Federal deep learning method for interpretability and sensitivity perception of 6G slices

Abstract

The invention discloses a federated deep learning method for interpretability and sensitivity perception of 6G slices in the technical field of network communication, and the method comprises the steps: predicting the call drop rate of each radio access network (RAN) slice through an interpretable federated learning local model; meanwhile, sensitivity perception and interpretability indexes are used as constraints in non-independent identically-distributed local data set setting, the credibility of an interpreter is quantitatively verified through a logarithmic probability score, and the score can be used as a constraint condition in a federated learning optimization task during operation; the performance of the method provided by the invention is obviously superior to that of traditional self-organizing federal deep learning after unconstrained integral gradient, and the complexity of network resource management can be effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of network communications, and in particular to a federated deep learning method for interpretability and sensitivity perception of 6G slices. Background Art

[0002] Currently, the more popular 6G network slicing technology is to adopt end-to-end (E2E) autonomous management and orchestration of network resources in the network domain, because the isolation of slices may lead to high costs in terms of efficiency. Therefore, the ETSI organization has begun to consider a standardized zero-touch network and service management framework, which refers to the automation and management of resources without human intervention. In addition, the development of cognitive slice management solutions in 6G networks is crucial for automatically orchestrating and managing network slices, especially network resources across different technical domains (TDs), and ensuring QoE and QoS for end users.

[0003] In addition, 6G network technology has given birth to AI-native network slice management solutions to support emerging AI services. Artificial intelligence algorithms are driven by the distributed nature of data sets to obtain the full potential of network slice automation to solve the problematic behaviors of traditional cloud-centric machine learning solutions. Therefore, a decentralized learning method is needed to effectively handle distributed network slices.

[0004] 6G network will be a "machine-centric" technology, which means that all corresponding "smart things" in 6G network will operate intelligently, just like a smart black box, which is opaque in its actions or decision-making process and may have an adverse impact on the network operation of 6G technology. In this regard, XAI provides human-interpretable methods to fully explain AI systems and their decisions to gain human trust in the loop. In view of this fact, zero-touch XAI-driven federated learning will be particularly valued for its automation and unique advantages, which are crucial for end-user trust and security procedures.

[0005] The present invention proposes a federated deep learning method for joint explainability and sensitivity perception of 6G slices, which is significantly better than the performance of self-organizing federated deep learning after unconstrained integral gradient, and can effectively reduce the complexity of network resource management. Summary of the invention

[0006] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of the present invention to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0007] Therefore, the purpose of the present invention is to provide a federated deep learning method for interpretability and sensitivity perception of 6G slices, predict the call drop rate of each radio access network (RAN) slice through an interpretable federated learning local model, and use sensitivity perception and interpretability indicators as constraints in such non-independent and identically distributed local data set settings. The credibility of the interpreter is quantitatively verified by the logarithmic probability score, which can be used as a constraint condition in the runtime federated learning optimization task. The method provided by the present invention is significantly better than the performance of traditional unconstrained integrated gradient post-self-organizing federated deep learning, and can effectively reduce the complexity of network resource management.

[0008] In order to solve the above technical problems, the present invention provides a federated deep learning method for interpretability and sensitivity perception of 6G slices, which adopts the following technical solution: the method described is mainly implemented through an application framework, and the application framework includes

[0009] A wireless access network, wherein the wireless access network is composed of K base stations, wherein each base station deploys N parallel slices, and each base station runs a local control closed loop, wherein the local control closed loop collects monitoring data and performs call drop rate prediction, wherein the local control closed loop is independently trained to obtain a local federated learning local model;

[0010] The method described runs iteratively in a closed-loop manner via a local control loop, with a runtime interpreter, model tester, and log-odds mapper,

[0011] For each local control loop, the local federated learning local model feeds the model graph to the model tester, which then tests the features. and the corresponding predictive features is fed to the explainer, which first generates feature attributes using the integrated gradient XAI method; the log-odds mapper then uses these attributes to select the top p% of features and then calculates the log-odds score θ k,n , the score is fed back to the local federated learning local model to be included in the local constraint optimization of the local federated learning local model.

[0012] It also includes a recall mapper that calculates the recall score ρ based on the predicted values ​​and true values ​​output by the model tester and explainer. k,n , incorporating it into the optimization of local constraints of the local model of local federated learning;

[0013] For each local control loop, the predicted call drop rate is classified into m=(1,…,D k,n ), combined with the local dataset D k,n Long-term statistical constraints defined on samples and corresponding recall scores ρ k,n and the log-odds score θk,n Under the constraints of The main loss function is minimized.

[0014] Optionally, the monitoring data collected by the local control closed loop is used to construct a local data set for slice n (n=1,...N), and the local data set includes: an average physical resource block, an average data transmission delay, and an SNR value characterizing the channel quality, which is expressed by the formula:

[0015] in, represents the i-th input feature in the local dataset, and represents the corresponding i-th output feature, k (k=1, .. . K), and the output features include: call drop rate.

[0016] Optionally, it also includes a federated learning layer, which is used to collect local federated learning local model information obtained by each local control closed loop being trained independently, aggregate the local federated learning local model information obtained by each local control closed loop being trained independently through an algorithm, and then feed back the obtained global model information to the aggregated cloud server to optimize the resource management of the wireless access network.

[0017] Optionally, in the tth round of closed-loop iteration, the weights of the local closed-loop optimization in the federated learning framework are The aggregate cloud server generates a global federated learning model for slice n, as shown in the following formula:

[0018]

[0019] in It represents the weight factor of the nth slice in the kth base station in the tth round, which is used in the subsequent constraint optimization to minimize the main loss function, thereby building the implementation framework of federated learning. is the total data samples of all data sets related to slice n. Then, the aggregation cloud server broadcasts the global federated learning model to all local K local control loops that use it to start the next round of local iterative optimization.

[0020] Optionally, the local federated learning local model information includes local federated learning local model parameters or local federated learning local model update parameters. After receiving the updated model graph, the model tester reconstructs the test prediction from the data subset extracted from the local data set. All model diagrams, test features and the corresponding predictive features will be provided to the interpreter.

[0021] Optionally, the calculation formula of the logarithmic probability score is:

[0022]

[0023] in, is the predictive feature, is the i-th input feature in the local dataset, Indicates the features in which the first p% features in the modified local dataset are zero-filled.

[0024] Optionally, the recall score ρ k,n The calculation formula is:

[0025]

[0026] Among them, π + (D k,n )Define D k,n The proportion of positive values, D k,n [*] is D that satisfies the expression * k,n A subset of .

[0027] Optionally, a lower bound α is imposed on the recall score n , setting an upper limit β on the log-odds score n , the constraint optimization is transformed into solving the constrained local classification problem in the round t,(t=0,…,t-1) of federated learning specified by the iteration cycle, that is,

[0028]

[0029] ρ k,n ≥α n ,

[0030] θ k,n ≤β n .

[0031] Optionally, two Lagrangian equation frameworks are constructed to solve the above constrained local classification problem, as follows:

[0032]

[0033] Among them, Φ1 and Φ2 represent the original constraints, ψ1 and ψ2 represent the smoothing substitutes, λ1 and λ2 represent the Lagrange multipliers,

[0034] The smoothing substitution formula is as follows:

[0035]

[0036] ψ2=Φ2=β n -θ k,n .

[0037] Optionally, the constraint optimization is a non-zero-sum two-player game strategy, and the local constraint optimization is calculated by the following formula:

[0038]

[0039] In summary, the present invention includes at least one of the following beneficial effects: predicting the call drop rate of each radio access network (RAN) slice through an interpretable federated learning local model, and taking sensitivity perception and interpretability indicators as constraints in such non-independent and identically distributed local data set settings, and quantitatively verifying the credibility of the interpreter through the logarithmic probability score, which can be used as a constraint condition in the runtime federated learning optimization task. The method provided by the present invention is significantly better than the performance of traditional unconstrained integrated gradient post-self-organizing federated deep learning, and can effectively reduce the complexity of network resource management. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0041] Figure 1 A flow chart of a federated deep learning method for interpretability and sensitivity perception of 6G slices according to the present invention;

[0042] Figure 2 The flowchart of the federated deep learning method for interpretability and sensitivity perception of 6G slices of the present invention is iteratively run in a closed-loop manner through a local control loop. DETAILED DESCRIPTION

[0043] The following will combine the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all the embodiments. All other embodiments obtained by ordinary technicians in this field without creative work based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0044] In the description of the present invention, it should be noted that the terms "upper", "lower", "inner", "outer", "top / bottom" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific position, be constructed and operated in a specific position, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0045] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "provided with", "mounted / connected", "connected", etc. should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0046] Embodiment 1

[0047] The present invention discloses a federated deep learning method for interpretability and sensitivity perception of 6G slices, wherein the method is mainly implemented through an application framework, such as Figure 1 As shown, the application framework includes

[0048] Radio Access Network,The Radio Access Network (RAN) consists of K base stations (BSs), each of which deploys a set of N parallel slices. Each BS runs a local control loop, which collects monitoring data and performs call drop rate prediction. The monitoring data collected by the local control loop is used to construct a local dataset for slice n (n=1,...N), that is,

[0049]

[0050] in, represents the i-th input feature in the local dataset, and represents the corresponding i-th output feature, base station k (k=1,...K), the local data set includes: average physical resource block, average data transmission delay, SNR value representing channel quality, etc., and the output features include: call drop rate, etc.

[0051] In detail, in this embodiment, a federated learning layer is also included, which is used to collect local federated learning local model information obtained by independently training each local control closed loop, aggregate the local federated learning local model information obtained by independently training each local control closed loop through an algorithm, and then feed back the obtained global model information to the aggregated cloud server to optimize the resource management of the wireless access network.

[0052] In the field of communication anomaly detection, to train an accurate classifier, ideally a very comprehensive data set that covers all possible situations is required. However, due to the different traffic distributions caused by heterogeneous user distribution and channel conditions, these accumulated data sets are actually non-independent and identically distributed data sets, and it is difficult to include all possible abnormal situations in one data set. In the method provided by the present invention, without directly sharing data, the data source (local control closed loop) from multiple slices is used to train the model. Since each local control closed loop may have some unique data, these data contain abnormal situations or features not covered by other data sets. The end-to-end slice-level federated learning layer is a key part in the federated learning application framework. It manages and coordinates the federated learning process in a slice-like manner from an end-to-end (from the source of data generation to the final model aggregation application) perspective, and collects the model parameters or model updates obtained by each local closed loop trained on its own data. Then, the model information from different local closed loops is aggregated through a specific algorithm (such as weighted average, etc.), and then the obtained global model information is fed back to the aggregation cloud server for resource management optimization of the radio access network (RAN).

[0053] The present invention provides a federated deep learning method for interpretability and sensitivity perception of 6G slices, wherein a constrained call drop detection classifier and an interpreter exchange attributes and predicted values ​​of features in a closed-loop manner to achieve transparent zero-touch service management of 6G network slices of the radio access network (RAN) under the condition of non-independent and identically distributed data sets. An integrated gradient explainable artificial intelligence (XAI) method is used to display the characteristic attributes of the radio access network (RAN) slice, and the generated attributes are used to quantitatively verify the credibility of the explanation through the logarithmic probability indicator, and the indicator is used as a constraint in the optimization task of federated learning. Furthermore, the present invention formulates the corresponding optimization problem of federated learning with joint recall and logarithmic probability constraints under the proxy Lagrangian framework, and optimizes and solves it through a non-zero-sum two-player game strategy. The specific processing flow is as follows Figure 2 shown.

[0054] The local control loop is trained independently to obtain a local federated learning local model; the method described is iteratively run in a closed-loop manner through the local control loop, and has a runtime interpreter, model tester and log-probability mapper. For each local control loop, the local federated learning local model feeds the model graph to the model tester, and the model tester tests the feature and the corresponding predictive features Provided to the explainer, which first generates feature attributes using the integrated gradient XAI method; then the log-odds mapper uses these attributes to select the top p% of features, and then calculates the log-odds score, which is fed back to the local federated learning local model to be included in the local constraint optimization of the local federated learning local model.

[0055] It also includes a recall mapper that calculates the recall score ρ based on the predicted values ​​and true values ​​output by the model tester and explainer. k,n , and incorporate it into the local constraint optimization of the local federated learning local model; for each local control closed loop, the predicted call drop rate is classified into m=(1,…,D k,n ), combined with the local dataset D k,n Long-term statistical constraints defined on samples and corresponding recall scores ρ k,n and the log-odds score θ k,n Under the constraints of The main loss function is minimized.

[0056] In detail, in this embodiment, in the tth round of iterative operation in the closed-loop mode, the weights after local closed-loop optimization in the federated learning framework are The aggregate cloud server generates a global federated learning model for slice n, as shown in the following formula:

[0057]

[0058] in It represents the weight factor of the nth slice in the kth base station in the tth round, which is used in the subsequent constraint optimization to minimize the main loss function, thereby building the implementation framework of federated learning. is the total data samples of all data sets related to slice n. Then, the aggregation cloud server broadcasts the global federated learning model to all local K local control loops that use it to start the next round of local iterative optimization.

[0059] In detail, in this embodiment, the local federated learning local model information includes local federated learning local model parameters or local federated learning local model update parameters. After receiving the updated model graph, the model tester reconstructs the test prediction from the data subset extracted from the local data set. All model diagrams, test features and the corresponding predictive features will be provided to the interpreter.

[0060] In detail, in this embodiment, the interpreter displays the characteristic attributes of the radio access network (RAN) slice by using a low-complexity integrated gradient explainable artificial intelligence (XAI) method, and the interpreter uses the attributes generated by the integrated gradient to explain the prediction results of the model. If the model predicts that the sample belongs to a certain category, the interpreter can find out the features that contribute most to the prediction result by analyzing the value, thereby providing an intuitive explanation to the user, explaining which features the model is based on to make decisions. The interpreter can explain the performance differences of the model on different slices by analyzing the integrated gradient distribution on different slice data. If the model performance is poor on a certain slice, the interpreter can check the feature importance distribution of the slice data to see if there are abnormal integrated gradients of some key features, or if the importance of some features on different slices varies greatly, so as to find out the reasons that may cause performance differences and take corresponding measures to improve them.

[0061] In detail, in this embodiment, in order to characterize the credibility of the local model, it is preferred to calculate the logarithmic probability score θ k,n , which measures the impact of the top attribute features on the model predictions. Specifically, the log-odds score is defined as the average difference in the negative log probability of the predicted class before and after masking the top p% of features with zero padding. The log-odds mapper first selects the top p% of features based on the attributes collected from the explainer and replaces them with zero padding. That is,

[0062]

[0063] in, is the predictive feature, is the i-th input feature in the local dataset, denotes the features where the first p% features in the modified local dataset are zero-padded. Finally, the log-odds mapper computes the log-odds score, which is used as one of the constraints to constrain the FL optimization task.

[0064] In detail, in this embodiment, the sensitivity perception score (recall mapping) preferably uses the call recall rate as the score of the sensitivity of the federated learning local classifier, expressed as ρ k,n ,Right now,

[0065]

[0066] Among them, π + (D k,n )Define D k,n The proportion of positive values, D k,n [*] is D that satisfies the expression * k,n A subset of .

[0067] In detail, in this embodiment, in order to make the abnormal detection / classification of dropped calls more reliable, an AI service quality protocol integrating federated learning is established between the slice tenant and the infrastructure provider, in which a lower limit α is imposed on the recall score. n , setting an upper limit β on the log-odds score n , the constraint optimization is transformed into solving the constrained local classification problem in the round t,(t=0,…,t-1) of federated learning specified by the iteration cycle, that is,

[0068]

[0069] ρ k,n ≥α n ,

[0070] θ k,n ≤β n .

[0071] In detail, in this embodiment, the constrained local classification problem can be solved by a Lagrangian framework, and two Lagrangian equations need to be further constructed as shown below:

[0072]

[0073]

[0074] Among them, Φ1 and Φ2 represent the original constraints, ψ1 and ψ2 represent the smoothing substitutes, λ1 and λ2 represent the Lagrange multipliers,

[0075] The smoothing substitution formula is as follows:

[0076]

[0077] ψ2=Φ2=β n -θ k,n , since the negative logarithm is already a convex function, this also confirms that the solution to the optimization problem is equivalent to the solution obtained when using only the original constraints.

[0078] In detail, in this embodiment, the constraint optimization is a non-zero-sum two-player game strategy, and the local constraint optimization is calculated by the following formula:

[0079]

[0080] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A federated deep learning approach for interpretability and sensitivity awareness of 6G slices, characterized by: The method is mainly implemented through an application framework, and the application framework includes A wireless access network, wherein the wireless access network is composed of K base stations, wherein each base station deploys N parallel slices, and each base station runs a local control closed loop, wherein the local control closed loop collects monitoring data and performs call drop rate prediction, wherein the local control closed loop is independently trained to obtain a local federated learning local model; The method described runs iteratively in a closed-loop manner via a local control loop, with a runtime interpreter, model tester, and log-odds mapper, For each local control loop, the local federated learning local model feeds the model graph to the model tester, which then tests the features. and the corresponding predictive features is fed to the explainer, which first generates feature attributes using the integrated gradient XAI method; the log-odds mapper then uses these attributes to select the top p% of features and then calculates the log-odds score θ k,n , the score is fed back to the local federated learning local model to be included in the local constraint optimization of the local federated learning local model. It also includes a recall mapper that calculates the recall score ρ based on the predicted values ​​and true values ​​output by the model tester and explainer. k,n , incorporating it into the optimization of local constraints of the local model of local federated learning; For each local control loop, the predicted call drop rate is classified into Combined local dataset D k,n Long-term statistical constraints defined on samples and corresponding recall scores ρ k,n and the log-odds score θ k,n Under the constraints of The main loss function is minimized.

2. The method for interpretability and sensitivity perception of 6G slices according to claim 1, characterized in that: The monitoring data collected by the local control closed loop is used to construct a local data set for slice n (n=1,...N), and the local data set includes: an average physical resource block, an average data transmission delay, and an SNR value representing the channel quality, which is expressed by the formula: in, represents the i-th input feature in the local dataset, and represents the corresponding i-th output feature, k (k=1, .. . K), and the output features include: call drop rate.

3. The federated deep learning method for interpretability and sensitivity perception of 6G slices according to claim 1, characterized in that: Also includes The federated learning layer is used to collect the local federated learning local model information obtained by each local control closed loop training independently, aggregate the local federated learning local model information obtained by each local control closed loop training independently through an algorithm, and then feed back the obtained global model information to the aggregated cloud server to optimize the resource management of the wireless access network.

4. The method for interpretability and sensitivity perception of 6G slices according to claim 3, characterized in that: In the tth round of iterative operation in the closed-loop mode, the weights of the local closed-loop optimization in the federated learning framework are The aggregate cloud server generates a global federated learning model for slice n, as shown in the following formula: in It represents the weight factor of the nth slice in the kth base station in the tth round, which is used in the subsequent constraint optimization to minimize the main loss function, thereby building the implementation framework of federated learning. is the total data samples of all data sets related to slice n. Then, the aggregation cloud server broadcasts the global federated learning model to all local K local control loops that use it to start the next round of local iterative optimization.

5. The method for interpretability and sensitivity perception of federated deep learning for 6G slices according to claim 3, characterized in that: The local federated learning local model information includes local federated learning local model parameters or local federated learning local model update parameters. After receiving the updated model graph, the model tester reconstructs the test prediction from the data subset extracted from the local data set. All model diagrams, test features and the corresponding predictive features will be provided to the interpreter.

6. The method for interpretability and sensitivity perception of 6G slices according to claim 1, characterized in that: The calculation formula for the log-odds score is: in, is the predictive feature, is the i-th input feature in the local dataset, Indicates the features in which the first p% features in the modified local dataset are zero-filled.

7. The method for interpretability and sensitivity perception of federated deep learning for 6G slices according to claim 1, characterized in that: The recall score ρ k,n The calculation formula is: Among them, π + (D k,n )Define D k,n The proportion of positive values, D k,n [*] is D that satisfies the expression * k,n A subset of .

8. The method for interpretability and sensitivity perception of federated deep learning for 6G slices according to claim 4, characterized in that: A lower limit α is imposed on the recall score n , setting an upper limit β on the log-odds score n , the constraint optimization is transformed into solving the constrained local classification problem in the round t,(t=0,…,t-1) of federated learning specified by the iteration period, that is, r k,n ≥a n , i k,n ≤β n 。 9. The method for interpretability and sensitivity perception of federated deep learning for 6G slices according to claim 8, characterized in that: Two Lagrangian equation frameworks are constructed to solve the above constrained local classification problem, as follows: Among them, Φ1 and Φ2 represent the original constraints, ψ1 and ψ2 represent the smoothing substitutes, λ1 and λ2 represent the Lagrange multipliers, The smoothing substitution formula is as follows: ψ2=Φ2=β n -θ k,n 。 10. The method for interpretability and sensitivity perception of federated deep learning for 6G slices according to claim 9, characterized in that: The constrained optimization is a non-zero-sum two-player game strategy, and the local constrained optimization is calculated by the following formula:

Images