基于对Bishop那本书的崇拜,去此牛人的主页瞻仰了一下,下了几篇文章
一篇是A New Framework for Machine Learning。其主要内容可以用introduction里的一句话来概括:
The new framework for machine learning is built upon three key ideas:
(i) the adoption of a Bayesian viewpoint,
(ii) the use of probabilistic graphical models,
(iii) the application of fast, deterministic inference algorithms.
这里的fast, deterministic inference algorithms指的是message passing。
受该结论的统治,其他几篇的特色也就很清楚。。。
基本思路:将所面对的问题在bayesian框架下刻画出图模型,然后进行inference
如:
“Bayesian Hierarchical Mixtures of Experts”
“Robust Bayesian Mixture Modelling”。
“Principled Hybrids of Generative and Discriminative Models”。
如此看来graphic model 与Variational Inference实在是有必要好好研究一下