- non-parametric model
- set σn =0, we can fit the data exactly, the mean predictive function pass training data
- log marginal likehood:
- the first term: −12log(|Σ|) is a complexity penalty term, measure and penalizes the complexity of the model
- the second tern: a negative quadratic, plays the role of a data-fit measure, it is the only term depend on the train-output y <script type="math/tex" id="MathJax-Element-241">y</script>
- the third term is a log normalization term, not so interesting
Gaussian Process
最新推荐文章于 2024-04-20 10:11:04 发布