Mostly good on content, now I just want to refresh on potential “derivation” questions:
Samples to prepare for:
- Optimal OLS weights
- GD weight updates
- Deriving cback propagation step using chain rule
- Kmeans clustering updates
- Hierarchical clustering grouping
- PCA by hand,eigenvalues/vectors
- Statistical significance of a clustering result
- Multiple hypothesis testing correction
- Derive MLE for linear regression
- How would you code gradient descent, provide pseudocode
- Regression and Optimization
- OLS
- SGD
- Logistic Regression updates
- Ridge/Lasso?
- ANN
- Calculate FFNN forward pass
- Calculate error signal
- Backprop
- Naive Bayes
- Posterior calculations
- Discriminant func?
- Unsupervised
- PCA
- K-means
- Hiearchical
- DBSCAN??
- Autoencoders
- Autoencoders, define components
- VAE
- Trees and significance
- Calculate impurity (Gini index or entropy)
- Statistical significance
- GMM & Modeling
- GMM model definitions, objective function, EM algorithm, and updates
- analyze Biclustering results
- Deep learning
- vanishing gradient
- weight explosion for images (MNIST) problem
- fix by using convolution
- Evaluation
- Regression: Predicted vs actual, residuals, etc
- ROC vs Precision recall
- False discovery rate tests
Backprop Proof:
- Activation of unit i - Weight matrix from layer j to j+1 - Activation function(ReLU) - Shorthand for the matrix mult of previous layer activations and weight matrix