In the quest for fault-tolerant quantum computation, the delicate nature of qubits demands the need for robust quantum error correction. Quantum error correcting codes, the quantum counterpart to classical error correcting codes in information theory, have been developed to address the unique challenges quantum bits face against the noise of their environment in storing information. These codes include but are not limited to the Kitaev surface code, toric code, rotated surface code, generalized bicycle code, and bivariate bicycle code. Fast and reliable decoders for these quantum error correcting codes are needed to correct errors and achieve fault tolerance on these qubits. The belief propagation algorithm, which is a efficient and reliable heuristic for decoding classical low-density parity check codes, has been generalized to different types of quantum error correction codes. Paired with the ordered statistics decoding algorithm as a post-processing step from the soft outputs of the belief propagation decoder, belief propagation and ordered statistics decoding has been one of the most promising avenues of efficient quantum error correction. This thesis combines the state-of-the-art belief propagation and ordered statistics decoding algorithms with the exploration of the neural belief propagation algorithm to leverage classical deep learning machines to learn message passing patterns in decoding the newest quantum codes, the rotated surface code and bivariate bicycle code. The neural network exploits the local lattice structure of these codes to adjust the trainable weights and biases to more learn message update patterns more efficiently than the vanilla belief propagation algorithm. In this thesis, we prove a new threshold for the physical error rate on the rotated surface code using a neural belief propagation algorithm and ordered statistics decoding that consistently outperforms the threshold established by the vanilla belief propagation algorithm with ordered statistics decoding. We additionally show the results for similar efforts on the bivariate bicycle codes, and discuss the difficulties in achieving a threshold using the current neural belief propagation algorithms.
In the quest for fault-tolerant quantum computation, the delicate nature of qubits demands the need for robust quantum error correction. Quantum error correcting codes, the quantum counterpart to classical error correcting codes in information theory, have been developed to address the unique challenges quantum bits face against the noise of their environment in storing information. These codes include but are not limited to the Kitaev surface code, toric code, rotated surface code, generalized bicycle code, and bivariate bicycle code. Fast and reliable decoders for these quantum error correcting codes are needed to correct errors and achieve fault tolerance on these qubits. The belief propagation algorithm, which is a efficient and reliable heuristic for decoding classical low-density parity check codes, has been generalized to different types of quantum error correction codes. Paired with the ordered statistics decoding algorithm as a post-processing step from the soft outputs of the belief propagation decoder, belief propagation and ordered statistics decoding has been one of the most promising avenues of efficient quantum error correction. This thesis combines the state-of-the-art belief propagation and ordered statistics decoding algorithms with the exploration of the neural belief propagation algorithm to leverage classical deep learning machines to learn message passing patterns in decoding the newest quantum codes, the rotated surface code and bivariate bicycle code. The neural network exploits the local lattice structure of these codes to adjust the trainable weights and biases to more learn message update patterns more efficiently than the vanilla belief propagation algorithm. In this thesis, we prove a new threshold for the physical error rate on the rotated surface code using a neural belief propagation algorithm and ordered statistics decoding that consistently outperforms the threshold established by the vanilla belief propagation algorithm with ordered statistics decoding. We additionally show the results for similar efforts on the bivariate bicycle codes, and discuss the difficulties in achieving a threshold using the current neural belief propagation algorithms.
Abstract To improve the decoding performance of quantum error-correcting codes in asymmetric noise channels, a neural network-based decoding algorithm for bias-tailored quantum codes is proposed. The algorithm consists of a biased noise model, a neural belief propagation decoder, a convolutional optimization layer, and a multi-objective loss function. The biased noise model simulates asymmetric error generation, providing a training dataset for decoding. The neural network, leveraging dynamic weight learning and a multi-objective loss function, mitigates error degeneracy. Additionally, the convolutional optimization layer enhances early-stage convergence efficiency. Numerical results show that for bias-tailored quantum codes, our decoder performs much better than the belief propagation (BP) with ordered statistics decoding (BP + OSD). Our decoder achieves an order of magnitude improvement in the error suppression compared to higher-order BP + OSD. Furthermore, the decoding threshold of our decoder for surface codes reaches a high threshold of 20%.
Fault-tolerant quantum computers will depend crucially on the performance of the classical decoding algorithm which takes in the results of measurements and outputs corrections to the errors inferred to have occurred. Machine learning models have shown great promise as decoders for the surface code; however, this promise has not yet been substantiated for the more challenging task of decoding quantum low-density parity-check (QLDPC) codes. In this paper, we present a recurrent, transformer-based neural network designed to decode circuit-level noise on Bivariate Bicycle (BB) codes. For the [ [ 72 , 12 , 6 ] ] BB code, at a physical error rate of p = 0.1 % , our model achieves logical error rates almost 5 times lower than belief propagation with ordered statistics decoding (BP-OSD), and roughly 5 times larger than a most-likely error decoder. Moreover, while BP-OSD has a wide distribution of runtimes with significant outliers, our model has a consistent runtime and is an order-of-magnitude faster than the worst-case times from a benchmark BP-OSD implementation. On the [ [ 144 , 12 , 12 ] ] BB code, our model obtains worse logical error rates but maintains the speed advantage. These results provide initial evidence that machine learning decoders can out-perform conventional decoders on small QLDPC codes, but suggest more complex architectures and/or training procedures are necessary to scale to larger code sizes.
Error correction is used to correct any errors that may arise when sending any message, as errors are prone to appear due to noise. For classical codes, there are many decoders used, for example, belief propagation. Unfortunately, these classical decoders are less efficient for quantum codes. A proposed remedy to improve the decoders efficiency is to use a neural network for decoding. In this thesis, we will implement belief propagation on the toric code and check its efficiency. We will see that for few errors, belief propagation works, but it fails for other cases. We also introduce the modifications needed for neural belief propagation, a modification of the original algorithm that integrates a neural network on the algorithm structure.
Efficient and accurate decoding of quantum error-correcting codes is essential for fault-tolerant quantum computation, however, it is challenging due to the degeneracy of errors, the complex code topology, and the large space for logical operators in high-rate codes. In this work, we propose a decoding algorithm utilizing generative modeling in machine learning. We employ autoregressive neural networks to learn the joint probability of logical operators and syndromes in an unsupervised manner, eliminating the need for labeled training data. The learned model can approximately perform maximum likelihood decoding by directly generating the most likely logical operators for k logical qubits with O(2k) computational complexity. Thus, it is particularly efficient for decoding high-rate codes with many logical qubits. The proposed approach is general and applies to a wide spectrum of quantum error-correcting codes including surface codes and quantum low-density parity-check codes (qLDPC), under noise models ranging from code capacity noise to circuit level noise. We conducted extensive numerical experiments to demonstrate that our approach achieves significantly higher decoding accuracy compared to the minimum weight perfect matching and belief propagation with ordered statistics on the surface codes and high-rate quantum low-density parity-check codes. Our approach highlights generative artificial intelligence as a potential solution for the real-time decoding of realistic and high-rate quantum error correction codes.
Decoding Bivariate Bicycle (BB) quantum error correction codes typically requires Belief Propagation (BP) followed by Ordered Statistics Decoding (OSD) post-processing when BP fails to converge. Whether BP will converge on a given syndrome is currently determined only after running BP to completion. We show that convergence can be predicted in advance by a single modulo operation: if the syndrome defect count is divisible by the code's column weight w, BP converges with high probability (100% at p <= 0.001, degrading to 87% at p = 0.01); otherwise, BP fails with probability >= 90%. The mechanism is structural: each physical data error activates exactly w stabilizers, so a defect count not divisible by w implies the presence of measurement errors outside BP's model space. Validated on five BB codes with column weights w = 2, 3, and 4, mod-w achieves AUC = 0.995 as a convergence classifier at p = 0.001 under phenomenological noise, dominating all other syndrome features (next best: AUC = 0.52). The false positive rate scales empirically as O(p^2.05) (R^2 = 0.98), confirming the analytical bound from Proposition 2. Among BP failures on mod-w = 0 syndromes, 82% contain weight-2 data error clusters, directly confirming the dominant failure mechanism. The prediction is invariant under BP scheduling strategy and decoder variant, including Relay-BP - the strongest known BP enhancement for quantum LDPC codes. These results apply directly to IBM's Gross code [[144, 12, 12]] and Two-Gross code [[288, 12, 18]], targeted for deployment in 2026-2028.
Post-selection improves the logical reliability of quantum computation by rejecting shots which are likely to result in logical failure. We introduce graph neural networks that predict decoder failure without additional decoder executions, using only syndromes, and for high-rate quantum low-density parity-check (qLDPC) codes, existing belief-propagation posteriors. We evaluate rotated surface and bivariate bicycle (BB) code memories under uniform depolarising circuit-level noise. Retaining 90% of shots gives approximately 3700x and 740x reductions in logical error rates for [[72,12,6]] and [[144,12,12]] BB codes at realistic physical error rates. For the distance 5 surface code at physical noise p=0.003, our neural post-selector achieves logical error suppression comparable to the leading complementary gap method without running the decoder.