Neuroscience and IA : what can IA learn from how the brain computes ?

There have been incredible advances in AI systems over the past few years, with deep learning-trained AI systems rivalling and even outperforming humans on various challenging tasks, including image and video analysis, speech processing and text generation. Such systems can avoid the temporal constraints imposed by the brain’s neural hardware, allowing them to perform tasks much faster than humans. But some of the brain’s key computational tricks are still missing from state-of-the-art AI. In this talk, Simon Thorpe discuss two particularly interesting features. First, he argue that using ultra-sparse spike-based coding schemes could be critical for explaining why the brain only needs 20 Watts of power. Second, he propose that the brain uses very efficient learning mechanisms that allow neurons to become selective to repeating patterns of activity in just a few presentations. Such features could allow the development of new types of brain-inspired AI systems that could be even more powerful than the start of the art.