Abstract
- Decision trees
- Linear/logistic regression
- Kernels and Support Vector Machines (SVM)
Textbook, Chapter 19
Problem Space
Classical ML is motivated by problems that are intractable to solve generally, meaning that there is no efficient solution for all cases.
Intractable
Roughly speaking, a problem is called intractable if the time required to solve instances of the problem grows exponentially with the size of the instances. (from textbook)
Go, protein folding
Some problems are also difficult as formulating the rules in a way that a computer can understand and process may be challenging.
Image recognition
Thus, the problems focused are:
- problems where function is difficult to specify
- solutions are intractable to compute in general
- typically episodic and sequential
Paradigm
Rather than solving the problem explicitly through a search or applying rules, classical machine learning focuses on learning a function which could identify patterns in the data and makes decisions or provides solutions based on what it has learned.
Machine Learning
Machine learning (ML)
Ability to learn without being explictly programmed
Types of machine learning:
- Supervised learning: learns from labeled data (input to output), to learn a mapping from inputs to outputs
- Unsupervised learning: learns from unlabeled data (input), and aims to find patterns or structure.
- Semi-supervised learning: both labeled and unlabeled data
- Reinforcement learning: learns to make decisions by interacting with an environment, receiving rewards or penalties