
MIT algorithm understands how memorable images are, feeding into artificial intelligence deep learning and shining new light on memorability.
MIT algorithm understands how memorable images are, feeding into artificial intelligence deep learning and shining new light on memorability.
The research team with MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) created the algorithm. But the team’s plans don’t stop here.The algorithm, available for try-out online could soon be turned into an app. What would we do with such an app? Well, for starters check out how our memorable our selfies and profile pictures are.
It might seem trivial, but the memorability checking algorithm could become a crucial factor in advertising, marketing and other largely imagery dominated sectors. The algorithm is called MemNet if you’re planning to check the memorability score of your own images online.
Aditya Khosla, graduate student with MIT’s CSAIL and lead author of the paper explaining the mechanism behind the MemNet algorithm declared that:
“Understanding memorability can help us make systems to capture the most important information, or, conversely, to store information that humans will most likely forget”.
Simply put, MemNet algorithm functions kind of like an instant focus group that will allow you to gain insight on how valuable a visual message is. MemNet algorithm is based on previous research conducted by the CSAIL team. The most exciting thing about the newly developed algorithm is that is can perform almost at the same level as humans.
Based on deep learning, an area of artificial intelligence, the MemNet algorithm makes use of neural connections to find relevant data and create unique patterns of recognition for each case. Facebook’s photo-tagging feature or Apple’s personal assistant Siri are also created based on deep learning. The neural networks underpinning MemNet algorithm’s functioning are organized in layers and layers of processing units. Each of the units is performing computations on the data fed into the algorithm in succession. The more data it receives, the better and more accurate the final deliberations are.
Thousands of images were analyzed with MemNet algorithm to test its capabilities. When the results were compared to how humans perceived the images in terms of memorability, MemNet algorithm was just a few percentage points behind human performance. The MIT algorithm understands how memorable images are or aren’t for that matter. One thing is certain. MemNet algorithm fared 30 percent above any other algorithm working on the same principle.
Photo Credits: i.kinja-img.com