quantum-feature-amplification

name

quantum-feature-amplification

description

Quantum Feature Amplification Network (QFAN) methodology for autoregressive quantum generative modeling. Decouples quantum register size from output dimension using fixed-size quantum circuits combined with classical autoregressive decoding. Use when designing scalable quantum generative models for high-dimensional data, quantum ML for scientific simulations, or hybrid quantum-classical generative architectures. arXiv: 2605.16044

Quantum Feature Amplification Network (QFAN)

Overview

QFAN is an autoregressive quantum generative model that solves the scaling bottleneck of direct-register quantum generative models, where output dimension is tied to quantum register size.

Core Insight

Traditional quantum generative models require the quantum register to scale with the full image/data dimension, making large-scale demonstrations infeasible. QFAN breaks this coupling by using:

1. A fixed-size quantum circuit that processes local feature patches
2. Classical autoregressive decoding that stitches patches into full outputs
3. Quantum-classical handoff at the feature level, not the pixel level

Architecture

Input → Fixed Quantum Circuit (small register) → Feature Patch → Autoregressive Classical Decoder → Full Output

Key Design Choices

• Quantum register size: Independent of output dimension
• Autoregressive structure: Each step conditions on previous outputs
• Hybrid quantum-classical: Quantum for feature extraction, classical for generation

Advantages

1. Scalability: Handles high-dimensional data without proportional qubit increase
2. Hardware-friendly: Works on near-term quantum devices (NISQ)
3. Scientific applications: Suitable for detector-scale geometries in high-energy physics
4. Generative quality: Autoregressive structure captures complex correlations

Application Domains

• Calorimeter shower simulation (high-energy physics)
• Medical image generation
• Molecular structure generation
• Scientific data synthesis

Implementation Pattern

# Conceptual architecture
class QFAN:
    def __init__(self, quantum_circuit_size, autoregressive_steps):
        self.quantum_circuit = FixedSizeQuantumCircuit(quantum_circuit_size)
        self.decoder = AutoregressiveDecoder(autoregressive_steps)

    def generate(self, latent):
        features = self.quantum_circuit.encode(latent)
        return self.decoder.autoregressive_decode(features)

Comparison with Alternatives

Approach	Register Scaling	Output Dimension	Hardware Demand
Direct-register	Linear with output	Limited	High
Latent-variable hybrid	Partial reduction	Medium	Medium
QFAN	Fixed	Unbounded	Low

Pitfalls

• Autoregressive decoding introduces sequential bottleneck
• Quality depends on feature patch size and overlap
• Quantum circuit still needs sufficient expressivity per patch

References

• arXiv: 2605.16044
• Related: [[qml-spiking-encoding]], [[quantum-neural-network-data-loading]]