Communications of the ACM
Deep-Learning Algorithm Predicts Photos' Memorability at "near-Human" Levels
Massachusetts Institute of Technology Computer Science and Artificial Intelligence Laboratory researchers say they have developed an algorithm that can predict how memorable or forgettable an image is, almost as accurately as humans can.
Credit: Massachusetts Institute of Technology Computer Science and Artificial Intelligence Laboratory
Researchers at the Massachusetts Institute of Technology's Computer Science and Artificial Intelligence Laboratory (CSAIL) say they have developed an algorithm that can predict how memorable or forgettable a given image is almost as accurately as humans.
The MemNet algorithm builds on previous work by CSAIL researchers to develop a similar algorithm for facial memorability. The researchers fed the algorithm, running on a deep-learning network, tens of thousands of images from several different datasets that had been given a "memorability score" by human subjects. When they pitted MemNet against human subjects by asking both to predict how memorable people would find a never-before-seen image, the algorithm performed 30% better than existing algorithms and within a few percentage points of the average human performance.
The researchers plan to use MemNet to build an app that can tweak images to make them more memorable. "It's like having an instant focus group that tells you how likely it is that someone will remember a visual message," says lead author Aditya Khosla, who presented the research this week at the International Conference on Computer Vision (ICCV) in Santiago, Chile.
Other potential applications for the technology include developing more effective teaching resources and more compelling advertisements.
From MIT News
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found