Robust quantum information processing: decoding, lattice surgery, and quantum associative neural networks
Year: 2026
Type: dissertation
Abstract: Scalable quantum computation requires protocols that protect quantum information from noise while remaining compatible with realistic hardware constraints. This thesis develops and analyzes methods for diagnosing, optimizing, and extending such protocols in active quantum error correction and related open-system memory models. First, neural-network and graph-matching-based decoders are studied for stabilizer codes under circuit-level noise. We show that machine-learning-based decoders can learn fault-tolerant decoding strategies, and interpretability methods are introduced to analyze their individual decisions. The resulting attribution patterns identify physically meaningful syndrome signatures, monitor the emergence of fault-tolerant behavior during training, and reveal architectural failure modes. Further, a syndrome-correlation-based noise-inference method is extended to high-weight error signatures and applied to repeated stabilizer-measurement data from a surface-17 superconducting-qubit experiment. Based on the inference, we adapt a graph-matching-based decoder to the observed noise environment. In addition, a correlated matching decoder is introduced, in which information from one syndrome sector is used to modify the decoding graph of the complementary sector. As a next step, lattice surgery is studied as a primitive for creating entanglement between encoded surface-code qubits in planar superconducting architectures. Numerical simulations for two distance-three surface-code patches identify optimized schedules that reduce the anticipated logical failure probability by about 50%. Adaptive stabilizer-measurement schedules and reduced-connectivity constructions are analyzed, showing how lattice-surgery performance depends jointly on circuit design, decoding information, classical feedback latency, and hardware connectivity. We present experimental results for lattice-surgery-based entanglement creation with two repetition codes; these results indicate that the encoded bit-flip error-corrected protocol outperforms the corresponding physical-qubit-based entanglement creation protocol. Finally, an open quantum Hopfield model is analyzed as an autonomous associative memory. Further, a method is developed to characterize how many distinct stationary memory states can be supported in this open quantum system. This method is then used to calculate the maximal storage capacity of the considered open quantum Hopfield model. Together, these results show how learned decoding, experimental noise inference, hardware-aware logical operations, and open-system memory dynamics can be used to control quantum information in noisy devices.
Source: RWTH Publications (RWTH Aachen)
Author: Bödeker, Lukas
Institution: RWTH Aachen University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225868394
Cites:
Cited by:
Related to: