NN AND ML——MLP

ADALINE Adaptive Linear classification machine 学习机
MLP:

1. Introduction

1. chap4.2 through chap4.7 discuss BP algorithm
   
   1. chap4.2 the derivation of the BP algorithm and credit-assignment.
   2. chap4.3 two kinds of methods of learning: Batch and on-line.
   3. chap4.4 the derivation of BP algorithm
   4. chap4.5 solve the XOP problem by BP algorithm.
   5. chap 4.6 heuristics and practical guidelines for improving BP algorithm performance.
   6. chap4.7 experiment on the multi-layer perceptron trained with BP algorithm
2. chap4.8 through 4.10 include:
   
   1. chap4.8 the error surface(误差曲面)
   2. chap4.9 computational issues relating to the Hessian of the error surface(误差曲面的Hession 矩阵的计算问题).
   3. chap4.10 How to fulfill optimal annealing and how to make the learning-rate parameter adaptive.(如何实现最优退火以及如何使得学习率参数自适应)
3. chap4.11-4.14 include: various matters relating to the performance of a MLP with BP.
   
   1. chap 4.11 the issue of generalization(泛化问题)——the essence of learning.
   2. chap 4.12 the approximation of continuous function by mean of MLP(通过多层感知实现连续函数的逼近问题)
   3. chap 4.13 Cross Validation(交叉验证)
   4. chap 4.14 The issue of complexity regularization and network-pruning techniques.
4. chap 4.15 The summarizes the advantages and limitations of lack-propagation learning. (反向传播学习的优点和局限性)
5. chap 4.16 Viewing supervised learning as an optimization problem.(从不同角度将监督学习看作最优化问题进行讨论)
6. chap 4.17 Convolutional Mutilayer Perceptron(cmp) (在解困难识别模式时得到成功的应用)
7. chap 4.18 Nonlinear filtering
8. chap 4.19 Small-scale versus large-scale learning problem
9. chap 4.20 Conclusion