NEC has developed new technology to help facilitate deep learning in its machine vision technologies, the company has announced.
The technology revolves around ‘regularization’, a concept concerning something like focus in artificial intelligence. As NEC explains, a deep learning system can become “excessively familiar” with data it is trained on, according to a statement; and this can result in the system being unable to recognize unfamiliar data. This unfortunate situation is one of “overtraining,” and can adversely affect accuracy in machine vision.
NEC’s solution is to regulate learning in order to prevent this from happening. As NEC Data Science Research Laboratories General Manager Akio Yamada explains, “This technology predicts the progress of learning at every layer based on the structure of an artificial neural network, and enables regularization to be automatically configured accordingly.” Yamada says this can reduce recognition errors by 20 percent compared to more conventional approaches.
And that, in turn, should lead to even more effective solutions for things like facial recognition, for which NEC is already well regarded. With these and other AI recognition technologies like behavioral biometrics seeing a growing range of applications, NEC’s deep learning upgrade could have a substantial impact on the operation of machine vision in the real world.
December 12, 2017 – by Alex Perala