quantum_noise_denoiser_v1

Overview

This model is a specialized Transformer designed to identify and suppress stochastic noise in superconducting quantum circuits. It processes raw pulse sequences or qubit telemetry to classify error types and provide a denoised representation of the underlying quantum state.

Model Architecture

The architecture is based on a Dense-Encoder Transformer optimized for 1D signal sequences:

• Encoding Layers: 6 Multi-head Self-Attention blocks.
• Dimensionality: Reduced 512-hidden size for low-latency inference.
• Objective: Minimization of the Mean Squared Error (MSE) between noisy input signals and pure state theoretical values: $$MSE=\frac{1}{n}\sum _{i=1}^n(Y_i-{\widehat{Y}}_i{)}^{2}MSE\; =\; \backslash frac\{1\}\{n\}\; \backslash sum\_\{i=1\}^\{n\}\; (Y\_i\; -\; \backslash hat\{Y\}\_i)^2$$

Intended Use

• Real-time Error Mitigation: Integrated into quantum controllers to correct bit-flip and phase-flip errors during gate execution.
• Telemetry Analysis: Post-hoc processing of quantum experiment logs to improve readout fidelity.
• Calibration: Assisting in the automated tuning of microwave pulse shapes.

Limitations

• Hardware Specificity: Trained primarily on transmon qubit data; performance on ion traps or photonic systems may vary.
• High-Frequency Noise: Limited by the sampling rate of the input signal; noise frequencies above 2GHz may result in aliasing artifacts.
• State Complexity: Accuracy decreases as the depth of the quantum circuit (number of gates) increases significantly beyond the training horizon.