CS 450 · Year 4 · Semester 1 · 3 credits · Computer Science

High-Performance & GPU Computing for Simulation

Transform overnight actuarial batches into sub-second GPU kernels: mastering parallel patterns, JAX vectorisation, and nested stochastic architectures.

Full practice exam · 30 Q · 120 min

Season 1 · 8 episodes

Silicon Solvency: The 72-Hour Valuation Crisis

When a sudden market shock forces Apex Life to produce a full nested stochastic solvency valuation on a forty-billion-dollar variable annuity book in three days, lead computational actuary Maya Chen must rebuild their legacy simulation engine on modern parallel hardware before regulators intervene.

Protagonist · Maya Chen, Lead Quantitative Actuary and High-Performance Simulation Architect at Apex Life & Annuity.

Setting · Apex Life & Annuity Headquarters, Actuarial Computing Infrastructure Lab, Chicago.

Stakes · Apex faces a mandatory 1.2 billion dollar capital add-on and regulatory supervisory sanction under VM-21 if the Solvency Capital Requirement cannot be computed accurately within statutory filing windows.

Pick an episode

Episodes

Season finale: Maya successfully executes a fully deterministic, bit-level reproducible, GPU-accelerated nested valuation of two million variable annuity contracts across ten thousand economic paths in twenty-eight minutes, slashing cloud spend by ninety-one percent and proving full solvency compliance under ASOP 56.

From the lecturer: Welcome to CS 450. In modern actuarial enterprise systems—whether computing Solvency II 1-in-200 balance sheet shocks, IFRS 17 dynamic hedging reserves, or million-path catastrophe simulations—computational bottlenecks directly constrain risk visibility. This course bridges mathematical finance and modern compute hardware. You should watch these lectures with a split screen: your terminal and Jupyter environment on one side, and pen and paper on the other. Use your notepad to sketch execution Directed Acyclic Graphs (DAGs), memory bandwidth hierarchies, and thread block layouts before writing a single line of code. While CS 450 does not align to a single preliminary exam paper, it directly empowers the computational infrastructure demanded by SOA and CAS research tracks, predictive modeling competitions, and senior actuarial engineering roles in industry. Treat every millisecond and byte transferred across the PCIe bus as a concrete economic cost. We will move rigorously from Amdahl's law and CPU multiprocessing to GPU hardware warps, JAX tensor compilation, nested stochastic valuation, and deterministic parallel pseudorandom generation.

Cold open

At 6:15 AM on the first Tuesday of quarterly reporting, chief risk officer Marcus Vance barges into Maya's lab with an emergency directive from the state insurance commissioner: equity volatility spiked forty percent overnight, invalidating proxy reserves on Apex's forty-billion-dollar variable annuity block. The legacy multi-threaded C++ engine estimates seventy-two hours to compute the one-in-two-hundred-year tail capital, but the regulatory filing deadline expires in forty-eight.

preparing lecture…

Transcript

Classify actuarial computational workloads into canonical algorithmic parallel patterns: Map, Reduce, Scan, Stencil, and Gather/Scatter.
Formulate and differentiate strong scaling (Amdahl's Law) and weak scaling (Gustafson's Law) for valuation engines.
Identify memory access bottlenecks, coalescing opportunities, and execution hazards across multi-core CPU and SIMD/GPU architectures.