How ANNs Learn？

ANNs typically start out with randomized weights for all their neurons. This means that they don't "know" anything and must be trained to solve the particular problem for which they are intended. Broadly speaking, backpropagation is one of the methods for training an ANN, depending on the problem it must solve.

A back-propagation ANN is trained by humans to perform specific tasks. During the training period, the teacher evaluates whether the ANN's output is correct. If it is, the neural weightings that produced that output are reinforced; if the output is incorrect, the weightings responsible are diminished. This process is most often used for cognitive research and for problem-solving applications.

AI in Zhongguancun

2012 was a big year for Baidu’s voice and image products, especially its voice recognition systems, which grew markedly more accurate. In early 2013, Li Yanhong, CEO of Baidu, announced the launch of the Institute of Deep Learning (IDL). The IDL team focuses on image recognition, machine learning, robotics, human-computer interaction, 3D vision and heterogeneous computing. IDL hopes to become one of the top research institutions around the world, like AT&T-Bell labs and Xerox PARC.