Quantum Error Mitigation using Machine Learning
This project demonstrates quantum error mitigation techniques using machine learning to improve the reliability of quantum computations in noisy quantum computers. The main goal is to mitigate measurement errors that occur during quantum computation, as measurement is a critical step for extracting information from quantum states.
The project is implemented using Qiskit, a popular quantum computing library, and Scikit-learn, a well-known machine learning library in Python. The error mitigation techniques explored are:
-
Complete Measurement Calibration (CMC): This technique uses the CompleteMeasFitter class from Qiskit's Ignis module to create an error mitigation filter. The filter is trained on calibration data obtained by running the quantum circuit on a noisy backend, and it is then applied to the noisy circuit results to improve fidelity.
-
Zero-Noise Extrapolation (ZNE): ZNE involves measuring the error probabilities for different noise levels by executing the circuit on a noisy backend. These error probabilities are used to create a MeasurementFilter object. The filter is then applied to the circuit results obtained for various noise levels, including extrapolating to zero noise, to achieve error mitigation.
-
Probabilistic Error Cancellation (PEC): This technique measures the error probabilities for the default noise level and uses them to create a MeasurementFilter object. The filter is then applied to the circuit results to mitigate errors.
The project provides an example of generating training data from the reduced density matrix of the first qubit and training a RandomForestClassifier to model the error probabilities. Other machine learning models can be experimented with as well.
By implementing these error mitigation techniques, the project aims to enhance the robustness of quantum algorithms, making them more practical and reliable in real-world noisy quantum computing environments. The results obtained from applying the different mitigation techniques are printed and analyzed to assess their effectiveness in reducing errors.