Advanced video recognition technology from MIT and IBM
The development of artificial intelligence -based technologies takes place through machine learning and deep learning . Even the simplest technological devices that we use today can use artificial intelligence. Therefore, devices that incorporate advanced artificial intelligence technology are now the only thing scientists and engineers want to develop. The MIT-IBM Watson Lab team wants to use advanced artificial intelligence in the field of video recognition technology .
As we mentioned before, it is seen in many areas where we use or monitor artificial intelligence. The MIT-IBM Watson Lab team is working on an algorithm that works quickly and accurately, even on mobile devices. If this work proceeds as intended, machine learning will also make it easier to operate video recognition technology on mobile devices.
Thanks to machine learning, artificial intelligence can identify faces on computers and show itself in many areas such as scanning traffic. But when it comes to real-world photos and videos, artificial intelligence can be confused.
For this reason, the team thinks that they can solve real-world problems thanks to their newly developed method. The team adds a new dimension to video identification technologies, accelerates training and improves performance on mobile devices.
When we look at the working principle of the model, it is seen that the video recognition models have changed the way we look at time. The method, which is already designed as the temporal shift module, saves time passing from larger coding lines, unlike existing methods. The algorithm trained in this context can become three times faster than the existing methods.
Normally, it was not easy for video recognition models to work on mobile devices. Thanks to the mentioned exchange module, video recognition models will also work on mobile devices. Song Han, a member of the team, says that their goal is to design a model that needs less energy and can run smoothly to make artificial intelligence accessible to everyone on low-powered devices.
This method, which will facilitate the detection of violent or harassment videos spread on platforms such as Facebook and YouTube, can also protect the security of local sensitive data, not hospital data in the cloud. The model will be presented at an international conference at the end of this month.