Machine Learning

Faisal Qureshi

Lesson Plan

  • What is Machine Learning (ML)
  • ML tasks
  • ML and deep learning

Figure from the Scientist

What Is Machine Learning?

  • Mitchell, 1997:

    A computer program is said to learn from experience $E$ with respect to some class of tasks $T$ and performance measure $P$ , if its performance at tasks in $T$ , as measured by $P$ , improves with experience $E$.

  • "Subfield of artificial intelligence"
  • More instructive: Machine learning is a means of building mathematical models of data.

Supervised vs.Unsupervised Learning

Supervised learning: Modeling relationships between measured data features and prescribed, associated targets (or labels)

  • "Given input $A$, predict target $B$"
  • Entails regression (continuous labels/targets) & classification (discrete targets/labels)
  • Using established data measurements to infer targets/labels in as-of-yet unseen data
  • Prediction accuracy can be evaulated from prescribed data

Unsupervised learning: Models applied to data without any targets or labels

  • "Letting the dataset speak for itself"
  • Entails clustering, dimensionality reduction, and density estimation
  • "What patterns can we infer from $A$? Which features of $A$ occur together frequently?"

Supervised Learning

Regression problems

  • Inputs features $\mathbf{X}$ and generic (continuous) targets/labels

Desired output

Classification problems

  • Inputs features $\mathbf{X}$ and categorical/discrete targets/labels

Desired output

Unsupervised Learning

Clustering problems

  • No input/output pairs, just raw features $\mathbf{X}$

Desired output

Other unsupervised learning problems

  • Dimensionality reduction
  • Fitting distributions
  • Recommender systems
  • Association analysis

Machine learning --> deep learning

  • model --> network, graphs
  • parameters --> weights
  • fitting --> learning
  • test set performance --> generalization
  • regression, classification --> supervised learning
  • density estimation, dimensionality reduction, clustering --> unsupervised learning
  • large grant ~50,000 dollars --> large grant ~1,000,000 dollars



In [ ]: