Week 4 - Algorithms Flashcards
What is a machine learning algorithm
It’s an algorithm that solves the problem of “learning to solve a problem” without explicitly being told how
What is the goal of MLA?
To predict something correctly - task or classification
How does explicit programming compare to machine learning
Explicit programming comes up with rules/instructions that processes the input/data and produce outputs to solve the problem.
Machine learning algorithms find the best approximation of function F that maps inputs/data to outputs to solve the problem
What’s the main difference between regression and classification
Regression is purely for numerical calculations. No fixed bounds
Classification has a closed set. You must pre-define the number of classes and classify accordingly
When is ML useful?
When it is hard to come up with explicit representations and instructions
How does the ML paradigm function
It functions with enough input and output
What is the math behind ML?
because we are mapping inputs to outputs, we estimate function f.
All models are wrong but some are useful. Explain
They only provide an estimation and they cannot reflect all the possibilities in the world. The math is based on statistics and a key in statistics is the sample size. You are only representing a subset to estimate what the population looks like.
Statistical approaches for estimating f
Parametric methods
1. Assumption about form of f
2. Apply algorithm to estimate the parameters for f that best fits the model to the data
Non parametric
1. Make no explicit assumption about functional form of f
What is an example of a parameter
Numbers added to the calculation to transform the input (e.g., m and c)
Neural networks
Complex formula variations of y=mx+c
Architecture and layers definition
Architecture - the way you do calculations, determines what the operations are
Layers - one set of inputs
