Loading Classical Data with Low-Depth Circuits

Loading Classical Data with Low-Depth Circuits

This PennyLane demo shows how to load classical image data into quantum states using shallow circuits, following the paper 'Typical Machine Learning Datasets as Low-Depth Quantum Circuits' by Kiwit and collaborators. Encoding arbitrary data is generally expensive, but real-world images carry structure that can be exploited. The tutorial first defines how an image with a power-of-two number of pixels is mapped to a quantum state whose address register holds pixel positions and whose extra color qubits encode intensities, using the flexible representation of quantum images (FRQI) for grayscale and the multi-channel representation (MCRQI) for color. It then explains why such image states are only weakly entangled and are therefore well approximated by matrix-product states, so that instead of the exponential gate count required for exact preparation, an approximate circuit inspired by an MPS in mixed-canonical form prepares them with a depth scaling only linearly in the number of qubits. Finally the demo trains and evaluates a small variational quantum classifier on the encoded dataset. It matters because efficient, low-depth data loading is a central bottleneck for near-term quantum machine learning, and this structure-aware approach makes encoding realistic datasets substantially more practical on hardware.

Run it

pip install -r requirements.txt
python demo.py

Source and license

Imported from demonstrations_v2/low_depth_circuits_mnist/demo.py in PennyLaneAI/demos at c52c0abeb5122218aa96b38eea848864cce7323f, under the Apache License 2.0. Original authors: Xanadu and the PennyLane community. The upstream LICENSE is included alongside this example.