Quantum Noise Processing: Baidu's QEP Toolkit for Quantum Error Mitigation

This article presents Dr. Wang Kun's technical report from Baidu Technology Salon #103 on quantum noise processing. The report addresses the critical challenge of quantum noise in the current NISQ (Noisy Intermediate-Scale Quantum) era, where quantum computers with 50-100 physical qubits face significant accuracy limitations due to noise.

The content introduces Baidu's Quantum Error Processing Toolkit (QEP), China's first open-source quantum noise processing toolkit. QEP provides three core functionalities: benchmark testing, noise characterization, and noise mitigation. Benchmark testing efficiently evaluates quantum computer performance and scalability through methods like randomized benchmarking. Noise characterization comprehensively profiles quantum noise details including quantum state tomography, quantum process tomography, and quantum detector tomography. Noise mitigation enhances computational accuracy by eliminating quantum gate and measurement noise through techniques such as quantum gate noise mitigation, measurement error mitigation, and the Neumann truncation series framework.

The report explains that quantum noise originates from hardware (readout noise, decoherence, crosstalk, leakage) and software (imprecise state preparation, gates, and measurements). Two approaches to handle quantum noise are discussed: bottom-up (improving hardware precision) and top-down (software-based noise characterization and mitigation). QEP implements the top-down approach, providing tools that work across different quantum hardware platforms including Baidu's Qian Shi, Google's Sycamore, and IBM's quantum computers.

Key technical highlights include hardware-adaptive noise characterization methods that address the "curse of dimensionality" by focusing on k-order crosstalk noise based on hardware topology, and a novel Neumann series-based quantum noise mitigation framework that represents the inverse of quantum noise as a linear combination of multiple noise compositions without requiring structural assumptions about the noise.