AI scientists in China use self-learning to train a neural net to train itself
Scientists at China’s Sun Yat-Sen University and Hong Kong Polytechnic University have developed a way for a neural network (a type of AI that mimics how a human brain works) to self-correct its understanding of a video of human movements by comparing its guesses to the guesses of other networks.
Up until now, to correctly identify human poses, neural networks required lots of data — cues about how a knee or elbow bends, and 2-D and 3-D renderings. The problem is that there isn’t always enough labeled data to feed a network. The researchers decided to demonstrate that a network could fine-tune its approximation of human movements by continually comparing the guesses of multiple networks with one another, and in this way teach itself.
The team included Keze Wang, Liang Lin, Chenhan Jiang, Chen Qian, and Pengxu Wei. Qian works with the Chinese AI startup SenseTime.