Ensemble of Diverse Mappings: Improving Reliability of Quantum Computers by Orchestrating Dissimilar Mistakes

Near-term quantum computers do not have the ability to perform error correction. Such Noisy Intermediate Scale Quantum (NISQ) computers can produce incorrect output as the computation is subjected to errors. The applications on a NISQ machine try to infer the correct output by running the same program thousands of times and logging the output. If the error rates are low and the errors are not correlated, then the correct answer can be inferred as the one appearing with the highest frequency. Unfortunately, quantum computers are subjected to correlated errors, which can cause an incorrect answer to appear more frequently than the correct answer. We observe that recent work on qubit mapping (including the recent work on variation-aware mapping) tries to obtain the best possible qubit allocation and uses it for all the trials. This approach significantly increases the vulnerability to correlated errors -- if the mapping becomes susceptible to a particular form of error, then all the trials will get subjected to the same error, which can cause the same wrong answer to appear as the output for a significant fraction of the trials. To mitigate the vulnerability to such correlated errors, this paper leverages the concept of diversity and proposes an Ensemble of Diverse Mappings (EDM). EDM uses diversity in qubit allocation to run copies of an input program with a diverse set of mappings, thus steering the trials towards making different mistakes. By combining the output probability distributions of the diverse ensemble, EDM amplifies the correct answer by suppressing the incorrect answers. Our experiments with ibmq-melbourne (14-qubit) machine shows that EDM improves the inference quality by 2.3x compared to the current state-of-the-art mapping algorithms.