Quiz #1 Topics and Structure

Structure

60 minutes to complete the quiz
20 multiple choice questions (or true/false) [4 points each]
Two (2) short answer questions [30 and 10 points]
DO NOT leave any questions blank.

Topics

AI and ML
- Artificial intelligence (overall field/science)
- ML: specific approach to AI (machine learning)
Types of ML
- Supervised
  - Labels ("answers") are given
- Unsupervised
  - NO labels
- Reinforcement
  - feedback from environment
Supervised Learning
- Classification
  - Discrete/fixed set of target variable categories (e.g., dog/not dog)
  - Ex: kNN, decision tree, random forest, neural net
- Regression
  - Continuous/infinite possible values for target variable (e.g., house prices)
  - linear regression
  - logistic regression--almost always used for binary classification
Unsupervised Learning
- Clustering
  - k-means
    - hyperparameter 'k' is the number of clusters you want to algo to find
    - starts with random positions for each 'centroid' (i.e., the center of each cluster); then moves them over time to optimize actual cluster centers
Hyperparameters vs. Parameters
- Parameters: variables inside the model that must be trained/optimized; trained/set by training algorithm
- Hyperparameters: configuration of training algos; set by us
Metrics
- Accuracy -- # correct / # total samples
- Confusion matrix -- grid of correct vs predicted classifications
Generalization (overfitting, underfitting, bias and variance)
- Generalization: how well does the model extend to unseen examples (things not in the training set)
- Overfit: 'memorizes' training set; high variance, low bias
- Underfit: does not capture data complexity enough; high bias, low variance
- training vs. test accuracy
  - - training: accuracy from the training data
    - test: accuracy from test data
  - High training accuracy, low test accuracy --> overfit / variance too high
  - Low training accuracy, low or high test accuracy --> underfit / bias too high