Workshop Description
Full-day workshop on QAOA for logistics combinatorial optimisation. Covers SKU clustering, demand segmentation, and calendar effect isolation as graph partitioning problems with honest NISQ benchmarks against classical competitors.
Many logistics decisions are discrete: which SKUs belong in the same replenishment cluster, which customers share ordering patterns, whether a demand spike is promotional or seasonal. These are graph partitioning and binary assignment problems where the Quantum Approximate Optimisation Algorithm (QAOA, Farhi et al. 2014) offers a structured quantum approach. QAOA uses alternating problem and mixer unitary layers with variational parameters optimised classically to find approximate solutions to combinatorial problems. This workshop teaches participants to formulate three logistics problems as MaxCut instances, implement QAOA circuits in Qiskit, and benchmark results against spectral clustering, k-means, and Goemans-Williamson relaxation. The workshop is direct about NISQ limitations. Qubit connectivity constraints force SWAP gate insertion that inflates circuit depth. Barren plateaus in the variational landscape make parameter optimisation difficult at moderate depths. For logistics-scale instances (hundreds of SKUs), classical methods currently match or exceed QAOA quality. The value of this workshop is understanding precisely what QAOA can do as hardware improves, so organisations can identify which of their discrete logistics problems will benefit first.
What participants cover
- QAOA circuit design: alternating problem/mixer unitaries, variational parameter optimisation, and circuit depth versus solution quality trade-offs (Farhi et al. 2014)
- SKU clustering: partitioning product portfolios by demand similarity, margin contribution, and replenishment frequency as weighted graph cuts
- Demand segmentation: identifying customer groups with distinct ordering patterns for differentiated service level policies
- Calendar effect isolation: separating promotional uplift from seasonal baseline using binary assignment formulations
- Classical benchmarking: comparing QAOA against spectral clustering, k-means, Goemans-Williamson, and simulated annealing on matched instances
- NISQ constraints: qubit connectivity overhead, barren plateaus in parameter optimisation, and scaling limits on current gate-based hardware