It is natural for people to predict what someone will do based on their body language, but this is not the case for computers. When we meet another person, they may say hello, shake hands, or even clench their fists. We may not know which gesture to use, but we are able to read the situation and respond appropriately.
Researchers at Columbia have now introduced a computer vision technique that involves humans, animals and utilizing higher-level associations between objects gives machines a more intuitive sense of what is going to happen.
“Our algorithm is a step towards machines being better able to predict human behavior and thus better coordinate Our results open up many opportunities for human-robot collaboration, autonomous vehicles and assistive technologies, “said Carl Vondrick, study leader at the International Conference on Computer Vision and Pattern Recognition.
researchers say this is the most accurate way to predict video action events for up to several minutes. After analyzing thousands of hours of movies, sports games and TV shows, the system learns to predict hundreds of activities, from handshakes to punches. If you can’t predict the specific action, you’ll find a higher-level concept that connects them, such as the word “greeting.”
Previous attempts at predictive machine learning, including those made by the team, they focused only on predicting one action at a time. The algorithms decide whether the action is classified as a hug, a patchwork, a handshake, or even a non-action such as “ignoring.” However, when there is a high degree of uncertainty, most machine learning models are unable to find commonalities between the possible options.
Researchers from Columbia Didac Suris and Ruoshi Liu have decided to look at the longer-term perspective from a different perspective. forecasting problem. “Not everything in the future is predictable. When one cannot predict exactly what will happen, he goes for sure and predicts a higher level of abstraction. Our algorithm is the first to learn this ability to think about future events in an abstract way. “- said Suris, co – lead author of the study.
The mathematical framework developed by the researchers allows machines to organize events according to how predictable they are in the future. We know, for example, that both swimming and running are a form of exercise. The new technique automatically learns how to categorize these activities. The system is aware of the uncertainty and, if there is certainty, provides more specific actions and, if not, provides more general predictions.
Technology can bring computers closer to being able to assess instead of pre-programmed action. a situation and make a nuanced decision. This is a critical step in building trust between people and computers. Confidence stems from the feeling that the robot really understands man. If machines can understand and predict our behavior, computers will be able to help people seamlessly in their daily activities, “said Liu, co-author of the study.
Hardware, software, tests, curiosities, and color news from the world of IT by clicking here
