A method and system for qubit loss identification and quantum error correction

Abstract

The application provides a quantum bit loss identification and quantum error correction method and system, the method comprising: extracting a space-time correlation feature from an obtained multi-round stabilizer measurement sequence; identifying a persistent random flicker mode caused by quantum bit loss according to the space-time correlation feature, and determining a logical decoding result and a quantum bit loss prediction result. The application can simultaneously realize logical decoding and quantum bit loss position identification, and improve the decoding accuracy and overall reliability of a quantum error correction system in a bit loss noise environment.

Description

Technical Field

[0001] This invention relates to the field of quantum computing technology, and in particular to a method and system for identifying lost qubits and for quantum error correction. Background Technology

[0002] Quantum computing systems are susceptible to various noises during operation, such as decoherence, gate operation errors, and measurement errors, which can cause deviations in the state of physical qubits. To improve the reliability of quantum information storage and processing, quantum error correction techniques are typically used to protect logical qubits. Among existing quantum error correction methods, a common approach is the stable subcode scheme, such as surface codes. This scheme encodes logical qubits onto multiple physical qubits and periodically measures the stable sub-operator to obtain time-evolving correction information. The decoder then infers the error and provides the recovery result based on this correction information. Stable subcodes, especially surface codes, are widely considered a crucial technological approach for achieving fault-tolerant quantum computing. Traditional stable subcode decoding schemes are primarily designed for Pauli-type errors and struggle to simultaneously achieve logical error correction and qubit loss detection. Summary of the Invention

[0003] This invention provides a method and system for qubit loss identification and quantum error correction, addressing the shortcomings of traditional stable subcode decoding schemes in effectively distinguishing between bit loss and Pauli errors. Based on multi-round stable subcode measurement information, this invention simultaneously achieves logical decoding and qubit loss location identification, improving the decoding accuracy and overall reliability of the quantum error correction system under qubit loss noise environments.

[0004] This invention provides a method for qubit loss identification and quantum error correction, comprising: extracting spatiotemporal correlation features from an obtained multi-round stable sub-measurement sequence; identifying a persistent random flickering pattern caused by qubit loss based on the spatiotemporal correlation features; and determining the logic decoding result and the qubit loss prediction result.

[0005] According to the present invention, a method for identifying lost qubits and quantum error correction is provided, wherein extracting spatiotemporal correlation features from the obtained multi-round stable qubit measurement sequence includes: obtaining a T+1 round stable qubit measurement sequence; wherein the first T rounds are from stable qubit measurements and the last round is from data qubit readout; T is an integer not less than 2; and extracting spatiotemporal features from the T+1 round stable qubit measurement sequence using a decoder.

[0006] According to the present invention, a method for identifying lost qubits and quantum error correction is provided, wherein obtaining the T+1 round of stable qubit measurement sequence includes: encoding logical qubits, performing multiple rounds of stable qubit measurement, and obtaining the T+1 round of stable qubit measurement sequence.

[0007] The method for qubit loss identification and quantum error correction provided by the present invention further includes: determining a flickering count feature by counting the detection events corresponding to each auxiliary qubit; adding the flickering count feature to the input representation of the decoder; the flickering count feature is used to distinguish continuous random flickering patterns.

[0008] According to a method for qubit loss identification and quantum error correction provided by the present invention, the decoder includes a feature embedding module, a spatiotemporal feature extraction module, and a dual-task output module; the method further includes: embedding features into the T+1 round stable sub-measurement sequence through the feature embedding module; extracting spatiotemporal features from the embedded features output by the feature embedding module through the spatiotemporal feature extraction module; and outputting the logical decoding result and the qubit loss prediction result through the dual-task output module based on the spatiotemporal features output by the spatiotemporal feature extraction module.

[0009] The quantum bit loss identification and quantum error correction method provided by the present invention further includes: superimposing auxiliary quantum bit index embedding features, round type embedding features, node type embedding features or task type embedding features based on the embedding features of the stable sub-measurement sequence through the feature embedding module.

[0010] According to the present invention, a method for identifying lost qubits and quantum error correction is provided, wherein the spatiotemporal feature extraction module adopts a recurrent neural network structure or a deep spatiotemporal graph neural network structure.

[0011] According to the present invention, a method for qubit loss identification and quantum error correction is provided, wherein the deep spatiotemporal graph neural network structure includes multi-layer cascaded deep network modules; each deep network module includes a graph neural network module, a temporal mixing module, and a spatial attention module; the method further includes: extracting spatial local features from the embedded features output by the feature embedding module through the graph neural network module; performing deep feature interaction in the temporal dimension on the spatial local features output by the graph neural network module through the temporal mixing module; and performing global spatial correlation on the temporal mixing features output by the temporal mixing module through the spatial attention module to output deep spatiotemporal extracted features.

[0012] The method for identifying and correcting lost qubits according to the present invention further includes: employing a delayed feedback mechanism to confirm the final result of qubit loss after a preset number of rounds; the delayed feedback mechanism confirms the result using a sliding window or an overlapping window method.

[0013] The present invention also provides a quantum bit loss identification and quantum error correction system, comprising: a first module for extracting spatiotemporal correlation features from the obtained multi-round stable sub-measurement sequence; and a second module for identifying a persistent random flickering pattern caused by quantum bit loss based on the spatiotemporal correlation features, and determining the logic decoding result and the quantum bit loss prediction result.

[0014] The present invention provides a quantum bit loss identification and quantum error correction method and system, which can simultaneously realize logic decoding and quantum bit loss position identification, thereby improving the decoding accuracy and overall reliability of the quantum error correction system in a bit loss noise environment. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating a quantum bit loss identification and quantum error correction method provided by the present invention.

[0017] Figure 2 This is a schematic diagram of an isolated flip pattern of a single Pauli error.

[0018] Figure 3 This is a schematic diagram of a continuous flickering pattern across multiple measurement rounds caused by the loss of a qubit.

[0019] Figure 4 This is one of the schematic diagrams of the spatiotemporal feature extraction module provided by the present invention.

[0020] Figure 5 This is the second schematic diagram of the spatiotemporal feature extraction module provided by the present invention.

[0021] Figure 6 This is a schematic diagram of the structure of a quantum bit loss detection and quantum error correction system provided by the present invention.

[0022] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0024] In practical quantum hardware, especially neutral atom platforms, in addition to conventional Pauli errors, qubit loss or qubit displacement from the computational subspace can also occur. For qubit loss scenarios, qubit loss and ordinary Pauli errors may exhibit similar compensator modes in a single round of stabilizer measurements, but their temporal behavior differs in multiple rounds of measurements. When a data qubit is lost, the relevant stabilizer measurement results will exhibit persistent random fluctuations in subsequent rounds, a phenomenon known as flickering.

[0025] The purpose of this invention is to provide a quantum bit loss identification and quantum error correction method to overcome the shortcomings of traditional stable subcode decoding schemes. By utilizing multi-round stable subcode measurement information, it can simultaneously achieve logic decoding and quantum bit loss position identification, thereby improving the decoding accuracy and overall reliability of the quantum error correction system in a bit loss noise environment.

[0026] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a quantum bit loss identification and quantum error correction method provided by the present invention.

[0027] Please refer to Figure 2 , Figure 2 This is a schematic diagram of an isolated flip pattern for a single Pauli error.

[0028] Please refer to Figure 3 , Figure 3 A schematic diagram of a continuous flickering pattern across multiple measurement rounds caused by the loss of a qubit.

[0029] This invention provides a method for identifying lost qubits and quantum error correction, comprising: 101: Extract spatiotemporal correlation features from the obtained multi-round stable sub-measurement sequences.

[0030] 102: Based on spatiotemporal correlation characteristics, identify the persistent random flickering pattern caused by the loss of qubits, and determine the logic decoding result and the qubit loss prediction result.

[0031] This invention provides a method for qubit loss identification and quantum error correction, applicable to stable subcode quantum error correction scenarios, especially suitable for continuous quantum error correction processes involving qubit loss or qubit loss noise. The basic idea of ​​this invention is to directly utilize the multi-round stable subcode measurement history (multi-round stable subcode measurement sequence) already obtained during the quantum error correction process to extract its spatiotemporal correlation features, identify the persistent random flickering pattern caused by qubit loss, and simultaneously output the logic decoding result and the qubit loss prediction result. While qubit loss and ordinary Pauli errors may be difficult to distinguish in a single-round corrector mode, the loss event will exhibit persistent random flickering characteristics in subsequent rounds. Therefore, this invention employs a delayed feedback processing mechanism, accumulating several rounds of corrector information before making a decision to improve the reliability of identification.

[0032] This invention is applied to quantum error-correcting code storage tasks (e.g., rotating surface code storage tasks). First, logical qubits are encoded, then multiple rounds of stabilizer measurements are performed. In each round, the stabilizer measurement results of auxiliary qubits and the detection events formed between adjacent rounds are collected. After the storage task is completed, the historical information from these multiple rounds is input into an artificial intelligence decoder. The decoder outputs two results: one is the logical state or logical error prediction result corresponding to the current sample, and the other is the loss prediction result for each round and each data qubit. Subsequently, based on the loss prediction results, supplementation, replacement, reinitialization, or other recovery operations can be performed on the corresponding qubits, and the relevant information is used for error correction processing in subsequent rounds. This invention can simultaneously complete error decoding and loss localization within a continuous quantum error correction framework.

[0033] As a preferred embodiment, spatiotemporal correlation features are extracted from the obtained multi-round stabilizer measurement sequence, including: obtaining a T+1 round stabilizer measurement sequence; wherein the first T rounds are from stabilizer measurements and the last round is from data qubit readout; T is an integer not less than 2; and spatiotemporal features are extracted from the T+1 round stabilizer measurement sequence through a decoder.

[0034] As a preferred embodiment, obtaining the T+1 round of stabilizer measurement sequence includes: encoding the logical qubits, performing multiple rounds of stabilizer measurements, and obtaining the T+1 round of stabilizer measurement sequence.

[0035] In this embodiment, the decoder processes a multi-round stabilizer measurement sequence rather than a single-round measurement. Let T be the cumulative number of processed measurement rounds. In addition to the aforementioned T rounds of stabilizer measurements, an additional round constructed from the final data qubit readout can be added, resulting in a total of T+1 rounds of input processed by the decoder. For each auxiliary qubit in each round, at least two binary signals are input: one is the stabilizer measurement result of that round, and the other is a detection event constructed from the measurement information of the current round and the previous round. The resulting input can be represented as a multidimensional binary tensor arranged in terms of time and auxiliary qubit dimensions, used to characterize the complete multi-round corrector history information. No additional missing flags, leakage flags, or other bypass prior information are required in the input, allowing this invention to directly perform inferences based on data already available in the standard quantum error correction process.

[0036] As a preferred embodiment, the method further includes: determining a flicker count feature by counting the detection events corresponding to each auxiliary qubit; adding the flicker count feature to the input representation of the decoder; and using the flicker count feature to distinguish continuous random flicker patterns.

[0037] To enhance the decoder's sensitivity to qubit loss flickering characteristics, this embodiment also incorporates flickering count features within a short time window into the input representation. Specifically, the number of detection events occurring for each auxiliary qubit in the most recent rounds can be statistically analyzed, and this count is input into the decoder as an additional discrete feature. Since persistent loss typically generates anomalous detection events repeatedly within several adjacent rounds, and ordinary Pauli errors are more often characterized by localized and brief flips, introducing flickering count features helps improve the decoder's ability to distinguish persistent loss patterns.

[0038] In a preferred embodiment, the decoder includes a feature embedding module, a spatiotemporal feature extraction module, and a dual-task output module; the method further includes: embedding features into the T+1 round stable sub-measurement sequence through the feature embedding module; extracting spatiotemporal features from the embedded features output by the feature embedding module through the spatiotemporal feature extraction module; and outputting the logical decoding result and the quantum bit loss prediction result through the dual-task output module based on the spatiotemporal features output by the spatiotemporal feature extraction module.

[0039] As a preferred embodiment, it further includes: superimposing auxiliary qubit index embedding features, round-type embedding features, node-type embedding features, or task-type embedding features based on the embedding features of the stable sub-measurement sequence through a feature embedding module.

[0040] In this embodiment, the decoder as a whole includes a feature embedding module, a spatiotemporal feature extraction module, and a dual-task output module. The feature embedding module maps the original binary input corresponding to each round and each auxiliary qubit to a high-dimensional implicit representation, and can further superimpose auxiliary qubit index embedding, round-type embedding, node-type embedding, or task-type embedding to enhance the model's ability to perceive physical location, temporal location, and task context. The spatiotemporal feature extraction module jointly models the correlation of historical information of multi-round correctors in the temporal and spatial dimensions, thereby learning the different statistical behaviors between ordinary Pauli errors and qubit loss. The dual-task output module simultaneously outputs the logical decoding result and the qubit loss prediction result.

[0041] The spatiotemporal feature extraction module is used to iteratively update the historical information of the multi-round corrector (T+1 round stable sub-measurement sequence). For the input of each measurement round, the embedded features of the current round are combined with the temporal information stored in other rounds, and further combined with the spatial correlation between qubits for joint processing, thereby obtaining a hidden representation that can characterize the difference between ordinary Pauli errors and qubit loss. Spatial correlation can be given by the adjacency relationship of stable sub-codes, Tanner graph connection relationship, physical distance relationship or other equivalent topological relationship; temporal processing can be implemented by recursive state update, temporal convolution, temporal attention, gated fusion or a combination thereof; spatial processing can be implemented by local message passing, convolutional mixing, global attention, attention with distance bias or a combination thereof. After several layers of spatiotemporal joint processing, a feature representation for dual-task prediction is obtained, where the representation corresponding to each data qubit is used for round-by-round qubit loss prediction, while the representation of the final round or after temporal aggregation is used for logical state or logical error prediction.

[0042] The dual-task output module includes a first output head and a second output head. The first output head is the qubit loss output head, used to provide a binary classification result or corresponding probability for each data qubit in each round, indicating whether the data qubit is in a lost state in the corresponding round. The second output head is the logic output head, used to predict the logic state or logic error in the final round. The logic output head can generate prediction results for several logic lines separately, and then synthesize them to obtain the final logic decision. Through this dual-task design, the decoder learns the spatiotemporal location of qubit loss while learning logic decoding, thus enabling the two tasks to complement each other. The qubit loss location result can not only serve as diagnostic information, but also, in turn, help improve the accuracy of logic decoding.

[0043] As a preferred embodiment, the spatiotemporal feature extraction module adopts a recurrent neural network structure or a deep spatiotemporal graph neural network structure.

[0044] Please refer to Figure 4, Figure 4 This is one of the schematic diagrams illustrating the principle of the spatiotemporal feature extraction module provided by the present invention.

[0045] In this embodiment, the spatiotemporal feature extraction module adopts a recurrent neural network (RNN) structure. The RNN structure mainly includes a syndromic attention module. The syndromic attention module comprises a self-attention module, a feedforward network, and a convolutional hybrid network. First, a stabilizer embedder maps the stabilizer measurement results and detection events of each round to a high-dimensional representation of the auxiliary qubits, then recursively updates this representation by combining it with the hidden state from the previous round. Next, the syndromic attention module processes the auxiliary qubit representation. Subsequently, a readout module converts the auxiliary qubit representation into a data qubit representation. Finally, based on the data qubit representation, a data bit loss prediction head and a logical prediction head are respectively connected. Specifically, the self-attention module can be used to capture global spatial correlations, the feedforward network can provide a nonlinear mapping function and enhance expressive power through high-dimensional projection, the convolutional hybrid network can extract local and medium-scale structures, and the recursive update part can accumulate multi-round temporal information.

[0046] As a preferred embodiment, the deep spatiotemporal graph neural network structure includes multiple cascaded deep network modules; each deep network module includes a graph neural network module, a temporal mixing module, and a spatial attention module; the method further includes: extracting spatial local features from the embedded features output by the feature embedding module through the graph neural network module; performing deep feature interaction in the temporal dimension on the spatial local features output by the graph neural network module through the temporal mixing module; and performing global spatial correlation on the temporal mixing features output by the temporal mixing module through the spatial attention module to output deep spatiotemporal extracted features.

[0047] Please refer to Figure 5 , Figure 5 The second schematic diagram illustrates the principle of the spatiotemporal feature extraction module provided by this invention.

[0048] In this embodiment, the spatiotemporal feature extraction module employs a deep spatiotemporal graph neural network structure for dual-task prediction. The temporal inputs from frame 0 to frame T are first mapped to high-dimensional features using a stable sub-embedder. These features then enter a deep network module containing N layers of cascaded processing. Within each layer, the features of each time frame are first extracted independently using a graph neural network for spatial local feature extraction. This step can establish edge connections based on the physical connections between qubits or the Tanner graph to transmit local structural information between nodes. Next, the output features of all frames processed by the graph neural network are uniformly and parallelly fed into the temporal mixing module. This module uses a fusion of one-dimensional convolution and gated attention to simultaneously capture local short-range temporal correlations and long-range temporal dependencies spanning multiple frames, achieving deep feature interaction in the temporal dimension. Subsequently, the temporally mixed features enter the spatial attention module, where global spatial correlation calculations further enhance the network's ability to identify relevant error clusters and flicker patterns. After N layers of deep spatiotemporal feature extraction and updating, the output features of each frame are connected to the corresponding data bit loss prediction header to complete the bit loss state prediction for each time frame; at the same time, the logic prediction header is connected based on the output features of the network in the Tth frame to achieve accurate prediction of logic errors.

[0049] The two types of network implementations described above differ in structure, but both revolve around the same technical concept: spatiotemporal joint modeling of multi-round corrector historical information, and simultaneous logic decoding and qubit loss identification through dual output heads. The main improvement of this invention lies in the overall technical approach of multi-round historical input, delayed feedback processing, spatiotemporal feature extraction, and dual-task joint output.

[0050] As a preferred embodiment, it further includes: employing a delayed feedback mechanism to confirm the final result of qubit loss after a preset number of rounds; the delayed feedback mechanism uses a sliding window or overlapping window method to confirm the result.

[0051] Since the identification of lost qubits relies on multiple rounds of observation, this invention employs a delayed feedback mechanism. Specifically, after receiving several consecutive rounds of stabilizer measurement results, the system does not require an immediate final confirmation of qubit loss in each round. Instead, it allows sufficient time evidence to accumulate before outputting a high-confidence decision. For loss events occurring just at the end of the window, the flickering pattern may not be immediately identified within the current window because it has not yet fully unfolded. To address this, a sliding window or overlapping window approach can be used, allowing the event to accumulate more collimator history in subsequent windows before being confirmed. Furthermore, the confirmation results obtained in subsequent windows can also correct the logic decoding judgments near the end of the previous window, thus balancing real-time performance and accuracy in continuous quantum error correction.

[0052] This invention employs supervised learning to train the decoder. Training samples can be generated by a stable subcircuit simulator, and each sample includes at least a multi-round corrector history, a logic label, and a qubit loss label. The logic label represents the logic state or logic operator value corresponding to the sample, and the qubit loss label represents the actual loss position of each data qubit in each round. During training, the model parameters are optimized using a multi-task loss function, enabling the model to simultaneously learn the logic decoding task and the qubit loss recognition task. Preferably, training data can be continuously generated in a streaming mode to improve sample diversity and reduce the risk of overfitting.

[0053] The training can employ a circuit-level noise model incorporating Pauli noise, measurement noise, and qubit loss noise. Auxiliary qubit loss can be considered as being refreshed after each round of measurement, thus its impact is primarily limited to a single round. Data qubit loss, however, is persistent; once lost, it will no longer participate in normal gate operations in subsequent rounds, and its effects will accumulate until it is recognized by the decoder and recovered by hardware. This setting more closely aligns with the practical problem this invention aims to solve: achieving joint decoding and localization based solely on multi-round collimator historical information under conditions of continuous data qubit loss.

[0054] The training objective employs a multi-task loss function. Let the logistic prediction loss be... The data bit loss prediction loss is The total loss can then be written as: , in, and , which is a weighting coefficient used to balance the convergence speed and optimization level of the logic decoding task and the quantum bit loss detection task. and These are the logit scores (logit values) of the actual logistic value and the logistic value predicted by the neural network. and These are the actual loss result and the network-predicted logit score of the loss, respectively. Preferably, both of the aforementioned loss terms can be cross-entropy loss. Alternatively, other loss functions suitable for classification tasks can be used. After training, only multi-round corrector history information needs to be input during the inference phase to simultaneously obtain the logic decoding result and the qubit loss position prediction result.

[0055] The present invention also includes: generating targeted recovery actions based on the quantum bit loss prediction results; and using the identified loss location information for updating the logical results of the current or subsequent windows.

[0056] The present invention also includes: performing supplementation, replacement, reinitialization or equivalent recovery operations on the data qubits that are determined to be lost, based on the qubit loss prediction results.

[0057] The following describes the quantum bit loss detection and quantum error correction system provided by the present invention. The quantum bit loss detection and quantum error correction system described below can be referred to in correspondence with the quantum bit loss detection and quantum error correction method described above.

[0058] Please refer to Figure 6 , Figure 6 This is a schematic diagram of the structure of a quantum bit loss detection and quantum error correction system provided by the present invention.

[0059] The present invention also provides a quantum bit loss identification and quantum error correction system, comprising: a first module 601 for extracting spatiotemporal correlation features from the obtained multi-round stable sub-measurement sequence; and a second module 602 for identifying a persistent random flickering pattern caused by quantum bit loss based on the spatiotemporal correlation features, and determining the logic decoding result and the quantum bit loss prediction result.

[0060] Existing traditional decoding schemes based on stable subcodes are primarily designed for Pauli errors. Their basic approach relies on local flip patterns in the stabilizer measurement results or the resulting event graph, combined with a pre-defined noise model to infer the most probable error chain and output the logical error correction result. This type of scheme performs well in handling instantaneous X, Y, and Z Pauli errors, but it typically requires that the physical qubits remain within the encoding space. When a data qubit is lost, the X-type and Z-type stabilizers previously involved by the lost qubit will exhibit anti-commutation relations in their neighborhoods. This causes the related stabilizer measurement results to randomly collapse between +1 and -1 in subsequent measurement rounds, forming a persistent random flickering characteristic. This phenomenon differs from the temporally localized, predictable flip patterns corresponding to ordinary Pauli errors, making existing traditional decoding schemes difficult to apply directly.

[0061] Furthermore, bit loss in a single-round stabilizer measurement may resemble ordinary Pauli noise, making it difficult to distinguish the error type. Existing traditional decoding algorithms typically rely on the corrector flipping pattern within a single round or a finite local range for judgment, while the random flickering characteristics caused by bit loss only become apparent after continuous observation over multiple rounds of stabilizer measurements. Therefore, decoding methods that rely solely on a single round or lack long-term temporal correlation analysis capabilities are prone to misclassifying bit loss as ordinary Pauli errors or high-frequency Pauli noise as loss-related anomalies, thereby reducing the accuracy of logical decoding. In other words, existing technologies lack an effective mechanism that can reliably distinguish between ordinary Pauli errors and bit loss errors based solely on the history of multiple rounds of corrector measurements.

[0062] Specifically, the system includes at least: a measurement data acquisition module for acquiring the measurement results and detection events of each round of stabilizers; a history window construction module for organizing and forming multi-round corrector history information (multi-round stabilizer measurement sequence); an artificial intelligence decoding module for extracting spatiotemporal features from the multi-round corrector history information and outputting logical decoding results and qubit loss prediction results; a recovery control module for generating targeted recovery actions based on the qubit loss prediction results; and a result update module for using the identified loss location information to update the logical results of the current or subsequent windows. The aforementioned modules can be implemented in software within an electronic device, or they can be implemented collaboratively by software and hardware.

[0063] This system can recover data qubits that are predicted to be lost, based on the qubit loss prediction results, by performing supplementation, replacement, reinitialization, or equivalent recovery operations on the control module. Even if some of the predicted lost qubits actually only exhibit high-frequency Pauli noise, as long as their collimator behavior is highly similar to the flickering pattern caused by the loss, reinitialization can still reduce local entropy accumulation and suppress error propagation to some extent. Therefore, this type of output still has practical control value. By combining the decoding results with the recovery actions in a closed loop, this invention not only achieves logical error correction but also enables hardware-oriented diagnosis and intervention.

[0064] The above are merely preferred embodiments of the present invention. The present invention is not limited to rotating surface codes, but can also be extended to other stable subcodes, other neutral atoms or superconducting quantum hardware platforms, and other quantum error correction tasks capable of generating multiple rounds of collimator history; nor is it limited to the two types of neural network structures mentioned above. As long as it can extract spatiotemporal correlation features based on multiple rounds of collimator history information and simultaneously output logical decoding results and qubit loss identification results, it falls within the protection scope of the present invention.

[0065] The key points of this invention are as follows: The first key point of this invention is that it no longer treats qubit loss simply as a special case of ordinary Pauli error, but rather captures the characteristic that qubit loss produces persistent random flickering during multiple rounds of stabilizer measurements. Ordinary Pauli errors typically correspond to discrete, local, and predictable collimator flips, while qubit loss continuously causes random fluctuations in the relevant stabilizer results across multiple measurement rounds. Therefore, this invention does not rely on a single round of collimator patterns for judgment, but rather utilizes the temporal correlation characteristics in the historical information of multiple rounds of collimators to distinguish between Pauli errors and qubit loss. This is the core inventive point that differentiates this invention from traditional decoding schemes.

[0066] The second key aspect of this invention lies in its use of a delayed feedback decoding mechanism. Since the flickering characteristics of a lost qubit require multiple rounds of accumulation before they can be identified with high confidence, this invention allows the decoder to collect several rounds of stable sub-measurement information before making a final judgment, rather than requiring an immediate final decision in each round. For qubit loss events occurring just after the current window, evidence can be accumulated through subsequent sliding or overlapping windows, and the logical judgment at the end of the previous window can be updated. This mechanism enables this invention to balance recognition accuracy and system operability in continuous quantum error correction scenarios.

[0067] The third key aspect of this invention lies in unifying logic decoding and qubit loss location identification within the same artificial intelligence decoding framework, simultaneously completing both tasks through a dual-task joint output structure. Specifically, the decoder outputs, on the one hand, the prediction result of the final logic state or logic operator error, and on the other hand, the prediction result of whether each round and each data qubit has been lost. Compared with existing decoders that only output logic error correction results, this invention not only improves logic decoding performance but also provides loss location diagnostic information with practical control significance, thereby forming a more complete quantum error correction processing capability.

[0068] The fourth key point of this invention is that the qubit loss identification results can be directly used to guide subsequent hardware recovery operations, thus forming a closed-loop mechanism combining decoding and recovery. The identified lost qubit locations can be used to supplement, replace, or reinitialize the corresponding qubits, reducing the continuous damage to logical information caused by qubit loss. Even if some marked locations actually correspond to high-frequency Pauli noise rather than actual loss, as long as their statistical behavior is close to the flickering pattern caused by loss, reinitializing them helps reduce local entropy accumulation and suppress error propagation. Therefore, this invention does not merely provide an offline analysis algorithm, but rather an online diagnostic and intervention scheme that can serve continuous quantum error correction operations.

[0069] The fifth key point of this invention lies in its core technical idea, which does not rely on a specific neural network structure but rather possesses good architectural generalization. Whether employing a recurrent neural network based on a recursive structure and attention mechanism, or a graph neural network structure with interleaved spatiotemporal processing, as long as it can perform spatiotemporal joint modeling of multi-round scintillation history information and output logical decoding results and qubit loss prediction results, the core idea of ​​this invention can be realized. The fact that different network architectures achieved similar performance in experiments indicates that the true innovation of this invention is not limited to any specific model detail, but lies in the overall technical approach of delay feedback joint decoding based on multi-round scintillation characteristics.

[0070] In summary, the essential key points of this invention can be summarized as follows: identifying qubit loss flickering characteristics based on multi-round corrector historical information; improving the reliability of loss identification by adopting a delayed feedback mechanism; simultaneously realizing logic decoding and qubit loss localization through dual-task joint output; and using the identification results to guide subsequent recovery operations, thereby improving the overall performance of the quantum error correction system in a qubit loss noise environment without excessively increasing hardware complexity.

[0071] The beneficial effects of this invention are as follows: Compared with existing technologies, this invention can significantly improve logic decoding performance in quantum error correction scenarios where qubits are lost. In the rotating surface code storage task and under a circuit-level noise model including Pauli noise, measurement noise, and qubit loss noise, the delayed feedback AI decoder employed in this invention significantly outperforms the traditional minimum weight perfect matching decoder in terms of logic accuracy, and also outperforms the delayed erasure MWPM decoder with spatial prior information about lost qubits. This result demonstrates that this invention can not only utilize multi-round corrector history information to identify qubit loss, but this identification capability can also, in turn, improve the logic decoding effect.

[0072] This invention also enables high-precision qubit loss localization. In one embodiment, after 10 rounds of quantum error correction, the STGNN decoder achieves a recall of 0.654 and a precision of 0.845 for identifying lost data qubits; the recurrent neural network decoder achieves a recall of 0.652 and a precision of 0.856. These results demonstrate that this invention can effectively identify the spatiotemporal location of lost qubits without relying on additional loss marker input, thus providing a reliable basis for subsequent recovery operations.

[0073] This invention further possesses strong hardware recovery support capabilities. Since it can output the location of lost qubits, targeted reinitialization, supplementation, or replacement operations can be performed on the corresponding data qubits, thereby reducing the damage to logical information caused by qubit loss. Related results also show that even if some qubits predicted to be "lost" actually still exist, they are usually located in the high-frequency Pauli noise region, and their statistical behavior is similar to the flickering characteristics caused by loss. Performing reinitialization on these qubits still helps reduce local entropy accumulation and suppress error propagation; therefore, the output results of this invention have direct engineering application value.

[0074] This invention also has the advantage of requiring no additional hardware detection overhead. It constructs inputs using only the multi-round stabilizer measurement results and detection events already present in the standard quantum error correction process, without relying on additional probe beams, additional teleportation detection circuits, or additional erasure / transformation steps, thus simultaneously completing logic decoding and qubit loss identification. Therefore, this invention is more suitable for deployment in continuous quantum error correction and complex quantum algorithm operating environments.

[0075] This invention also demonstrates good architectural versatility and robustness. Using recurrent neural networks and STGNN networks with different inductive biases, similar results were achieved in both logical accuracy and qubit loss detection metrics. The effectiveness of this invention lies in its technical approach of extracting qubit loss flickering features based on multi-round collimator historical information and performing joint decoding with delay feedback. This feature facilitates the transfer and application of this invention across different quantum hardware platforms and different artificial intelligence implementation structures.

[0076] In summary, this invention can simultaneously achieve high-precision logic decoding and qubit loss location without excessively increasing hardware complexity, thereby improving the logic reliability, diagnostic capability, and continuous operation applicability of quantum error correction systems in qubit loss noise environments.

[0077] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown, the electronic device may include a processor 701, a communications interface 702, a memory 703, and a communication bus 704. The processor 701, communications interface 702, and memory 703 communicate with each other via the communication bus 704. The processor 701 can call logic instructions in the memory 703 to execute a qubit loss identification and quantum error correction method. This method includes: extracting spatiotemporal correlation features from the obtained multi-round stable quantum measurement sequence; identifying the persistent random flickering pattern caused by qubit loss based on the spatiotemporal correlation features; and determining the logic decoding result and the qubit loss prediction result.

[0078] Furthermore, the logical instructions in the aforementioned memory 703 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0079] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the qubit loss identification and quantum error correction methods provided by the above methods. The method includes: extracting spatiotemporal correlation features from the obtained multi-round stable sub-measurement sequence; identifying the persistent random flickering pattern caused by qubit loss based on the spatiotemporal correlation features; and determining the logical decoding result and the qubit loss prediction result.

[0080] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the qubit loss identification and quantum error correction methods provided by the above methods. The method includes: extracting spatiotemporal correlation features from the obtained multi-round stable sub-measurement sequence; identifying a persistent random flickering pattern caused by qubit loss based on the spatiotemporal correlation features; and determining the logical decoding result and the qubit loss prediction result.

[0081] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0082] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0083] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting lost qubits and quantum error correction, characterized in that, include: Spatiotemporal correlation features are extracted from the obtained multi-round stable sub-measurement sequences; Based on the spatiotemporal correlation characteristics, the persistent random flickering pattern caused by the loss of qubits is identified, and the logical decoding result and the qubit loss prediction result are determined.

2. The method for qubit loss detection and quantum error correction according to claim 1, characterized in that, The extraction of spatiotemporal correlation features from the obtained multi-round stable sub-measurement sequences includes: Obtain the T+1 rounds of stabilizer measurement sequence; where the first T rounds come from stabilizer measurements and the last round comes from data qubit readout; T is an integer not less than 2; The temporal and spatial features of the T+1 round stabilizer measurement sequence are extracted using a decoder.

3. The quantum bit loss detection and quantum error correction method according to claim 2, characterized in that, The acquisition of the stable sub-measurement sequence for round T+1 includes: The logical qubits are encoded and subjected to multiple rounds of stabilizer measurements to obtain a T+1 round stabilizer measurement sequence.

4. The method for qubit loss detection and quantum error correction according to claim 3, characterized in that, Also includes: The flickering counting characteristics are determined by counting the detection events corresponding to each auxiliary qubit; The flicker count feature is added to the input representation of the decoder; The flicker counting feature is used to distinguish continuous random flickering patterns.

5. The quantum bit loss detection and quantum error correction method according to claim 2, characterized in that, The decoder includes a feature embedding module, a spatiotemporal feature extraction module, and a dual-task output module; the method further includes: The feature embedding module is used to embed features into the T+1 round stable sub-measurement sequence; The spatiotemporal feature extraction module extracts spatiotemporal features from the embedded features output by the feature embedding module. The dual-task output module outputs the logical decoding result and the quantum bit loss prediction result based on the spatiotemporal features output by the spatiotemporal feature extraction module.

6. The quantum bit loss detection and quantum error correction method according to claim 5, characterized in that, Also includes: Through the feature embedding module, based on the embedding features of the stable sub-measurement sequence, auxiliary qubit index embedding features, round type embedding features, node type embedding features, or task type embedding features are superimposed.

7. The quantum bit loss detection and quantum error correction method according to claim 5, characterized in that, The spatiotemporal feature extraction module adopts a recurrent neural network structure or a deep spatiotemporal graph neural network structure.

8. The method for qubit loss detection and quantum error correction according to claim 7, characterized in that, The deep spatiotemporal graph neural network structure includes multiple cascaded deep network modules; each deep network module includes a graph neural network module, a temporal mixing module, and a spatial attention module; the method further includes: Spatial local feature extraction is performed on the embedded features output by the feature embedding module through the graph neural network module. The temporal mixing module performs deep feature interaction in the temporal dimension on the spatial local features output by the graph neural network module. The spatial attention module performs global spatial correlation on the temporal blending features output by the temporal blending module to output deep spatiotemporal extraction features.

9. The quantum bit loss detection and quantum error correction method according to any one of claims 1 to 8, characterized in that, Also includes: A delayed feedback mechanism is adopted to confirm the final result of the loss of qubits after a preset number of rounds; The delayed feedback mechanism uses a sliding window or overlapping window method to confirm the result.

10. A quantum bit loss detection and quantum error correction system, characterized in that, include: The first module is used to extract spatiotemporal correlation features from the obtained multi-round stable sub-measurement sequences; The second module is used to identify the persistent random flickering pattern caused by the loss of qubits based on the spatiotemporal correlation characteristics, and to determine the logic decoding result and the qubit loss prediction result.

Images