MemNet was able to find patterns for itself, it was tested against human subjects. "It performed 30 percent better than existing algorithms and was within a few percentage points of the average human performance," the CSAIL researchers said of the algorithm, which also creates a heat map of each image to highlight the most memorable section.

Examples from the La Mem gallery with high memscores. // La Mem

The researchers hope that the technology will help reveal more about how and what people remember. "This sort of research gives us a better understanding of the visual information that people pay attention to," UC Berkley Associate Professor Alexei Efros said. "For marketers, movie-makers and other content creators, being able to model your mental state as you look at something is an exciting new direction to explore."

To see how memorable your photos are, try the LaMem demo site created by the team at MIT.

[h/t: Science Alert]