Machine Learning Flashcards
What is Machine Learning? (ML)
The study of computer algorithms that improve automatically through experience and the use of data. It’s part of AI
How does ML work?
Machine learning algorithms build a model based on sample data, known as “training data”, in order to make
predictions or decisions without being explicitly programmed to do so.
How do you formalize ML?
ML can be described as a function Y=H(X) where the goal is to find the most simple H which predicts Y using X as input for a given prediction accuracy
What do you call the performance of H in matching Y using X?
The Objective function
How do you find the objective function?
Obj(H) = L(H) + omega(H)
where L is the matching error
and Omega is the regularization term/complexity of H
What does ML consist of in terms of the objective function?
Minimizing the Obj(H) as the best potential compromise between prediction accuracy and complexity
What are the main categories of Machine Learning?
Supervised: classification & regression
Unsupervised: clustering, association & dimension reduction (generalization)
What is the difference between supervised and unsupervised ML?
Supervised: data is pre-categorized
Unsupervised: data is not labeled
What are the main ML application/tasks?
Forecasting and classification
What are the main categories of ML engines
-Linear/non-linear regressions
-Random forests and boosted trees
-Deep learning and neural networks
What is a linear regression?
You model the relationship between two variables Y and X where X explains Y such that:
Y= aX+b
where a=Cov(Y,X)/Var(X)
and B=E(Y)-aE(X)
(remember Y is what you want to predict and X is the explanatory variable)
What do you need for the regression to be complete?
The mean of the residue should be normally distributed with a mean of 0
What are the steps in training AI predictive models?
Building the model
Training the model on sample data
Testing the model on different sample data
What is one of the main challenges in training ML algorithms?
Avoiding overfitting so that it only works on the training data sample
How do you avoid overfitting?
You keep the model as simple as possible (few parameters)
