Quantum Benchmarking
summary
Classifies exactly which parameters of gate noise are learnable for Clifford operations using a pattern transfer graph, shows a CNOT has 13 learnable, 2 unlearnable Pauli-error degrees of freedom, and proves Pauli twirling plus cycle benchmarking (with a simple single-qubit rotation trick) recovers all learnable combinations. For non‑Clifford layers introduces RCS benchmarking (random circuits + XEB), maps average fidelity to a classical spin model and fits an exponential decay e^{-T d} to estimate total error T, giving evidence that Google’s FRB/XEB match implies uncorrelated noise.
AI-written summary · about
chapters (16)
- 0:00Start
- 3:06Why benchmarking?
- 3:54What is noise?
- 5:18Challenges in benchmarking
- 9:50Outline
- 11:24What's the issue?
- 13:46Trivial example
- 14:24Main idea
- 19:16Building the pattern transfer graph
- 22:10Cycle benchmarking with trick
- 26:32The learnability of Pauli noise: example
- 33:25Scalable noise benchmarking methods
- 34:50Motivation: Google's quantum supremacy experiment [Arute et al 19]
- 40:57Fidelity estimation via cross entropy
- 44:09Conclusion
- 44:42Summary
related talks
53mFundamental limits to quantum computation
53mFrom randomized benchmarking to cycle error reconstruction
01h 04mLearning Beyond Stabilizer States
46mQuantum Learning Algorithms Imply Circuit Lower Bounds
49mEvidence for the utility of quantum computing before fault tolerance
28mThe Problem with Qubits
01h 10mQuantum Advantage with Shallow Clifford Circuits
01h 08mPutting Quantum Supremacy on a Rigorous Footing