From Learning to Machine Learning
Learning:
Acquiring skills with experience accumulated from observations
observation -> learning -> skill
Machine Learning
Acquiring skills with experience accumulated/computed from data
computers simply simulate the learning process of human beings.
data -> ML -> skill
So what is skill?
Skill: Improve some performance measure
data -> ML -> performance measure
An application in computational finance
stock data -> ML -> more investment gain
Another application: Tree Recognition
- Define trees and hard-program is Difficult
- Learn from data(observations) and recognize is Easy
- “ML-based recognition system” can be easier to build than hand-programmed(rule-based) system
ML is an alternative route to build complicated systems.
Use scenarios
- when human cannot program the system manually(navigating on mars)
- when human cannot “define the solution” easily(speech/visual recognition)
- when needing rapid decisions that humans cannot do(HFT)
- when needing to be user-oriented in a massive scale(consumer-targetd marketing)
Machine Learning(redefinition)
improve some performance measure with experience computed from data
data -> ML -> imporved performance measure
Feasibility
- exists some underlying pattern to be learned(performance measure can be improved)
- no programmable(no easy) definition(ML is needed)
- there is data about the pattern(resources to learn)