Andi Gu

With @AndiGu00, we introduce Cascade; a state-of-the-art quantum error correction decoder. We show that utility-scale logical error rates are achievable with modest code sizes at current physical error rates, with real-time compatible latency. 📄 scirate.com/arxiv/2604.083…

@ShohamJacoby @letonyo Also, just to clarify - as Anthony noted, the figure from Bravyi et al. plots error rate per round on the y-axis, while ours shows error rate per round *per logical qubit*, so the difference is a factor of 12 (not 144)

Hey Shoham, thanks for the interest. The gap comes from decoder settings: IBM runs a much heavier BP+OSD configuration than the stimbposd (github.com/oscarhiggott/s…) defaults we used. We stuck with defaults because BP+OSD is already the slowest point on our Pareto plot (Fig 1c); Relay (for which we used the recommended tuning + configuration from github.com/trmue/relay) is faster and more accurate. Matching IBM's BP+OSD settings would've made sweeping noise rates and code sizes intractable for us.

Might be time to retire BP+OSD soon. Many new decoders for qLDPC codes with really impressive performance these days, e.g. the Cascade decoder from the Harvard group

We also put a working Cascade model in the browser. You can inject errors, and watch the network think 🔬 andigu.github.io/projects/casca… Paper: arxiv.org/abs/2604.08358 w/ @J_Pablo_Bonilla, Mikhail Lukin, and Susanne Yelin

On the [[144,12,12]] Gross code, Cascade reaches ~10⁻¹⁰ logical error rates at physical error rates current hardware already achieves. That's 17x more accurate than the best practical decoders, and 1000-100,000x faster.

New preprint with @J_Pablo_Bonilla! We built a neural network that decodes quantum errors — and discovered that fault-tolerant quantum computing might need far fewer qubits than we thought. 📄 Paper: arxiv.org/abs/2604.08358 Website: andigu.github.io/projects/casca…

Thanks Anthony - really appreciate the engagement! Just to clarify: we don't claim the waterfall is unprecedented for qLDPC codes. In the intro we cite prior indications in quantum codes (e.g., Komoto et al. arXiv:2412.21171) and note the phenomenon is well established classically. What we contribute is resolving the effect deep below threshold on high-rate codes like [[144, 12, 12]], where it has quantitative consequences for resource estimates. The same point applies to surface codes, all mainstream resource estimates (Gidney–Ekerå, Litinski, Beverland et al.) still bake in the distance bound, so quantifying how much steeper the true scaling is directly changes qubit counts. Thanks again for the kind words on Cascade!

strange discussion though about the waterfall region, which the authors claim was never observed before for qLDPC!?? arxiv.org/pdf/2604.08358

With @AndiGu00 we built machine learning (ML) decoders for the recent neutral atom experiment on error-corrected quantum memory and logical computation (arxiv.org/abs/2506.20661), which led us to ask: how general can these decoders be? It turns out, quite general...

I am thrilled to announce our new work at the intersection of many-body physics and computational complexity! In this work, we construct a set of Hamiltonians that are computationally indistinguishable from the GUE ensemble. scirate.com/arxiv/2410.181…

All you need is quantum chaos. Except that you do not need #quantumchaos. scirate.com/arxiv/2410.181… It took me a while to accept that the main result of this work is not wrong, which I still find surprising.

Our paper on classically hard simple quantum systems is out (with Varun Menon and Andi Gu). These systems are promising for giving verifiable such advantages: scirate.com/arxiv/2408.160… #ResearchPapers #QuantumComputing #quantum

What’s the difference between low-magic entanglement and high-magic 🪄entanglement? Is there one? Check our new preprint: scirate.com/arxiv/2403.196… 1/9

New preprint✨ Are you wondering if doped stabilized states occur in nature, and if it's possible to use Clifford+T simulations in many-body physics? scirate.com/arxiv/2403.149…👈 1/6🧵