Learning From Data

• Basics of Data Science

  • When is the mean useful?
  • Correlation---intuition and problems

• Cost

  • Functional Estimation as Cost Minimization
  • What's the cost?
  • How to minimize costs
  • Derivation of simple linear regression

• Error, Bias, Variance

  • Decomposing Error in Bias, Variance, Irreducible Error
  • Practical Trade-offs between Bias and Variance
  • Regularization, Dropout, etc.

• Evaluating Models

  • Metrics for evaluating models
  • Clearing up confusion about confusion matrices
  • How to construct your test data

• What Data to Collect?

  • Active learning
  • Generating types

• Causal Inference

  • Causal Inference
  • Experimental Inference
  • Power

• Computer Vision

  • Detection, segmentation, and evaluation

• Fair ML

  • Concerns
  • Solutions

• Interpretable ML

  • Why?
  • How?
  • Concerns

• Problem Solving With Data

  • Why?
  • How?
  • Concerns

• Model Testing

  • Why cross-validation isn't enough
  • Lessons from SWE

Author

Gaurav Sood

Build and check

Install R, Pandoc, and the R packages rmarkdown, revealjs, and knitr:

install.packages(c("rmarkdown", "revealjs", "knitr"))

Run make check to verify the numerical identities and execute every R lesson chunk. Run make slides to regenerate the HTML presentations and the Markdown copy of the first lesson from the R Markdown sources. Open an HTML file in a browser to present it; equations use MathJax and need an internet connection.