Quantum-Inspired Tensor Networks

Why standard models struggle with sparse, correlated logistics data.

MPS, DMRG, and tensor train decomposition for logistics applications. Key areas include: Matrix product states (MPS) as efficient representations of high-dimensional probability distributions; Density matrix renormalisation group (DMRG): White (1992) algorithm adapted for time-series data compression; Tensor train decomposition: reducing exponential parameter counts to linear scaling in feature dimension.

Demand sensing, anomaly detection, and multi-variate forecasting. Key areas include: Multi-SKU demand modelling: capturing correlations across thousands of products with sparse sales signals; Anomaly detection in sensor networks: tensor decomposition of IoT time-series from fleet telematics and warehouse systems; Multi-variate forecasting: jointly modelling weather, traffic, inventory, and demand as entangled time series.

Building a tensor network model for logistics time-series data. Key areas include: Constructing an MPS model for a 200-SKU demand dataset using the ITensor library; Comparing compression quality and forecast accuracy against PCA, autoencoders, and LSTM baselines; Visualising bond dimensions and entanglement structure to identify which product correlations the model captures.

How tensor networks bridge quantum theory and deployable classical methods. Key areas include: Quantum circuit simulation via tensor contraction: why MPS methods originated in quantum many-body physics; Classical hardware deployment: tensor network models run on standard GPUs and CPUs without quantum hardware; When quantum hardware helps: variational tensor network training on quantum devices as an acceleration strategy for the fault-tolerant era.

Deploying tensor network models in logistics data pipelines.

Q&A and Action Planning: this session covers the core principles and technical underpinnings relevant to the subject area.