CS 301 · Year 3 · Semester 1 · 4 credits · AI & Machine Learning

Machine Learning

Statistical rigour meets algorithmic power: from loss surface gradients to calibrated, explainable insurance intelligence.

Full practice exam · 30 Q · 120 min

Season 1 · 8 episodes

The Black Box Margin

When an aggressive algorithmic rival cherry-picks Northstar Casualty's prime commercial auto fleets and spikes their loss ratio to 118 percent, lead predictive pricing actuary Marcus Vance must rebuild the carrier's entire underwriting engine from first principles—optimisation, boosting, neural nets, and interpretability—before the state insurance commissioner shuts down the filing and forces a solvency downgrade.

Protagonist · Marcus Vance, ACAS, Lead Predictive Modeler at Northstar Commercial Casualty

Setting · Northstar Commercial Casualty headquarters in downtown Chicago, during an intense six-month filing review under the Illinois Department of Insurance.

Stakes · If Marcus fails to deliver an actuarially sound, calibrated, and interpretable machine learning pricing engine, Northstar faces a two-notch AM Best downgrade, a 42 million dollar capital impairment, and regulatory sanction under ASOP 12 and ASOP 56.

Pick an episode

Episodes

Season finale: Marcus defends a fully audited, monotonic Combined Actuarial Neural Network and TreeSHAP pipeline before the Insurance Commissioner, proving mathematical compliance and driving Northstar's commercial combined ratio from 118 percent down to 94.2 percent.

From the lecturer: Welcome to CS 301. To master machine learning as an actuary or AI specialist, passive watching will not suffice. Keep an engineering notebook and an active Python environment open beside you. On paper, derive the gradients, loss functions, and dual formulations by hand before we write a single line of code. In your terminal, verify vector dimensions and implement core algorithms from scratch using pure NumPy before reaching for scikit-learn or XGBoost. This course bridges statistical learning theory with rigorous actuarial deployment, directly preparing you for SOA Exam PA (Predictive Analytics) and CAS Exam MAS-II (Modern Actuarial Statistics II). We will scrutinize the bias-variance tradeoff not just as an abstract curve, but as the mathematical boundary between underfitting pure premiums and overfitting catastrophe tails. Pay intense attention to probability calibration, deviance metrics, and SHAP interpretability—in insurance, a black-box model without actuarial justification and compliance with ASOP No. 56 is unratable. Treat every optimization step as a decision under uncertainty.

Cold open

Marcus Vance stares at the Q3 commercial fleet loss run on his monitor. Red cells glow across sixty pages: three regional trucking fleets defected to an algorithmic startup last week, while Northstar's residual book just posted twenty-four severe liability claims. Chief Actuary Diane Sterling walks into his office, shuts the door, and sets a regulatory warning letter on his desk. The company has exactly six months to retrain its pricing engine or face an AM Best rating downgrade.

preparing lecture…

Transcript

Formulate machine learning parameter estimation as empirical risk minimisation with regularisation
Derive the first-order gradient descent update rule via Taylor series expansion and analyse convergence under Lipschitz smoothness
Contrast batch, stochastic, and mini-batch gradient descent in terms of computational complexity and variance
Connect first-order gradient methods to second-order Newton-Raphson and Iteratively Reweighted Least Squares used in actuarial GLMs