机器学习基石笔记 Lecture 3 - Types of Learning

Lecture 3 - Types of Learning

---

Learning with Different Output Space $Y$

binary classification

core and important problem with many tools as building block of other tools

Multiclass Classification

many applications in practice,especially for ‘recognition’

Regression

also core and important with many ‘statistical’tools as building block of other tools

Structured Learning

a fancy but complicated learning problem
可以看作大规模的多分类问题，但是没有明确的类定义

Learning with Different Data Label $y_n$

Supervised

every $x_n$ comes with corresponding $y_n$

Unsupervised

Learning without $y_n$
unsupervised multiclass classification $\Longleftrightarrow$‘clustering’

Semi-supervised

semi-supervised learning: leverage unlabeled data to avoid ‘expensive’ labeling
由于标记成本比较高，或者说根本就没有这么多标记

Reinforcement Learning

a ‘very different’ but natural way of learning reinforcement: learn with ‘partial/implicit information’ (often sequentially)
训练机器，好比训练一条狗，哈哈，好好玩

增强学习我了解的太少了，具体怎么反馈的？？

Learning with Different Protocol $f \Rightarrow (x_n,y_n)$

Batch Learning

a very common protocol，learn from all known data

Online

最开始一点数据也不要

Active Learning

Learning by ‘Asking’，相当于我们高中自习的时候，有问题问老师
improve hypothesis with fewer labels (hopefully) by asking questions strategically

A photographer has 100, 000 pictures, each containing one baseball
player. He wants to automatically categorize the pictures by its player inside. He starts by categorizing 1, 000 pictures by himself, and then writes an algorithm that tries to categorize the other pictures if it is ‘confident’ on the category while pausing for (& learning from) human input if not. What protocol best describes the nature of the algorithm?

Learning with Different Input Space $X$

对人来说，越抽象的特征，越难理解，对于机器来说，也是越难学习

concrete features

each dimension of $X \subseteq R^d$ represents ‘sophisticated physical meaning’,the ‘easy’ ones for ML
More on Concrete Features:

Raw Features

image pixels, speech signal, etc.often need human or machines
to convert to concrete ones

Abstract Features

again need ‘feature conversion/extraction/construction’

fun time