Machine Learning from Scratch: The Systems View (Junior Edition)

Apply linear algebra operations (matrix multiplication, decomposition, eigenvalues) to understand data transformations and dimensionality reduction in ML contexts

Compute gradients and partial derivatives to explain how optimization algorithms update model parameters

Use probability distributions and Bayes' rule to model uncertainty and interpret probabilistic predictions

Explain how loss functions quantify prediction error and guide the learning process for different ML tasks

Describe gradient descent optimization and its variants, including how learning rates and batch sizes affect convergence

Analyze the bias-variance tradeoff to diagnose underfitting and overfitting, and apply regularization strategies to improve generalization

Implement linear regression, logistic regression, and decision trees from mathematical principles, explaining how each algorithm makes predictions

Build and evaluate classification and regression models using scikit-learn, applying proper train-test splits and cross-validation

Select appropriate algorithms for different problem types based on data characteristics, interpretability needs, and performance requirements

Design and implement reproducible data preprocessing pipelines handling missing values, scaling, and encoding

Engineer informative features from raw data using domain knowledge and automated techniques

Detect and handle data quality issues (outliers, imbalanced classes, leakage) that compromise model validity

Explain how neural networks compose non-linear transformations through layers, and how backpropagation computes gradients for all parameters

Implement feedforward and convolutional neural networks in PyTorch, understanding data loaders, loss functions, and optimization loops

Diagnose and address common training challenges (vanishing gradients, overfitting, slow convergence) using appropriate techniques

Design reproducible ML experiments with proper versioning, logging, and documentation practices

Implement model evaluation strategies that detect overfitting, distribution shift, and performance degradation

Deploy a trained model as a containerized service with basic monitoring and understand the path toward production MLOps

Build an end-to-end ML pipeline solving a real-world problem, from data exploration through model deployment

Document design decisions, trade-offs, and evaluation methodology in a professional technical report

Present and defend modeling choices to a technical audience, demonstrating systematic thinking about ML systems