Forewords
I am recently starting the “Introduction to Machine Learning” course at college. Therefore, I decide to start a new series on the topic of machine learning, in which I will note down important take aways from ML.
All about distributions
The essence of machine learning lies in statistics and optimization (an argument borrowed from my professor). The datasets that we look at have some inherent patterns or distributions, while we create probabilistic models to fit these inherent patterns. We then use optimization tools to actually do the fitting.
I know this summary can be abstract, so next let’s look at some examples together.
The examples
We first consider the classic example of linear regression:
Another example worth looking at is classification, which will be introduced in more details in the following logs.
For now, the most important takeaway is: Regression and classification are two of the most important and fundamental applications of machine learning.
Appendix: KL-Divergence and Cross-entropy
Remarks: It’s worth noting that KL-divergence and cross-entropy both characterize the difference between two distributions. This may sound familiar to you.
Spoiler: Cross-entropy can be used as an objective of optimization too! (Or you may also call it a loss function)
发表回复