Quantum computers hold promise for solving problems infeasible for classical machines. However, due to their susceptibility to errors, quantum error correction (QEC) is essential. This thesis investigates surface code decoders based on graph and recurrent neural networks, analyzing their predictions through soft outputs, uncertainty estimation, and hard case identification. For this purpose, we developed a pipeline for the analysis of models soft-outputs, revealing underlying properties of the data and behavior of models. We also established methods for detecting challenging syndromes and explored additional training on these cases. Furthermore, we also implemented a Bayesian approach to quantify uncertainty. Our results show that for lower code distances, decoders approximate maximum-likelihood decoding well, while the larger distance codes succumb to the sparsity of data. The hard case fine-tuning and the Bayesian approach did not significantly enhance performance of our models. Overall, our findings demonstrate the value of the analysis tools and training strategies developed in this work. These methods enabled us to better understand decoder behavior, identify sources of error, and outline concrete pathways for improving quantum error correction in future studies.
Quantum computers hold promise for solving problems infeasible for classical machines. However, due to their susceptibility to errors, quantum error correction (QEC) is essential. This thesis investigates surface code decoders based on graph and recurrent neural networks, analyzing their predictions through soft outputs, uncertainty estimation, and hard case identification. For this purpose, we developed a pipeline for the analysis of models soft-outputs, revealing underlying properties of the data and behavior of models. We also established methods for detecting challenging syndromes and explored additional training on these cases. Furthermore, we also implemented a Bayesian approach to quantify uncertainty. Our results show that for lower code distances, decoders approximate maximum-likelihood decoding well, while the larger distance codes succumb to the sparsity of data. The hard case fine-tuning and the Bayesian approach did not significantly enhance performance of our models. Overall, our findings demonstrate the value of the analysis tools and training strategies developed in this work. These methods enabled us to better understand decoder behavior, identify sources of error, and outline concrete pathways for improving quantum error correction in future studies.
Quantum computers has the potential to outperform classical computers for certain tasks, but the practical use of quantum computers is limited by noise and errors in quantum systems. Quantum Error Correction (QEC) provides a way for detecting and correcting these errors, but in the presence of complex and correlated noise, efficient decoding still remains a significant challenge. In this thesis, neural network-based decoders are explored as an approach to decode QEC data. In particular, recurrent neural network (RNN) architectures, including Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM), are investigated as their are able to capture temporal dependencies and model sequential data. GRU and LSTM models are trained using the data generated with the help of the Stim simulator for quantum circuits that are based on experimental setups. The models are trained to classify error occurrences from detection event data and evaluated under different configurations, including varying number of rounds, code distance and input representation. The results show that both GRU and LSTM models can learn error patterns for shorter sequences and provide competitive decoding performance. However, for longer sequences the models struggle. Overall, this work demonstrates that relatively simple neural network models can be used as alternatives to traditional decoding methods and highlights the potential of machine learning approaches for improving quantum error correction.
As quantum computers remain susceptible to noise, QEC is essential for preserving logical information during computations. However, the performance of QEC codes breaks down beyond certain noise thresholds, revealing fundamental limits on their ability to protect quantum information. These limits can be characterized using information-theoretic measures such as the coherent information, which quantifies the maximum rate at which logical information can be reliably transmitted through a noisy quantum channel. In this work, we establish a direct connection between the CI and the binary cross-entropy loss used when training neural network decoders. Specifically, we show that the CI constitutes a sharp lower bound on the achievable loss for decoders that track logical operators across noisy channels. To this end, we develop a transformer-based neural network model based on maximum likelihood decoding. We train this network to estimate the CI and evaluate its performance on the surface code under three noise models: code capacity, phenomenological, and circuit-level noise. Our results demonstrate that the network accurately predicts CI and yields threshold estimates that closely match known theoretical limits. When used as a decoder, the network significantly outperforms the minimum weight perfect matching decoder in terms of logical error rate. We also introduce a novel soft post-selection scheme that independently treats uncertainty in both logical operators and relies on confidence-based filtering of the network's output. We prove that such post-selection strategies, based on the MLD cosets, are optimal, and demonstrate their scalability in terms of both logical error rate and abort probability. These findings establish transformer-based architectures as powerful tools for QEC and provide the first numerical evidence supporting the optimality and scalability of MLD-based post-selection.
Quantum error correction (QEC) is key to have reliable quantum computation and storage, due to the fragility of qubits in current quantum technology and the imperfect application of quantum operations. In order to have efficient quantum computation and storage, active QEC is required. QEC consists of an encoding and a decoding process. The way that encoding protects quantum information is through grouping many unreliable physical qubits into one more reliable logical qubit. Then, computation occurs based on the logical qubits, however, errors still occur on the physical qubits. Decoding is the process of identifying the location and type of errors occurring on the physical qubits. The decoder proposes corrections against the errors that have been identified. In this thesis, we are exploring novel ways to design decoders for QEC codes, focusing on the surface code. We began our investigation by implementing a rule-based decoder for the smallest surface code, which consists of 17 qubits. We incorporated this decoder to a platform that we created, called Quantum Platform Development Framework (QPDO), in order to study the working principles of a Pauli frame and to quantify its potential effect on the decoding performance. The Pauli frame unit keeps track of errors on physical qubits without the need to apply corrections constantly. We quantified through simulation the benefits in terms of the decoding performance and the execution schedule of QEC, minimizing the idle time. Minimizing the execution time is critical, due to the limited time budget of quantum error correction, thus requiring a high speed decoder capable of still reaching high decoding performance. We show that when the decoding time is equal to the time required to run a surface code cycle, the decoder reaches its maximum performance. However, such a rule-based decoder cannot easily scale to larger quantum systems, therefore other decoding approaches should be considered. Most of the classical decoders that have been developed so far, do not have a good balance between short execution time and high decoding performance. Therefore, we proposed decoders that incorporate neural networks to keep the execution time small, while keeping the decoding performance high. We designed a two-module decoder, which included a classical module and a neural network. We named this configuration neural network based decoder (NNbD). We compare different designs of NNbDs with classical decoders and prove that NNbDs can reach similar or better decoding performance compared to classical decoders while having constant execution time. Furthermore, we quantified the execution time of a NNbD and argued about the speed that can be achieved in a hardware chip like a Field Programmable Gate Array (FPGA) or an Application-Specific Integrated Circuit (ASIC). Both the classical module and the neural network are highly parallelizable and fast modules by construction, leading to constant execution time for a given code distance. We proved that neural network based decoders can adapt to any noise model, since the neural network functionality is based on creating a map between the input and output data, requiring no knowledge about the underlying error model. Following that, a comparison between different NNbD design approaches was performed. We show that it is advantageous to start with a classical decoding module and improve on its decoding performance with a neural network rather than having a neural network perform the decoding on its own. Also, in the latter case, the execution time of such a decoder is non-constant and on average larger than the decoder containing a classical module and a neural network. Moreover, we show that for the design containing a classical module and a neural network, the execution time is increasing linearly as the code distance increased, which was mainly attributed to the increase of the size of the neural network. However, there is a fundamental difference between NNbDs and classical decoders in that NNbDs require sampling and training based on data obtained from the problem, unlike classical decoders. As the code distance increases, the amount of data required to be gathered and trained are exponentially increasing, imposing a limit to the size of the quantum system that can be efficiently decoded. We proposed as a solution to have a distributed decoding approach that divides the code into small regions and then decodes each region locally. We show that using such a distributed decoding approach for small code distances does not lead to significant loss in decoding performance, while simultaneously providing a way to decode large code distances. Thus, we were able to create a decoder that can achieve high decoding performance with constant execution time. However, there are still some issues to keep in mind with such kind of decoders. The main challenge of NNbDs is that they are a dedicated decoder for a given problem. Every time that some aspect of the problem changes (quantum error correcting code, code distance, error model), sampling, training and evaluating the decoder needs to be repeated. Moreover, there is a large number of neural network parameters that need to be specifically tuned when the problem changes. A careful study of the design choices is required to maximize the performance of the decoder. We envision that when sampling and training are performed in hardware, the time required for these processes will be decreased compared to the time required in software. Finally, if the hardware resources allow us to include multiple neural networks, then this can potentially increase the decoding performance. As we presented, dividing the task of decoding to smaller tasks that are distributed to many neural networks can be beneficial.
Quantum error correction (QEC) is indispensable for realizing fault-tolerant quantum computation, yet its effectiveness hinges critically on the classical decoding algorithm that interprets noisy syndrome measurements. Among all possible decoding strategies, maximum likelihood decoding (MLD) is provably optimal, since it identifies the logical group with largest likelihood by summing over all possible errors within logical class consistent with the observed syndrome. Despite its optimality, MLD is computationally intractable in general (#P-hard), motivating a rich landscape of exact and approximate algorithms. In this topical review, we provide a unified perspective on MLD by surveying recent advances through three complementary lenses: statistical mechanics, tensor networks, and artificial intelligence. From the statistical mechanics viewpoint, the MLD problem maps onto evaluating partition functions of disordered spin models, enabling exact solutions for certain codes and noise models as well as threshold estimation via phase-transition analysis. From the tensor network perspective, approximate contraction of tensor networks on the code's factor graph yields decoders that closely approach MLD accuracy with polynomial computational cost. From the artificial intelligence perspective, neural-network-based decoders, including autoregressive generative models and recurrent transformers, learn to approximate the MLD distribution from data, achieving high accuracy with the parallelism afforded by modern hardware accelerators. We discuss the connections among these three approaches, review their application to both simulated and experimental quantum hardware, and outline open challenges including real-time decoding, scalability to large code distances, and generalization to high-rate quantum low-density parity-check codes.
Quantum error correction is essential to achieve fault-tolerant quantum computation in the presence of noisy qubits. Among the most promising approaches to quantum error correction is the surface code, thanks to a scalable two-dimensional architecture, only nearest-neighbor interactions, and a high error threshold. Decoding the surface code, i.e. finding the most likely error chain given a syndrome measurement outcome is a computationally complex task. Traditional decoders rely on classical algorithms, which, especially for larger systems, can be slow and may not always converge to the optimal solution. This thesis presents a novel approach to decoding the surface code using graph neural networks. By mapping the syndrome measurements to a graph and performing graph classification, we find that the graph neural networks can predict the most likely error configuration with high accuracy. Our results show that the GNN-based decoder outperforms the classic minimum weight perfect matching (MWPM) decoder in terms of accuracy. With a phenomenological noise model with depolarizing noise and perfect syndrome measurements, our networks beat MWPM up to code-size 15 across all relevant error rates. Furthermore, the GNN is capable of surpassing MWPM under circuit-level noise up to code size 7. We also show that training the network on repetition code data from a recent experiment [Google Quantum AI, Nature 614, 676 (2023)] produces per-step error rates comparable to those achieved with a matching decoder specifically adapted to the error rates of the physical qubits. This indicates that graph neural network decoders are capable of learning the underlying error distribution on the qubits. Our findings advance the field of quantum error correction and provide a promising new direction for the development of efficient decoding algorithms.
Abstract Building a large-scale quantum computer requires effective strategies to correct errors that inevitably arise in physical quantum systems 1 . Quantum error-correction codes 2 present a way to reach this goal by encoding logical information redundantly into many physical qubits. A key challenge in implementing such codes is accurately decoding noisy syndrome information extracted from redundancy checks to obtain the correct encoded logical information. Here we develop a recurrent, transformer-based neural network that learns to decode the surface code, the leading quantum error-correction code 3 . Our decoder outperforms other state-of-the-art decoders on real-world data from Google’s Sycamore quantum processor for distance-3 and distance-5 surface codes 4 . On distances up to 11, the decoder maintains its advantage on simulated data with realistic noise including cross-talk and leakage, utilizing soft readouts and leakage information. After training on approximate synthetic data, the decoder adapts to the more complex, but unknown, underlying error distribution by training on a limited budget of experimental samples. Our work illustrates the ability of machine learning to go beyond human-designed algorithms by learning from data directly, highlighting machine learning as a strong contender for decoding in quantum computers.