VQC architectures for probabilistic demand and supply forecasting in logistics operations.
Why logistics needs probabilistic demand distributions, not single numbers.
Parameterised circuits as probabilistic distribution generators. Key areas include: VQC structure: data encoding layers, parameterised rotation gates, entangling layers, and measurement-based output; Born machine interpretation: VQC measurement statistics as probability distributions over demand outcomes; Expressibility versus trainability: how circuit ansatz design affects the distributions a VQC can represent (Sim et al. 2019).
SKU-level, channel-level, and daily granularity probabilistic models. Key areas include: SKU-level demand distributions: generating full probability distributions for safety stock and service level calculations; Channel-level forecasting: capturing online versus offline demand patterns and their cross-channel correlations; Daily granularity: modelling intra-week and intra-month demand variation for workforce and delivery planning.
Training a VQC for probabilistic demand forecasting. Key areas include: Building a 10-qubit VQC using PennyLane with hardware-efficient ansatz for a 50-SKU demand dataset; Training via MMD (maximum mean discrepancy) loss to match empirical demand distributions; Comparing output quality against quantile regression, DeepAR, and Gaussian process baselines on CRPS metric.
Barren plateaus, noise, and what limits VQC forecasting today. Key areas include: Barren plateau problem: Cerezo et al. (2021) results showing gradient vanishing in deep parameterised circuits and practical mitigation strategies; Noise impact on output distributions: how NISQ device errors corrupt probability estimates and when error mitigation helps; Classical alternatives with equivalent expressibility: normalising flows and VAEs that produce probabilistic forecasts without quantum hardware.
Connecting VQC research to production forecasting pipelines.
Q&A and Action Planning: this session covers the core principles and technical underpinnings relevant to the subject area.
Discuss this topic with senior peers.
