Backpropagation Algorithm 2006 11 13 12222021 1 Perceptron

Backpropagation Algorithm 郝红侠 2006. 11. 13 12/22/2021 1

Perceptron : Single Layer Feedforward Rosenblatt’s Perceptron: a network of processing elements (PE): Input layer of source nodes 12/22/2021 Output layer of neurons 2

Perceptron : Multi Layer Feedforward • • • • Input layer Output layer • • Hidden Layer 12/22/2021 4

Learning Rule Measure error l Reduce that error l l 12/22/2021 By appropriately adjusting each of the weights in the network 8

BP Network Details l Forward Pass: l l l Error is calculated from outputs Used to update output weights Backward Pass: l l Error at hidden nodes is calculated by back propagating the error at the outputs through the new weights Hidden weights updated 12/22/2021 9

BP Algorithm l 正向过程: l 输出层所有神经元的误差能量总和 sum squared error, SSE l 权值修正: delta学习规则 12/22/2021 10

Case 2: 隐层权值修正 nj aj δj 12/22/2021 wji ni wji i δi 12

THANK YOU! 12/22/2021 17