Restricted Boltzmann machine architectures on quantum hardware: sampling, energy landscapes, and logistics applications.
Where classical probabilistic methods underperform under uncertainty.
From classical RBMs to quantum-enhanced sampling. Key areas include: Classical Restricted Boltzmann Machines (RBMs): energy-based models, contrastive divergence training, and sampling limitations; Quantum Boltzmann Machines (QBMs): transverse-field terms, quantum tunnelling for faster mixing, and the Amin et al. (2018) framework; Variational quantum thermalisation: preparing Gibbs states on gate-based hardware versus annealing approaches.
Promotional planning, new product launches, and demand shocks. Key areas include: Scenario generation for promotional planning: sampling from joint distributions over correlated product categories; New product launch modelling: generating demand trajectories with limited historical data using quantum-enhanced priors; Demand shock simulation: stress-testing supply chains under extreme weather, port disruptions, and pandemic scenarios.
Full-day format only. Key areas include: Facilitator-led walkthrough: training a small QBM on a logistics demand dataset and sampling scenarios; Comparing QBM-generated scenarios against classical RBM and VAE baselines for distributional fidelity; Discussion: identifying use cases from your organisation where classical generative models struggle with tail events.
Current capabilities and the fault-tolerant frontier. Key areas include: QBM on quantum annealers: using annealing hardware as a sampler for Boltzmann distributions and current qubit connectivity constraints; Gate-based QBMs: variational Gibbs state preparation on superconducting and trapped-ion hardware, circuit depth limitations; Quantum-inspired approaches: tensor network methods and classical Boltzmann sampling accelerators as near-term alternatives.
Inserting QBMs into existing planning systems. Key areas include: Architecture patterns: QBM as a scenario generator feeding into classical optimisation and simulation layers; Data requirements: what training data quality and volume QBMs need versus classical generative models; Decision criteria: when QBM exploration is justified based on problem dimensionality and distributional complexity.
Q&A and Pilot Planning: this session covers the core principles and technical underpinnings relevant to the subject area.
Discuss this topic with senior peers.
