机器学习基石 - Three Learning Principles
机器学习基石下 (Machine Learning Foundations)—Mathematical Foundations
Hsuan-Tien Lin, 林轩田，副教授 (Associate Professor)，资讯工程学系 (Computer Science and Information Engineering)
Occam’s Razor
- entities must not be multiplied beyond necessity
- trim down unnecessary explanation
- simple: small hypothesis/model complexity
- direct action: linear first; always ask whether data over-modeled
- simple is good
Sampling Bias
- If the data is sampled in a biased way, learning will produce a similarly biased outcome.
- match test scenario as much as possible 测试的环境和训练的环境尽可能接近
Data Snooping
- your brain’s ‘model complexity’
- data snooping by shift-scale all values
- later author snooped data by reading earlier papers