The MIT team combined a data learning algorithm with a deep neural network that uses e.g.
To make artificial intelligence systems resilient to inconsistent data, researchers have tried to implement supervised learning defenses.
Traditionally, a neural network learns to connectspecific labels or actions with given inputs. For example, a neural network that receives thousands of images tagged as cats, along with images tagged as houses and hot dogs, should correctly label the new image as a cat.
In robust artificial intelligence systems, thosethe same supervised learning methods can be tested with partially modified versions of the image. If the net hits the same mark - a cat - there is a high chance that the image and changes or not is a cat.
To use neural networks in criticalfor security scenarios, we had to figure out how to make real-time decisions based on worst-case assumptions, the paper's authors explain.
Therefore, the team sought to rely on one morea form of machine learning that does not require the binding of labeled inputs to outputs, but rather aims at enhancing certain actions in response to inputs. This approach is commonly used to teach computers to play chess and Go.
The authors believe the new CARRL algorithm could help robots safely deal with unpredictable interactions in the real world.
Read more
Physicists have created an analogue of a black hole and confirmed Hawking's theory. Where it leads?
Algorithm has discovered a new mysterious layer inside the Earth
Because of the Sun, the Earth's atmosphere will lose all free oxygen