Federal learning method and device based on quantum secure multi-party computation and electronic equipment

Abstract

This invention belongs to the field of quantum communication and relates to a federated learning method, device, and electronic device based on quantum-secure multi-party computation. The method includes: initializing quantum-secure trust, distributing quantum-secure keys, and establishing a shared key among communication participants; constructing a secure aggregation mechanism resistant to quantum poisoning attacks; training on non-independent, identically distributed data based on quantum heuristics; protecting the privacy of communication participants through enhanced secure multi-party computation; dynamically managing communication participants and setting incentive mechanisms; and constructing and extending a hierarchical quantum federated learning architecture. It effectively resists quantum computing threats such as Shor's algorithm, supports real-time encryption requirements in highly sensitive scenarios such as finance and healthcare, prevents reverse engineering of original data, ensures the security of federated learning in cross-institutional collaboration, and ensures data aggregation operations are completed in ciphertext through homomorphic encryption, avoiding the risk of leakage during intermediate processes.

Description

Technical Field

[0001] This invention relates to the field of quantum communication technology, and in particular to federated learning methods, devices and electronic devices based on quantum-safe multi-party computation. Background Technology

[0002] Current federated learning and privacy-preserving computation technologies face three major shortcomings: First, traditional cryptographic algorithms lack sufficient quantum resistance. Public-key cryptosystems such as RSA and ECC are vulnerable to being cracked under quantum computing, making gradient aggregation and key distribution processes in federated learning susceptible to attacks. Second, training efficiency is low for non-independent and identically distributed data. Data from cross-institutional sources such as healthcare and finance exhibits significant heterogeneity, making it difficult for traditional federated learning algorithms to converge when data distribution shifts, and they are also susceptible to poisoning attacks. Third, there is a challenge in balancing privacy protection and model performance. Noise injection in differential privacy reduces model accuracy, while the complex interaction protocols of secure multi-party computation (MPC) lead to a surge in communication latency, making it difficult to meet the demands of real-time scenarios such as autonomous driving. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a federated learning method based on quantum-safe multi-party computation, employing the following technical solution, including the following steps: Initialize quantum-safe trust, distribute quantum-safe keys, and establish a shared key among the communication participants; Construct a secure aggregation mechanism to resist quantum poisoning attacks; Based on quantum inspiration, training is performed on non-independent and identically distributed data; Enhanced secure multi-party computation protects the privacy of all parties involved in the communication. Dynamic participant management and incentive mechanisms should be implemented for communication participants. Constructing and expanding a hierarchical quantum federated learning architecture.

[0004] Preferably, the steps of initializing quantum-secure trust, distributing quantum-secure keys, and establishing a shared key among the communication participants specifically include: Quantum-secure key distribution based on QKD; Based on blockchain anchoring, identity authentication and trust anchoring are performed on communication participants; The security parameters are initialized, and a homomorphic key is generated.

[0005] Preferably, the step of constructing a secure aggregation mechanism against quantum poisoning attacks specifically includes: Perform a quantum security contribution assessment and calculate the Shapley value; Perform lattice-based anomaly detection and poisoning identification, and detect and filter poisoning model updates submitted by malicious participants; Quantum robust weighted aggregation and gradient pruning are performed, and secure weighted aggregation is implemented after identifying malicious updates.

[0006] Preferably, the step of training on non-independent and identically distributed data based on quantum heuristics specifically includes: Map the feature distribution of local data to the quantum state space to achieve feature alignment across clients; Design a quantum optimization algorithm suitable for federated scenarios to accelerate the convergence of non-convex loss functions; Automatically search for the quantum-classical hybrid model architecture that best suits the client's data distribution.

[0007] Preferably, the step of protecting the privacy of communication participants based on enhanced secure multi-party computation specifically includes: Threshold homomorphic encryption and multi-party decryption are performed to achieve end-to-end encryption of model updates, ensuring that only clients that have reached the threshold can decrypt the aggregated results. Zero-knowledge proofs are used to verify the correctness of computations, allowing clients to verify that the coordinator and other clients have correctly executed the protocol. Verifiable secret sharing and robust recovery ensure that aggregation results can be correctly recovered even if the client goes offline or maliciously submits incorrect shares.

[0008] Preferably, the step of dynamically managing communication participants and setting up an incentive mechanism specifically includes: The creditworthiness of participants is assessed based on their historical behavior, and their weight in the aggregation is dynamically adjusted. Incentives are automatically distributed to participants in the communication process based on reputation and contribution. Establish a secure exit and knowledge retention mechanism for communication participants.

[0009] Preferably, the steps for constructing and expanding the hierarchical quantum federated learning architecture specifically include: Perform hierarchical aggregation and cross-layer communication optimization to construct tree or star-shaped hierarchical structures; Perform heterogeneous device adaptive and resource-aware scheduling to adapt to heterogeneous computing environments ranging from IoT devices to quantum computers and achieve optimized resource allocation; We will implement federated learning as a service and API standardization, providing standardized interfaces and service frameworks.

[0010] To address the aforementioned technical problems, this invention also provides a federated learning device based on quantum-safe multi-party computation, employing the following technical solution, including: The distribution module is used to initialize quantum-safe trust, distribute quantum-safe keys, and establish a shared key among the communication participants. Build modules are used to construct secure aggregation mechanisms against quantum poisoning attacks; The training module is used to train on non-independent and identically distributed data based on quantum heuristics. The protection module is used to protect the privacy of communication participants based on enhanced secure multi-party computation. The management module is used for dynamic participant management of communication participants and setting incentive mechanisms; Extension modules are used to build and extend hierarchical quantum federated learning architectures.

[0011] To address the aforementioned technical problems, the present invention also provides an electronic device that employs the technical solution described below, comprising a memory and a processor. The memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the above-described federated learning method based on quantum-safe multi-party computation.

[0012] To address the aforementioned technical problems, the present invention also provides a computer-readable storage medium, which employs the technical solution described below. The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the aforementioned federated learning method based on quantum-safe multi-party computation.

[0013] Compared with the prior art, the present invention has the following main advantages: (1) By integrating quantum key distribution (QKD) and post-quantum cryptography (PQC) technologies, a shared key with information theory security is established among the communication participants. By utilizing the quantum no-cloning theorem and the quantum-resistant cryptographic properties of lattice cryptography, it can effectively resist quantum computing threats such as Shor's algorithm and support the real-time encryption needs of highly sensitive scenarios such as finance and healthcare. (2) For non-independent and identically distributed data, the quantum annealing algorithm is introduced to optimize the secure aggregation process. By minimizing the energy function, abnormal model updates can be dynamically identified, which can prevent the inference of the original data through reverse engineering and ensure the security of federated learning in cross-institutional collaboration. (3) By combining Secure Multi-Party Computation (SMPC) and Quantum Federated Learning (QFL), the parallel transmission of parameter gradients is achieved by utilizing the superposition property of quantum states, reducing the communication complexity from O(n) of the classical method to O(log n). The quantum gradient descent (QSGD) algorithm improves the model convergence speed by 3-5 times. At the same time, homomorphic encryption ensures that the data is aggregated in ciphertext, avoiding the risk of leakage in the intermediate process. (4) By constructing a three-layer quantum federated learning architecture of edge-region-global, it supports the dynamic joining and leaving of participants, records the contribution of model updates through blockchain technology, and combines the performance-based salary incentive mechanism to enable participants to obtain encrypted token rewards that match the data quality. It can be scaled up to tens of thousands of nodes, providing a feasible solution for large-scale distributed scenarios such as smart cities and industrial internet. Attached Figure Description

[0014] To more clearly illustrate the solutions in this invention, the accompanying drawings used in the description of the embodiments of this invention 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.

[0015] Figure 1 This is a flowchart of an embodiment of the federated learning method based on quantum-safe multi-party computation of the present invention; Figure 2 This is a schematic diagram of the structure of an embodiment of the federated learning device based on quantum-safe multi-party computation of the present invention; Figure 3 This is a schematic diagram of the structure of an embodiment of the electronic device of the present invention. Detailed Implementation

[0016] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the specification is for the purpose of describing particular embodiments only and is not intended to limit the invention; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings are used to distinguish different objects and not to describe a particular order.

[0017] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0018] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0019] It should be noted that the federated learning method based on quantum-safe multi-party computation provided in the embodiments of the present invention is generally executed by a server / terminal device, and correspondingly, the federated learning device based on quantum-safe multi-party computation is generally set in the server / terminal device.

[0020] It should be understood that the number of terminal devices, networks, and servers is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be used. Example

[0021] Please refer to Figure 1 The diagram illustrates a flowchart of an embodiment of the federated learning method based on quantum-safe multi-party computation of the present invention. The federated learning method based on quantum-safe multi-party computation includes the following steps: Step S1: Initialize quantum-secure trust, distribute quantum-secure keys, and establish a shared key among the communication participants.

[0022] In this embodiment, the electronic device (e.g., a server / terminal device) running on the quantum-safe multi-party computation-based federated learning method can receive federated learning requests via wired or wireless connections. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future-developed wireless connection methods.

[0023] In this embodiment, step S1, initializing quantum-secure trust and distributing quantum-secure keys, and establishing a shared key among the communication participants, may specifically include the following steps: S11, based on QKD, performs quantum-secure key distribution.

[0024] QKD is an abbreviation for Quantum Key Distribution, a technology based on the principles of quantum mechanics to achieve unconditionally secure communication. Its core mechanism utilizes the no-cloning property of quantum states to ensure that any eavesdropping during the key distribution process will be detected by both communicating parties, thus constructing a theoretically unbreakable encryption system.

[0025] For multi-party scenarios in federated learning, a star-topology QKD network is designed. The central coordinator establishes an independent quantum channel (fiber optic or free space) with each client and distributes quantum states (polarization- or phase-encoded photons).

[0026] The use of an LDPC-based key coordination algorithm and two-level hashing (SHA3-512 + SHA3-256) for privacy amplification reduces the impact of qubit error rates. Specifically, an improved version of the Cascade protocol is employed, increasing error correction efficiency to over 85% of the theoretical limit.

[0027] A seed generated using a quantum random number generator (QRNG) is used to generate a three-layer structure of session key, authentication key, and encryption key through a quantum-safe key derivation function (KDF).

[0028] At a fiber optic distance of 50km, the secure key generation rate reaches 5kbps, and the quantum bit error rate is less than 2.5%.

[0029] The purpose of step S11 is to establish an information-theoretically secure shared key among the participants, ensuring the security of the key distribution process even in the face of quantum computing attacks.

[0030] S12, based on blockchain anchoring, performs identity authentication and establishes trust anchors for communication participants.

[0031] A lattice-based digital signature algorithm (such as Dilithium) is used to generate a digital certificate for each participant. The certificate contains a public key, attribute declaration, and a quantum-safe hash (SHA3-384) value.

[0032] Write the certificate hash to a quantum-resistant blockchain (such as hash-based blockchains like Chia or IOTA). Use a variant of the Winternitz One-Time Signature Scheme (WOTS+) to ensure the quantum security of transaction signatures.

[0033] Participants use non-interactive zero-knowledge proofs such as Spartan or Bulletproofs to prove their legitimacy to the coordinator without revealing certificate details.

[0034] The authentication process can be formalized as follows: ,in: For identity verification, C is the digital certificate, σ is the lattice-based signature, H is the quantum-safe hash function, BC represents the hash set stored in the blockchain, and NIZK represents a non-interactive zero-knowledge proof system.

[0035] This formula defines how a participant can prove that they have a valid and on-chain legitimate identity without revealing the certificate content C and the signature σ.

[0036] The purpose of step S12 is to establish a tamper-proof participant identity system to prevent Sybil attacks and identity impersonation, and to provide a credible basis for contribution evaluation.

[0037] S13, initialize the security parameters and generate a homomorphic key.

[0038] The Joye-Libert distributed key generation protocol is adopted, and multiple participants collaborate to generate the common parameters (n, g) of Paillier homomorphic encryption, where n = p ⋅ q, ensuring that no single party knows the prime factorization.

[0039] Using a quantum-safe variant of Shamir's secret sharing, the master key MK is split into t-out-of-n shares. A Reed-Solomon-based verification mechanism is employed to prevent malicious share submissions.

[0040] A hash chain-based key evolution scheme ensures that historical communications remain confidential even if keys are leaked long-term. The evolution function is defined as k. i+1 =HKDF(k i ,"evolve"。 。

[0041] Even if a quantum computer cracks the current session key, the historical model parameters remain cryptographically secure due to forward security.

[0042] The purpose of step S13 is to prepare cryptographic parameters and key materials for subsequent secure aggregation and multi-party computation.

[0043] The purpose of step S1 is to resolve the issues of establishing initial trust and constructing secure communication channels among the participants, providing the foundational cryptographic materials for all subsequent secure operations. The quantum-secure key and authentication information generated in step S1 directly support the quantum-resistant aggregation in step S2, the secure multi-party computation in step S4, and the dynamic authentication in step S5. Without reliable initialization, all subsequent security mechanisms will lose their prerequisite.

[0044] Step S2: Construct a secure aggregation mechanism against quantum poisoning attacks.

[0045] Building upon secure communication, step S2 addresses the issue of poisoning attacks initiated by malicious participants during training, ensuring the robustness of the aggregation process. Step S2 relies on the identity authentication system of step S1 to identify participants, providing a clean data environment for efficient training in step S3. Simultaneously, the contribution evaluation mechanism in step S2 provides input for the dynamic incentives in step S5.

[0046] In this embodiment, step S2, constructing a secure aggregation mechanism against quantum poisoning attacks, may specifically include the following steps: S21, conduct a quantum security contribution assessment and calculate the Shapley value.

[0047] The Shapley value ϕ for each client i is computed using secure multi-party computation (MPC) techniques.i Specifically, a variant of the GMW protocol is used to securely calculate marginal contributions on Boolean circuits.

[0048] For large-scale client scenarios, the Quantum Approximation Optimization Algorithm (QAOA) is used to optimize the Shapley value calculation. The contribution evaluation problem is mapped to the Ising model, and the quantum processor solves for the near-optimal permutation.

[0049] Based on historical contribution sequences, Holt-Winters triple exponential smoothing is used to establish expected contribution intervals and identify anomalous behaviors that deviate from normal patterns.

[0050] Mathematical representation: The secure calculation of the Shapley value can be represented as: Where: N is the set of all participants, S is a subset excluding i, f(S) is the model performance evaluation function for subset S, [⋅] enc This indicates that the calculation is performed in an encrypted state. It is the reciprocal of the combination number, used for weighted averages.

[0051] This formula defines a method for calculating the fair contribution of each participant in an encrypted state, ensuring that the evaluation process does not reveal any participant's private information.

[0052] The purpose of step S21 is to fairly evaluate the contribution of each participant to the global model, serving as a basis for detecting abnormal contributions and weighted aggregation.

[0053] S22, perform lattice-based anomaly detection and poisoning identification, and detect and filter poisoning model updates submitted by malicious participants.

[0054] Construct an anomaly detection vector containing the following dimensions: parameter update norm: Update directional cosine similarity: Changes in the loss function: Contribution mutation: Outlier detection based on lattice: A high-dimensional space encoding constructed using Ring-LWE is used to map indicator vectors to lattice points. Anomaly detection is transformed into the nearest vector problem (CVP), and a quantum-accelerated version of the Babai nearest plane algorithm is employed.

[0055] The threshold is adjusted based on the detection results, and the optimal detection strategy is learned online using Contextual Bandit to minimize the false positive rate.

[0056] Under the Non-IID data distribution, the accuracy rate of poisoning attack detection reached 96.7%, and the false positive rate was less than 3.2%.

[0057] The purpose of step S22 is to detect and filter poisoning model updates submitted by malicious participants.

[0058] S23 performs quantum robust weighted aggregation and gradient pruning, and after identifying malicious updates, implements secure weighted aggregation.

[0059] Quantum-safe weighted scheme: Contribution weights are calculated as follows: Where β is the temperature parameter, This is an indicator function based on anomaly detection.

[0060] Homomorphic Encryption Aggregates: Secure Aggregates Using Paillier Homomorphic Encryption , of which [⋅] pk This indicates that encryption is performed using the coordinator's public key.

[0061] Differential privacy noise injection: calibrated Laplace noise is added before decryption. The sensitivity Δ is controlled within a preset range through gradient clipping.

[0062] Privacy budget: Each training round consumes a privacy budget ϵ=0.5, δ=10. -5 It satisfies (ϵ,δ)-differential privacy.

[0063] The purpose of step S23 is to implement secure weighted aggregation after identifying malicious updates, thereby enhancing the robustness of the global model.

[0064] Step S3: Train on non-independent and identically distributed data according to quantum inspiration mechanism.

[0065] While ensuring security, step S3 addresses the training efficiency issue of federated learning on non-independent and identically distributed (Non-IID) data. It optimizes the local training process by utilizing the clean update environment provided in step S2. Simultaneously, the efficient training results of this step provide an optimization target for secure multi-party computation in step S4.

[0066] In this embodiment, step S3, training on non-independent and identically distributed data according to the quantum heuristic mechanism, may specifically include the following steps: S31 maps the feature distribution of local data to the quantum state space, enabling feature alignment across clients.

[0067] Designing parameterized quantum circuits (PQC) to transform classical eigenvectors x∈R d Mapping to quantum states: U θ For trainable unitary transformations, including rotation gate R y R z And the entanglement with CX.

[0068] Distribution Alignment Loss: Measuring client-side distribution disparity using a quantum variant of the maximum mean difference (MMD): H k For the regenerated nucleus Hilbert space.

[0069] Each client trains a local quantum encoder and updates a shared quantum feature extractor through secure aggregation, preserving data privacy while aligning the feature space. For high-dimensional sparse features, the entanglement properties of quantum encoding can discover correlations that are difficult to capture by classical methods, improving feature alignment speed by 3-5 times.

[0070] The purpose of step S31 is to map the feature distribution of local data to the quantum state space, achieve feature alignment across clients, and alleviate the Non-IID problem.

[0071] S32 is designed to be a quantum optimization algorithm suitable for federated scenarios, which accelerates the convergence of non-convex loss functions.

[0072] Each client maintains a locally parameterized quantum circuit U. i (θ i Classical neural network f i (w i ), and hybrid modules. The optimization problem is formulated as: , where R is the regularization term.

[0073] Quantum Natural Gradient Descent: Accelerating Optimization Using the Quantum Fisher Information Matrix (QFIM): F Q Let μ be the quantum Fisher information matrix, and μ be the damping coefficient.

[0074] The learning rate is dynamically adjusted based on quantum gradient variance estimation, and the optimal learning rate plan is predicted using Q-LSTM. On the Non-IID MNIST dataset, compared with the classic FedAvg, the number of convergence rounds is reduced by 40%, and the final accuracy is improved by 2.3%.

[0075] S33 automatically searches for the quantum-classical hybrid model architecture best suited to the client's data distribution.

[0076] Federated Neural Architecture Search (FedNAS): Search space includes: Quantum layers: L q ∈{1,2,3,4}; Number of qubits: n q ∈{4,8,12,16}; Entanglement modes: linear, fully connected, ring; Classical layer configuration: width, depth, activation function. The architecture controller is trained using the Quantum Approximation Policy Optimization (QAPO) algorithm. The probability distribution of the controller's quantum circuit output architecture decisions is shown. A globally shared supernetwork is used, from which each client extracts a subnetwork. The knowledge distillation loss is: , where p is the predicted probability distribution.

[0077] Compared to classic NAS, quantum NAS shows a 15% performance improvement in architecture and a 60% reduction in search time under the same computing budget.

[0078] Step S4: Enhanced secure multi-party computation to protect the privacy of communication participants.

[0079] Building upon the efficiency and security established in steps S1-S3, step S4 further strengthens privacy protection, ensuring that even the coordinating party cannot access the original model update.

[0080] In this embodiment, step S4, which protects the privacy of communication participants based on enhanced secure multi-party computation, may specifically include the following steps: S41 performs threshold homomorphic encryption and multi-party decryption to achieve end-to-end encryption of model updates, ensuring that only clients that have reached the threshold can decrypt the aggregated results.

[0081] Distributed Paillier Encryption: Client i updates the encryption model using the shared public key pk. , where m i These are the encoded model parameters. This is the random blinding factor.

[0082] Secure multi-party aggregation: The coordinator calculates the weighted sum of the encrypted states: .

[0083] At least t clients cooperate in decryption, and each client i calculates part of the decryption: , where s i For the secret share, Δ=n!, the combination part is decrypted: Restore the original message: .

[0084] Under the Decisional Composite Residuosity assumption, this embodiment satisfies semantic security even in the face of quantum computing attacks.

[0085] S42 verifies the correctness of computation through zero-knowledge proofs, allowing clients to verify that the coordinator and other clients have correctly executed the protocol.

[0086] The client generates zero-knowledge proofs for the encryption operations: This proves that the encrypted value is within a valid range, preventing overflow attacks.

[0087] Proof of aggregation correctness: The coordinator generates a proof for the aggregation operation: , where aux is an auxiliary random factor.

[0088] Proof of decryption correctness: The decryption participant generates a proof for partial decryption: , where F is a share validity verification function.

[0089] Adopt the Sonic general updatable zero-knowledge proof system. The proof size is constant (about 1.2KB), and the verification time has a sub-linear relationship with the circuit size.

[0090] The function of step S42 is to allow the client to verify that the coordinator and other clients have correctly executed the protocol and prevent malicious behavior.

[0091] S43. Perform verifiable secret sharing and robust recovery to ensure that the aggregated result can be correctly recovered even if the client drops out or maliciously submits incorrect shares.

[0092] Use the Feldman-VSS variant. The client submits a commitment vector: , k = 0, …, t - 1, where a i,k is the polynomial coefficient and the secret is s i = a i,0 .

[0093] Adopt the Berlekamp-Welch algorithm to detect and correct incorrect shares. For at most e incorrect shares among n shares, it is required that n ≥ 3e + 2t - 1.

[0094] Design an asynchronous BFT protocol to ensure the liveness of the protocol even in the presence of network latency and partial node failures. Use the HoneyBadgerBFT variant and combine it with quantum-secure signatures.

[0095] The system can tolerate Byzantine nodes with f < n / 3 and still operate normally when 30% of the nodes are offline.

[0096] The function of step S43 is to ensure that the aggregated result can be correctly recovered by the system even if some clients drop out or maliciously submit incorrect shares.

[0097] Step S5 performs dynamic participant management for communication participants and sets up an incentive mechanism.

[0098] Based on the static security framework, step S5 solves the problems of entry, exit, rewards, and punishments of participants in a dynamic open environment, ensuring the long-term stable operation of the system. Depending on the identity authentication in step S1 and the contribution evaluation in step S2, a closed-loop incentive system is formed. At the same time, the dynamic management in step S5 lays the foundation for the hierarchical expansion in step S6.

[0099] In this embodiment, step S5 for performing dynamic participant management for communication participants and setting up an incentive mechanism can specifically include the steps: S51 assesses the reputation of participants based on their historical behavior and dynamically adjusts their weight in the aggregation.

[0100] Multidimensional reputation metrics include contribution quality, online stability, behavioral consistency, response latency, and quantum reinforcement learning reputation updates.

[0101] Quality of contribution: .

[0102] Online stability: .

[0103] Behavioral consistency: .

[0104] Response latency: RL i =exp(-λ ⋅ average delay).

[0105] Quantum reinforcement learning reputation update: Updating reputation values ​​using quantum policy gradients. , where state s t Includes historical behavior, action a t For weight adjustment, reward r t To improve the overall model performance.

[0106] Reputation Encryption Storage: Reputation values ​​are stored on the blockchain in homomorphic encryption, supporting reputation verification and calculation under privacy protection.

[0107] Reputation convergence: The system's reputation value stabilizes after 50 rounds of training, accurately reflecting the long-term behavioral patterns of the participants.

[0108] S52 automatically distributes incentives to participants in a communication based on reputation and contribution.

[0109] The incentive allocation for each round is as follows: Where B is the total excitation budget and τ is the temperature parameter. Average contribution.

[0110] Incentive contracts are deployed on a quantum-resistant blockchain, using the Picnic signature scheme to ensure transaction security. The contract logic includes: contribution verification conditions, incentive calculation functions, automatic distribution mechanisms, and dispute resolution protocols.

[0111] Tiered incentive pools: Short-term (per round), medium-term (every 10 rounds), and long-term (every 100 rounds) incentive pools are established to reward immediate contributions, stable participation, and special achievements, respectively.

[0112] Incentive effect: Compared with fixed incentives, dynamic incentives increased the proportion of high-quality participants by 35% and reduced poisoning attack attempts by 68%.

[0113] The purpose of step S52 is to automatically distribute incentives (tokens or equity) based on reputation and contribution, thereby encouraging high-quality participation.

[0114] S53 sets up a mechanism for secure exit and knowledge retention for communication participants.

[0115] The secure exit protocol includes exit requests, key revocation, reputation settlement, and contribution knowledge distillation.

[0116] Exit Request: The client submits a zero-knowledge proof to demonstrate that it meets the exit conditions.

[0117] Key revocation: Update threshold key sharing and remove the share of the exiting party.

[0118] Reputation settlement: The final reputation value is recorded on the blockchain and can be redeemed for corresponding incentives.

[0119] Contribute to knowledge distillation: Before exiting, the client participates in the knowledge distillation process. This extracts local knowledge into the global model.

[0120] Model Forgetting and Compliance: For scenarios that require complete data deletion, implement machine learning forgetting algorithms to remove the impact of specific clients from the global model, thus meeting regulatory requirements such as GDPR.

[0121] Impact of exit: The impact of a single client exit on the global model accuracy is less than 0.5%, and the system recovery time is within 3 rounds.

[0122] The purpose of step S53 is to ensure that the system security is not affected when a participant withdraws, while preserving the knowledge they have contributed.

[0123] Step S6: Construct and extend the hierarchical quantum federated learning architecture.

[0124] In this embodiment, step S6, constructing and expanding the hierarchical quantum federated learning architecture, may specifically include the following steps: S61 performs hierarchical aggregation and cross-layer communication optimization to construct tree or star hierarchical structures.

[0125] Based on network topology and computing power, an optimal hierarchical structure is dynamically constructed. The minimum communication cost tree is solved using a quantum approximation algorithm. Where T is the spanning tree, c uv For communication costs, d comm This is due to communication delay.

[0126] Cross-layer secure aggregation: Leaf layer: local aggregation from client to edge server; Middle layer: regional aggregation from edge server to regional center; Root layer: global aggregation from regional center to cloud center. Each layer uses independent threshold homomorphic encryption and zero-knowledge proof.

[0127] Different layers are allowed to update at different frequencies: the edge layer updates every round, the region layer every 5 rounds, and the global layer every 20 rounds. A time-warp algorithm is used to synchronize the state.

[0128] Compared to a planar architecture, a layered architecture reduces communication overhead by 62% and training time by 45%.

[0129] The purpose of step S61 is to construct a tree-like or star-like hierarchical structure, reduce communication overhead, and improve the efficiency of large-scale federated learning.

[0130] S62 performs heterogeneous device adaptive and resource-aware scheduling, adapting to heterogeneous computing environments ranging from IoT devices to quantum computers, and achieving optimized resource allocation.

[0131] Device Capability Vector C i =(CPU, Memory, Bandwidth, Number of Qubits, Fidelity), Task Complexity Vector T j =(Data volume, model size, accuracy requirement). Use the Hungarian algorithm to find the optimal allocation: , where x ij ∈{0,1} is the assignment indicator variable.

[0132] Hybrid quantum-classical computational flow: Lightweight devices: participate only in forward propagation and gradient calculation; Medium-weight devices: participate in classical backpropagation; Quantum devices: execute quantum subroutines (feature encoding, optimization); Cloud servers: coordinate global aggregation and model storage; Dynamic load balancing: adjust task allocation using control theory methods based on real-time monitoring data.

[0133] PID controller adjustment parameters: , where e(t) is the load imbalance degree.

[0134] The utilization rate of heterogeneous resources increased from 45% in the traditional solution to 82%, and the task completion time was reduced by 38%.

[0135] The purpose of step S62 is to adapt to heterogeneous computing environments ranging from IoT devices to quantum computers and achieve optimal resource allocation.

[0136] S63 enables federated learning as a service and API standardization, providing standardized interfaces and service frameworks.

[0137] Layered API Design: Basic Layer: Secure communication, key management, authentication; Core Layer: Local training, secure aggregation, contribution evaluation; Service Layer: Model management, monitoring and alerting, billing and settlement; Application Layer: Vertical industry interfaces (medical, financial, industrial); Quantum-Safe Middleware: Develop open-source middleware packages providing: quantum-resistant cryptography library (based on the NIST post-quantum standard), secure multi-party computation engine, zero-knowledge proof compiler, and quantum machine learning simulator, etc.

[0138] Service Level Agreements (SLAs) are automated and managed using smart contracts: Performance guarantees: accuracy, latency, throughput; Security guarantees: privacy budget, attack protection level; Economic terms: incentive allocation, penalty conditions; Compliance: data sovereignty, audit requirements, etc.

[0139] Deployment Case: Deployed within a medical alliance, connecting 15 hospitals, training a medical imaging diagnostic model with an accuracy rate of 96.5% at the expert level, meeting HIPAA compliance requirements.

[0140] The purpose of step S63 is to provide standardized interfaces and service frameworks, reduce the deployment threshold for enterprises, and promote ecosystem development.

[0141] The beneficial effects of implementing this embodiment are: (1) By integrating quantum key distribution (QKD) and post-quantum cryptography (PQC) technologies, a shared key with information theory security is established among the communication participants. By utilizing the quantum no-cloning theorem and the quantum-resistant cryptographic properties of lattice cryptography, it can effectively resist quantum computing threats such as Shor's algorithm and support the real-time encryption needs of highly sensitive scenarios such as finance and healthcare. (2) For non-independent and identically distributed data, the quantum annealing algorithm is introduced to optimize the secure aggregation process. By minimizing the energy function, abnormal model updates can be dynamically identified, which can prevent the inference of the original data through reverse engineering and ensure the security of federated learning in cross-institutional collaboration. (3) By combining Secure Multi-Party Computation (SMPC) and Quantum Federated Learning (QFL), the parallel transmission of parameter gradients is achieved by utilizing the superposition property of quantum states, reducing the communication complexity from O(n) of the classical method to O(log n). The quantum gradient descent (QSGD) algorithm improves the model convergence speed by 3-5 times. At the same time, homomorphic encryption ensures that the data is aggregated in ciphertext, avoiding the risk of leakage in the intermediate process. (4) By constructing a three-layer quantum federated learning architecture of edge-region-global, it supports the dynamic joining and leaving of participants, records the contribution of model updates through blockchain technology, and combines the performance-based salary incentive mechanism to enable participants to obtain encrypted token rewards that match the data quality. It can be scaled up to tens of thousands of nodes, providing a feasible solution for large-scale distributed scenarios such as smart cities and industrial internet.

[0142] This invention can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0143] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0144] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0145] Example 2 Further reference Figure 2 As a response to the above Figure 1 The present invention provides an embodiment of a federated learning device based on quantum-safe multi-party computation, which implements the method shown. Figure 1 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0146] like Figure 2 As shown, the federated learning device 70 based on quantum-safe multi-party computation described in this embodiment includes: a distribution module 71, a construction module 72, a training module 73, a protection module 74, a management module 75, and an expansion module 76. Wherein: The distribution module 71 is used to initialize quantum-safe trust, distribute quantum-safe keys, and establish a shared key among the communication participants. Module 72 is used to build a secure aggregation mechanism against quantum poisoning attacks; Training module 73 is used to train on non-independent and identically distributed data based on quantum heuristics. Protection module 74 is used to protect the privacy of communication participants based on enhanced secure multi-party computation; Management module 75 is used for dynamic participant management of communication participants and setting incentive mechanisms; Extension module 76 is used to build and extend hierarchical quantum federated learning architectures.

[0147] The beneficial effects of implementing this embodiment are: (1) By integrating quantum key distribution (QKD) and post-quantum cryptography (PQC) technologies, a shared key with information theory security is established among the communication participants. By utilizing the quantum no-cloning theorem and the quantum-resistant cryptographic properties of lattice cryptography, it can effectively resist quantum computing threats such as Shor's algorithm and support the real-time encryption needs of highly sensitive scenarios such as finance and healthcare. (2) For non-independent and identically distributed data, the quantum annealing algorithm is introduced to optimize the secure aggregation process. By minimizing the energy function, abnormal model updates can be dynamically identified, which can prevent the inference of the original data through reverse engineering and ensure the security of federated learning in cross-institutional collaboration. (3) By combining Secure Multi-Party Computation (SMPC) and Quantum Federated Learning (QFL), the parallel transmission of parameter gradients is achieved by utilizing the superposition property of quantum states, reducing the communication complexity from O(n) of the classical method to O(log n). The quantum gradient descent (QSGD) algorithm improves the model convergence speed by 3-5 times. At the same time, homomorphic encryption ensures that the data is aggregated in ciphertext, avoiding the risk of leakage in the intermediate process. (4) By constructing a three-layer quantum federated learning architecture of edge-region-global, it supports the dynamic joining and leaving of participants, records the contribution of model updates through blockchain technology, and combines the performance-based salary incentive mechanism to enable participants to obtain encrypted token rewards that match the data quality. It can be scaled up to tens of thousands of nodes, providing a feasible solution for large-scale distributed scenarios such as smart cities and industrial internet.

[0148] Example 3 To address the aforementioned technical problems, embodiments of the present invention also provide an electronic device. Please refer to [link / reference needed]. Figure 3 , Figure 3 This is a basic structural block diagram of the electronic device in this embodiment.

[0149] The aforementioned electronic device 8 includes a memory 81, a processor 82, and a network interface 83 that are interconnected via a system bus. It should be noted that only the electronic device 8 with components 81, 82, and 83 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the electronic device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0150] The aforementioned electronic devices can be computing devices such as desktop computers, laptops, handheld computers, and cloud servers. These electronic devices can interact with users via keyboards, mice, remote controls, touchpads, or voice-activated devices.

[0151] The aforementioned memory 81 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the aforementioned memory 81 may be an internal storage unit of the aforementioned electronic device 8, such as the hard disk or memory of the electronic device 8. In other embodiments, the aforementioned memory 81 may also be an external storage device of the aforementioned electronic device 8, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 8. Of course, the aforementioned memory 81 may also include both internal storage units and external storage devices of the aforementioned electronic device 8. In this embodiment, the aforementioned memory 81 is typically used to store the operating system and various application software installed on the aforementioned electronic device 8, such as computer-readable instructions based on the federated learning method of quantum-safe multi-party computation. In addition, the aforementioned memory 81 can also be used to temporarily store various types of data that have been output or will be output.

[0152] In some embodiments, the processor 82 described above may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 82 is typically used to control the overall operation of the electronic device 8. In this embodiment, the processor 82 is used to execute computer-readable instructions stored in the memory 81 or to process data, for example, to execute computer-readable instructions based on the quantum-safe multi-party computation federated learning method.

[0153] The aforementioned network interface 83 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the aforementioned electronic device 8 and other electronic devices.

[0154] The beneficial effects of implementing this embodiment are: it effectively resists quantum computing threats such as Shor's algorithm, and can support the real-time encryption needs of highly sensitive scenarios such as finance and healthcare; it can prevent the inference of original data through reverse engineering, ensuring the security of federated learning in cross-institutional collaboration; it ensures that data is aggregated in a ciphertext state through homomorphic encryption, avoiding the risk of leakage in the intermediate process; and it can be scaled up to tens of thousands of nodes, providing a feasible solution for large-scale distributed scenarios such as smart cities and industrial internet.

[0155] Example 4 The present invention also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the federated learning method based on quantum-safe multi-party computation as described above.

[0156] The beneficial effects of implementing this embodiment are: it effectively resists quantum computing threats such as Shor's algorithm, and can support the real-time encryption needs of highly sensitive scenarios such as finance and healthcare; it can prevent the inference of original data through reverse engineering, ensuring the security of federated learning in cross-institutional collaboration; it ensures that data is aggregated in a ciphertext state through homomorphic encryption, avoiding the risk of leakage in the intermediate process; and it can be scaled up to tens of thousands of nodes, providing a feasible solution for large-scale distributed scenarios such as smart cities and industrial internet.

[0157] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0158] Obviously, the embodiments described above are merely some embodiments of the present invention, not all embodiments. The accompanying drawings show preferred embodiments of the present invention, but do not limit the patent scope of the present invention. The present invention can be implemented in many different forms; rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the patent protection scope of this invention.

Claims

1. A federated learning method based on quantum secure multi-party computation, characterized in that, Includes the following steps: Initialize quantum-safe trust, distribute quantum-safe keys, and establish a shared key among the communication participants; Construct a secure aggregation mechanism to resist quantum poisoning attacks; Based on quantum inspiration, training is performed on non-independent and identically distributed data; Enhanced secure multi-party computation protects the privacy of all parties involved in the communication. Dynamic participant management and incentive mechanisms should be implemented for communication participants. Constructing and expanding a hierarchical quantum federated learning architecture.

2. The federated learning method based on quantum-safe multi-party computation according to claim 1, characterized in that, The steps of initializing quantum-secure trust, distributing quantum-secure keys, and establishing a shared key among the communication participants specifically include: Quantum-secure key distribution based on QKD; Based on blockchain anchoring, identity authentication and trust anchoring are performed on communication participants; The security parameters are initialized, and a homomorphic key is generated.

3. The federated learning method based on quantum-safe multi-party computation according to claim 1, characterized in that, The steps for constructing a secure aggregation mechanism against quantum poisoning attacks specifically include: Perform a quantum security contribution assessment and calculate the Shapley value; Perform lattice-based anomaly detection and poisoning identification, and detect and filter poisoning model updates submitted by malicious participants; Quantum robust weighted aggregation and gradient pruning are performed, and secure weighted aggregation is implemented after identifying malicious updates.

4. The federated learning method based on quantum-safe multi-party computation according to claim 1, characterized in that, The steps for training on non-independent and identically distributed data based on quantum heuristics specifically include: Map the feature distribution of local data to the quantum state space to achieve feature alignment across clients; Design a quantum optimization algorithm suitable for federated scenarios to accelerate the convergence of non-convex loss functions; Automatically search for the quantum-classical hybrid model architecture that best suits the client's data distribution.

5. The federated learning method based on quantum-safe multi-party computation according to claim 1, characterized in that, The steps for protecting the privacy of communication participants based on enhanced secure multi-party computation specifically include: Threshold homomorphic encryption and multi-party decryption are performed to achieve end-to-end encryption of model updates, ensuring that only clients that have reached the threshold can decrypt the aggregated results. Zero-knowledge proofs are used to verify the correctness of computations, allowing clients to verify that the coordinator and other clients have correctly executed the protocol. Verifiable secret sharing and robust recovery ensure that aggregation results can be correctly recovered even if the client goes offline or maliciously submits incorrect shares.

6. The federated learning method based on quantum-safe multi-party computation according to claim 1, characterized in that, The steps for dynamic participant management and setting up incentive mechanisms for communication participants specifically include: The creditworthiness of participants is assessed based on their historical behavior, and their weight in the aggregation is dynamically adjusted. Incentives are automatically distributed to participants in the communication process based on reputation and contribution. Establish a secure exit and knowledge retention mechanism for communication participants.

7. The federated learning method based on quantum-safe multi-party computation according to any one of claims 1 to 6, characterized in that, The specific steps for constructing and expanding the hierarchical quantum federated learning architecture include: Perform hierarchical aggregation and cross-layer communication optimization to construct tree or star-shaped hierarchical structures; Perform heterogeneous device adaptive and resource-aware scheduling to adapt to heterogeneous computing environments ranging from IoT devices to quantum computers and achieve optimized resource allocation; We will implement federated learning as a service and API standardization, providing standardized interfaces and service frameworks.

8. A federated learning device based on quantum-safe multi-party computation, characterized in that, include: The distribution module is used to initialize quantum-safe trust, distribute quantum-safe keys, and establish a shared key among the communication participants. Build modules are used to construct secure aggregation mechanisms against quantum poisoning attacks; The training module is used to train on non-independent and identically distributed data based on quantum heuristics. The protection module is used to protect the privacy of communication participants based on enhanced secure multi-party computation. The management module is used for dynamic participant management of communication participants and setting incentive mechanisms; Extension modules are used to build and extend hierarchical quantum federated learning architectures.

9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the federated learning method based on quantum-safe multi-party computation as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions that, when executed by a processor, implement the steps of the federated learning method based on quantum-safe multi-party computation as described in any one of claims 1 to 7.

Images