005 - Maximizing Performance on Noisy Quantum Devices with Pedro Rivero: Qiskit Summer School 2024

Finished

Link

https://www.youtube.com/watch?v=l5V8J_yMEis&list=PLOFEBzvs-Vvr-GzDWlZpAcDpki5jUqYJu&type=snipo

Note

Status

Completed

Key points:

Quantum computers are noisy "in every way possible." This emphasizes that noise affects quantum systems in multiple forms, making calculations less reliable.

Fault tolerance is still unfeasible today. Fault tolerance refers to the ability of a system to continue operating correctly in the presence of faults or errors. In quantum computing, fault-tolerant quantum computers are not yet available.

Interim solutions for improving signal quality:

Limit the amount of noise: Efforts are made to minimize the noise during quantum computations.

Clean the signal by filtering the noise out: Post-processing techniques are applied to extract meaningful data from noisy quantum computations.

Accomplishing these goals:

Run modified noisy quantum computations: Adjust the quantum computations to handle noise better.

Process collected outputs on a classical computer: Use classical computing power to refine and process the noisy results from quantum computers.

Compute an improved result: After processing, the result is improved, reducing the impact of noise on the final computation.

The slide also includes a cartoon with a humorous quote: "Well, your quantum computer is broken in every way possible simultaneously." This reflects the complexity of managing noise in quantum systems.

This slide is focused on fighting noise in quantum systems and presents three strategies for dealing with errors:

1. Suppression

Goal: Reduce or avoid the impact of errors before or during execution.

When it happens: Typically before or during the quantum computation.

Resources required: Additional classical resources are needed to implement suppression techniques.

2. Mitigation

Goal: Filter errors out after they occur.

When it happens: After or during execution.

Resources required: Requires additional quantum resources to apply the mitigation techniques effectively.

3. Correction

Goal: Detect and fix errors as they occur during execution.

When it happens: During quantum computation.

Resources required: Both quantum and classical resources are needed for error correction.

This hierarchical approach shows how quantum systems currently manage errors at different stages—from preventing them before they occur (suppression) to fixing them during computation (correction). Each method requires varying levels of computational resources, with the most complex (correction) needing both classical and quantum systems working together.

错误处理类型	定义	技术/方法	应用范围	优点	缺点
错误抑制 (Error suppression)	利用对不希望效应的知识，通过定制化来预测和避免潜在影响	动态解耦（如自旋回波）<br>DRAG（衍生移除通过绝热门）<br>Qiskit Pulse自定义脉冲	硬件层面	用户无需了解，自动处理	需要深入了解硬件
错误缓解 (Error mitigation)	使用电路集合的输出来减少或消除噪声对期望值估计的影响	概率错误抵消<br>零噪声外推（ZNE）<br>M3<br>Twirled Readout Error eXtinction (TREX)	近期实用的量子计算机	能计算无噪声的期望值	每种方法都有其开销，且运行时间随问题规模指数增长
错误校正 (Error correction)	通过冗余编码信息，即使部分量子比特出错，系统也能返回准确的答案	量子错误校正码（如表面码）	容错量子计算	实现硬件依赖的最小错误率	需要大量物理量子比特，目前不现实

错误处理类型	目标	研究进展	挑战
错误抑制	改善硬件控制，增加硬件冗余	几十年前的技术，如自旋回波	需要与硬件紧密结合
错误缓解	实现近期实用的量子计算机	研究社区和IBM量子网络正在开发新算法和应用	需要权衡准确性和开销
错误校正	实现容错量子计算	发现了线性扩展的代码，减少了开销	需要降低物理错误率，发现新的代码

Source:

‣

Why This Quantum Pioneer Thinks We Need More People Working on Quantum Algorithms | by Qiskit | Qiskit | Medium

Aharonov says that the search for new approaches to quantum algorithm design have not received the same attention in recent years. She suggested that researchers may avoid these areas because they require deep knowledge of many different topics in advanced mathematics, including combinatorics, random walks, and group representation theory, to name just a few.

‣

Error mitigation is the path to quantum computing usefulness | IBM Quantum Computing Blog

IBM Quantum roadmap to build quantum-centric supercomputers | IBM Quantum Computing Blog

Error mitigation is the path to quantum computing usefulness | IBM Quantum Computing Blog

Differences in error suppression, mitigation, and correction | IBM Quantum Computing Blog

The two URLs provided in the image are:

For Sampler options: https://docs.quantum.ibm.com/api/qiskit-ibm-runtime/qiskit_ibm_runtime.options.SamplerOptions

For Estimator options: https://docs.quantum.ibm.com/api/qiskit-ibm-runtime/qiskit_ibm_runtime.options.EstimatorOptions

These links lead to the official IBM Quantum documentation for the SamplerOptions and EstimatorOptions classes in the Qiskit IBM Runtime module, providing detailed information on how to configure these components when using Qiskit for quantum computations.

combination of first 2

The image discusses noise amplification (ZNE) in quantum computing, presenting three main techniques: pulse stretching, gate folding, and probabilistic error amplification (PEA).

Pulse stretching: This method requires costly pulse level calibration of the hardware. It involves scaling pulse duration via calibration.

Gate folding: Described as a largely heuristic approach, it offers a good trade-off between result quality and resource requirements. The technique involves repeating gates in identity cycles, represented by the formula U(U^(-1)U)^(λ-1)/2.

Probabilistic error amplification (PEA): This method requires learning circuit-specific noise but has general applicability and strong theoretical backing. It adds noise via sampling Pauli channels.

The image includes diagrams for each technique, citing research papers by Kandala et al. (2019), Shultz et al. (2022), and Li & Benjamin (2017).

The image discusses extrapolation techniques in the context of Zero Noise Extrapolation (ZNE) for quantum computing. Here are the key points:

Theoretical and experimental results predict an exponential decay in observed expectation values.

Exponential extrapolation is aggressive in mitigating noise but unstable due to unknown scale.

Polynomial extrapolation is stable but less effective at mitigating noise as it retains the scale of noisy data.

The method requires careful attention but shows great potential.

The main graph shows an exponential decay curve y(x) = A exp(-x/L) with A = 1.0 and L = 1.0, illustrating the expected behavior of quantum system performance under noise.

Four smaller graphs at the bottom demonstrate different scenarios:

A = 0.0; L ∈ ℝ+: Shows linear extrapolation working

A = ±0.2; L = 0.1: Displays noise saturation

A = ±0.5; L = 0.1: Also shows noise saturation

A = ±1.0; L = 0.1: Again demonstrates noise saturation

I'll summarize the key points about Probabilistic Error Amplification (PEA) from the image without reproducing any potentially copyrighted text verbatim:

The slide discusses PEA as a noise amplification technique for Zero Noise Extrapolation (ZNE) in quantum computing. It involves executing statistical ensembles of circuits and has two main tasks per layer: noise learning and noise injection.

The process is broken down into three steps:

Designing twirl layers of 2-qubit gates

Repeating identity pairs of layers and learning the noise

Deriving a fidelity error for each noise channel

The method is described as having general applicability and strong theoretical backing.

Diagrams illustrate the circuit structure and the process of noise amplification.

This image discusses strategies for combating noise in quantum computing before applying error correction. The key points are:

Different types of noise require different suppression and mitigation techniques.

Various techniques can be combined to address different noise sources.

The diagram illustrates a quantum circuit with annotations for different noise types and mitigation strategies:

Environmental noise: Addressed using Dynamical Decoupling (DD)

Gate errors: Mitigated using Zero Noise Extrapolation (ZNE) and Pauli Twirling (PT)

Readout errors: Handled with Readout Error Mitigation (TREX)

Here's the code shown in the image:

This code demonstrates how to combine different quantum error mitigation techniques using the Qiskit framework. It configures options for Dynamical Decoupling, Twirling, TREX (Readout Error Mitigation), and ZNE (Zero Noise Extrapolation).