Here's some good news for those who love taking selfies and want keep those 'likes' coming along. A group of programmers from the Massachusetts Institute of Technology (MIT) Computer Science and Artificial Intelligence Lab (CSAIL) came up with an algorithm that can determine how memorable a photo can be and they are giving everyone a free demo access to it right now!
You may shy away from it in the meantime since the algorithm is used by artificial intelligence and it can't be as good as human memory, right? Actually, the proponents of the study, Aditya Khosla, Akhil S. Raju, Antonio Torralba and Aude Oliva, tested their algorithm several times and fed the program thousands of photos with memorability scores provided by actual humans so it can learn to determine an image's memorability as well as humans can.
Think of it as an image version of the autocomplete function in search engines. When more people search for a certain topic, the algorithm takes note of it and when you try to search for the same thing, you already see it listed down as a suggestion. This time, though, the A.I. was fed with information about which part of photos humans deem the most memorable and the algorithm used these results to determine what would grab a human's attention when you upload a new photo to its system.
Just how good is this algorithm?
"We show that fine-tuned deep features outperform all other features by a large margin, reaching a rank correlation of 0.64, near human consistency (0.68)," the LaMem website explains. To put it simply: almost as good as humans and it's still learning.
Why did the team choose to study visual memorability? It's not for selfies. Their goal was to objectively understand how the human brain works in terms of memory and try to apply it to technology in the hopes that, from the result of their study, new devices and applications that can allow humans to consume and remember information more efficiently could be made.
For a more in-depth discussion, check out this article from Tech Times.
You can test out LaMem for yourself in its demo website or read the full study, "Understanding and Predicting Image Memorability at a Large Scale" by Khosla, et al. They also announced that LaMem is now available for download.
