人工智能入门（四）：uncertainty&基于统计的学习

最新推荐文章于 2025-06-05 23:55:53 发布

转载最新推荐文章于 2025-06-05 23:55:53 发布 · 209 阅读

0 ·

CC 4.0 BY-SA版权

原文链接：http://www.cnblogs.com/yizhaoAI/p/9944780.html

文章标签：

#人工智能

本文深入探讨了概率图模型的基础概念，包括信念网络和马尔科夫网络，详细讲解了独立性、条件化和D-分离等原理。涵盖了变量消元、桶消除算法、消息传递、MAP和ML估计、贝叶斯分类器等内容。讨论了处理缺失变量的方法，如EM算法，并介绍了多种采样技术，如吉布斯采样和粒子滤波。最后，涉及了动态模型，如隐马尔可夫模型和卡尔曼滤波。

1.belief networks (indenpendence, collider，conditioning / marginalization，connection graph，independence in belief networks，D-separation，uncertain and unreliable evidence）Belief and Markov Networks

2.inference, general inference(variable elimination，bucket elimination algorithm), message passing idea(sum-product algorithm,`belief propagation' or `dynamic programming',max-product algorithm,loop-cut conditioning)

for singly connected graphs: sum-product, max-product;

for multiply connected graphs: loop-cut conditioning, bucket elimination;

3.MAP,ML,(KL Divergence),Naive Bayes Classier,Using a Beta prior

4.dealing with miss variables: Missing Completely at random (MCAR), Missing at random(MAR),Missing NOT at random (MNAR),Expectation Maximisation(EM algorithm)

5.sampling（univariate，rejection，multi-variate,ancestral, Gibbs, importance, sequential importance,particle filter）

6.dynamical models(HMM(filtering, smoothing,prediction),Viterbi, Kalman, particle Filtering (bootstrap filtering)

转载于:https://www.cnblogs.com/yizhaoAI/p/9944780.html