ML Prep

Machine Learning Screening

Answer each question independently. Difficulty varies, and you can complete the list over several sittings.

Review results

Regression

Derive and interpret linear regression objectives and estimators.

  1. Squared loss and gradientEasy · 6 min · 3 skills
  2. Solve normal equationsMedium · 8 min · 3 skills
  3. One-dimensional ridge estimateHard · 10 min · 3 skills

Classification

Work with logistic models, losses, decision rules, and metrics.

  1. Logistic probability and lossEasy · 5 min · 4 skills
  2. Metrics from a confusion matrixMedium · 6 min · 3 skills
  3. Cost-sensitive thresholdHard · 10 min · 2 skills

Gradient Descent

Compute updates and reason about scale, learning rate, and convergence.

  1. Two gradient updatesEasy · 4 min · 2 skills
  2. Learning-rate stabilityMedium · 8 min · 3 skills
  3. SGD unbiasedness and varianceHard · 9 min · 2 skills

Optimization

Relate objectives, stationary points, convexity, and regularization.

  1. Stationary point and convexityEasy · 4 min · 3 skills
  2. Compare L1 and L2 solutionsMedium · 8 min · 4 skills
  3. Diagnose optimization tracesHard · 8 min · 3 skills

Neural Networks

Perform forward passes and backpropagate through small networks.

  1. ReLU forward passEasy · 5 min · 2 skills
  2. Backpropagate through tanhMedium · 7 min · 5 skills
  3. Parameters and output gradientHard · 10 min · 4 skills

Convolution

Compute convolution outputs, receptive geometry, and parameter counts.

  1. Valid convolution outputEasy · 5 min · 1 skills
  2. Convolution and pooling geometryMedium · 7 min · 5 skills
  3. Receptive field through a stackHard · 9 min · 3 skills

Dimensionality Reduction

Connect PCA objectives to covariance eigendecomposition and SVD.

  1. Center data and find PCA directionEasy · 6 min · 3 skills
  2. Connect SVD to PCAMedium · 8 min · 3 skills
  3. Optimal rank-two PCAHard · 7 min · 3 skills

Unsupervised Learning

Optimize and interpret clustering and latent-variable objectives.

  1. One k-means iterationEasy · 5 min · 2 skills
  2. GMM responsibilities and M-stepMedium · 9 min · 2 skills
  3. Compare k-means fixed pointsHard · 11 min · 2 skills

Theory

Reason about generalization, model complexity, evaluation, and bias–variance.

  1. Train, validation, and test rolesEasy · 5 min · 2 skills
  2. Compute bias and varianceMedium · 7 min · 2 skills
  3. Select model complexityHard · 8 min · 5 skills