Reformulating DP as GPU kernels delivers 100- to 100,000-fold speedups for stochastic vehicle routing and inventory problems, enabling much larger scenario sets and stronger first-stage decisions.
We construct a transition cost matrixA ω ∈R 3×3 where the(p, i)entry represents the cost of serving customersσ p+1 toσ i in one route, if the cumulative demand is within capacity
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
math.OC 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
From Sequential to Parallel: Reformulating Dynamic Programming as GPU Kernels for Large-Scale Stochastic Combinatorial Optimization
Reformulating DP as GPU kernels delivers 100- to 100,000-fold speedups for stochastic vehicle routing and inventory problems, enabling much larger scenario sets and stronger first-stage decisions.