A team from MIT-IBM has developed a faster way to train video recognition AI, which could make it easier to run machine learning on mobile devices.
Machine learning has given computers the ability to do things likeidentify facesandread medical scans. But when it's tasked with interpreting videos andreal-world events, the models that make machine learning possible become large and cumbersome. A team from the MIT-IBM Watson Lab believe they have a solution. They've come up with a method that reduces the size of video-recognition models, speeds up training and could improve performance on mobile devices.
The trick is in shifting how video recognition models view time. Current models encode the passage of time in a sequence of images, which creates bigger, computationally-intensive models. The MIT-IBM researchers designed atemporal shift module, which gives the model a sense of time passing without explicitly representing it. In tests, the method was able to train the deep-learning, video recognition AI three times faster than existing methods.
The link for this article located at Engadget is no longer available.