Machine Learning Concepts Quiz

Machine Learning 1 - Week 1

• Course Introduction
  • Topics include course introduction and supervised machine learning.
  • The suggested textbook: An Introduction to Statistical Learning, is by Gareth James, Daniela Witten, Trevor Hastie, Robert Tibshirani, Jonathan Taylor
  • The optional textbook: Introduction to Machine Learning with Python, is by Andreas C. Müller & Sarah Guido
• Course Tour
  • The course includes syllabus, course site, course outline, and a data camp.

Supervised Machine Learning

• What is Supervised Machine Learning and Why Do We Use It?

  • Supervised machine learning aims to learn patterns from data
  • The goal is to predict an output given an input.

• Learning from Data

  • Classical programming discovers rules from data to produce output
  • Machine learning models create rules from data inputs and outputs.

• Learning from Data and Examples

  • Algorithms discover patterns, create models, and use iterative adjustments to refine models.
  • The output, or the learned model, signifies a decision boundary. Common algorithms include Decision Tree, SVM (Gaussian kernel), Neural Network, and Random Forest.

• Biologically Inspired (Neural Network)

  • Machine learning techniques can be inspired by biological models
  • Neural network architectures are biologically inspired systems, such as input, hidden-layer 1, hidden-layer 2, and output layers.

• Learning by Trial and Error

• Machine Learning Definition (Arthur Samuel 1959)

  • Machine learning allows computers to learn without explicit programming.

• Machine Learning Definition (Herb Simon 1978)

  • Machine learning focuses on improving performance through experience.

• Machine Learning Definition (Tom Mitchell 1997)

  • A computer program learns from experience with respect to some task T and some performance measure P, if its performance on T as measured by P improves with experience E.

• Task, Performance Measure, and Training Experience

  • Examples include task T = recognizing digits; performance measure P = accuracy in predicting the correct digit and training experience E = an input-output dataset containing digit samples.

• Artificial Intelligence and Turing Test

  • Alan Turing (1950) introduced the Imitation Game (Turing Test), an approach to determine whether a machine can think.

• AI and Notable Figures

  • The 1956 Dartmouth Workshop was a seminal event in AI.
  • Important figures include John McCarthy, Marvin Minsky, Claude Shannon, Ray Solomonoff, and Alan Newell, Herbert Simon, Arthur Samuel, Oliver Selfridge, Nathaniel Rochester, and Trenchard More.

• Artificial Intelligence & Games

  • Examples of AI in games include Arthur Samuel's checkers program (1959), Garry Kasparov vs. Deep Blue (1997), Ken Jennings vs. IBM Watson (2011), and AlphaGo vs. Lee Sedol (2016).

Why Now?

• More Data, Computational Power
• Progress on algorithms, theory, and tools.
• Accessible Computing

Categories of Machine Learning

• Supervised Learning

  • Use labeled data.
  • Predict outcomes/future outcomes using direct feedback.
  • Regression (continuous variable); classification(discrete variable)
  • Supervised learning algorithms learn from labeled data to predict outcomes. Example figures in this presentation are a graph for regression ( a straight line) and another graph for classification (a line dividing two clusters of datapoints).

• Unsupervised Learning

  • No labels or targets.
  • Example is clustering.

• Reinforcement Learning

  • Decision-process, reward system.
  • Learn series of actions.

Evaluating Models

• Evaluating Regression Models

  • Mean Squared Error (MSE): measure of the difference/error
  • Mean Absolute Error (MAE): measure of the magnitude/error

• Evaluating Classification Models

  • Accuracy = proportion of correct predictions.
  • Confusion Matrix identifies TP, TN, FP, FN for assessing the model (accuracy)

• ROC curve (Receiver Operating Curves)

  • Helps to evaluate the model. It plots True Positive Rate (TPR) against False Positive Rate (FPR)
  • Used to select the best threshold.

• Regression Model Overfitting - Using data to create more complex models, resulting in worse performance with testing examples.

• Bias-Variance Tradeoff

  • Simpler models: Underfitting
    • Complex models: Overfitting

• Structure of Training and Prediction

  • Data split into training (red points) and testing (blue points) sets.
  • Model creation and training on training set
  • Prediction using model to obtain results for testing set.
  • Assessment of accuracy on testing set