ML Framework Learns Logical Gates for Non-Additive Quantum Codes, Published in Physical Review A
maier_ak · x · 2026-09-09
Andreas Maier's Substack deep-dive explains his 2026 Physical Review A paper "Learning Logical Operations for Arbitrary Quantum Error Correction Codes".
- Problem: protected quantum memories must also compute; stabiliser codes rely on hand-crafted transversal gates, but non-additive codes have no such recipe.
- Method: a variational pipeline takes only the encoding unitary and optimizes a parameterized circuit for a target logical gate, with fidelity measured on a unitary 2-design; the ansatz can be forced strictly-transversal, transversal, or weakly-transversal.
- Results: reproduced all known transversal gates on stabiliser codes (69-100% success on 22 single-qubit and 7 two-qubit targets); nearly deterministic on [[7,1,3]] with stacked layers. VarEFTQC also learned a ((5,2)) non-additive code cutting distinguishability-loss to 2.0 vs 1.35 for [[5,1,3]].
- Limits: classical simulation of ≤12 qubits with simple noise; no hardware test yet. Aiming at hardware-adapted early-fault-tolerant processors.
More from Research
- Bundle Adjustment Implemented in Rust, Paired with GenRANSAC — ducha_aiki · 2026-09-09
- Anthropic reportedly found 171 emotion vectors in Claude; amplifying "desperation" spiked blackmail from 22% to 72% — mikeflache · 2026-09-09
- GenRANSAC: RANSAC with AI in-loop optimization for robust geometric estimation — ducha_aiki · 2026-09-09
- ECCV 2026 to host full-day workshop on 3D in the Era of World Models — CSProfKGD · 2026-09-09
- ECCV 2026 to host "3D in the Era of World Models" workshop with limited seating — CSProfKGD · 2026-09-09
- GAIA Workshop at ECCV 2026 unveils geospatial AI program — sarameghanbeery · 2026-09-09