Evan Hockings

Congratulations!!

New paper outlining my Julia package QuantumACES now out in the Journal of Open Source Software!

Improving error suppression with noise-aware decoding The performance of quantum error correction can be improved with noise-aware decoders, which are calibrated to the likelihood of physical error configurations in a device. Averaged circuit eigenvalue sampling (ACES) is a Pauli noise characterisation technique that can calibrate decoders at the scales required for fault-tolerant quantum computation. We demonstrate that ACES is practically capable of calibrating a fast correlated matching decoder, enabling noise-aware decoding, in circuit-level numerical simulations of the surface code. We find that noise-aware decoding increases the error suppression factor of the code, yielding reductions in the logical error rate that increase exponentially with the code distance. Our results indicate that noise characterisation experiments performed and processed in seconds will suffice to calibrate decoders for fault-tolerant superconducting quantum computers. This establishes the practicality and utility of noise-aware decoding for quantum error correction at scale.

For more, check out my new paper with @acdoherty.bsky.social and Robin Harper! And look forward to some experimental results soon :)
arxiv.org/abs/2502.21044

GitHub - evanhockings/QuantumACES.jl: Design scalable noise characterisation experiments for quantum computers Design scalable noise characterisation experiments for quantum computers - evanhockings/QuantumACES.jl

Stim and PyMatching make this super easy. Characterise a circuit-level Pauli noise model with ACES, throw the noise estimates into your Stim circuit, and then it all just works—thanks @craiggidney.bsky.social and @oscarhiggott.bsky.social!

Code for this now in QuantumACES
github.com/evanhockings...

Yes! Gate times in superconducting architectures indicate that ACES noise characterisation experiments performed and processed in just seconds should suffice. At tens of seconds, ACES noise estimates are nearly indistinguishable from the true noise model for decoding.

This means the reduction in logical error rates from noise-aware decoding increases exponentially with the code distance. While gains are limited for small codes, they're substantial for large ones.

But is noise-aware decoding practical at the scales where it's most helpful?

Why characterise noise in syndrome extraction circuits? One reason: directly improving quantum error correction!

In simulations of the surface code, we find that noise-aware decoding—calibrating the decoder with noise estimates—improves the code's error suppression factor.

My first paper—with @acdoherty.bsky.social and Robin Harper—is now out in PRX Quantum! More to come soon :)
journals.aps.org/prxquantum/a...

Yeah, I have to imagine it’s a tokenisation problem (similar to the ARC-AGI benchmark) and I sort of wonder if the labs find it convenient for these issues to stick around right now (reduced alarm, regulation, etc)…or maybe LLMs just aren’t that smart?