Joint method for D2D pairing and device scheduling facing federated edge learning

Abstract

The invention discloses a federated edge learning-oriented D2D pairing and device scheduling joint method, and belongs to the technical field of device participation and data sharing in a federated edge learning environment. The invention provides a federated edge learning architecture supporting D2D communication, mainly focuses on an association mode and an entity scheduling strategy of a D2D device pair, focuses on the influence of a communication packet loss error rate and a data heterogeneous characteristic of a mobile device on the convergence performance of a model, and aims to improve the problem of device access quantity limitation caused by limited base station spectrum resources. According to the method, an equipment association scheduling strategy is introduced, so that the elasticity of a federated edge learning system to fluctuation communication conditions is enhanced; by introducing a fairness constraint, it is ensured that each device has an opportunity to participate in training, so that global model learning data is more comprehensive; the base station access amount is increased by introducing D2D communication, and the performance of edge federated learning is improved; the global loss is minimized under the condition that the receiving capacity limitation of the edge base station is met.

Description

Technical Field

[0001] The present invention relates to the technical field of device participation and data sharing in a federated edge learning environment, and specifically to a joint method for D2D pairing and device scheduling for federated edge learning. Background Art

[0002] Thanks to the rapid development of 5G / IoT technology, the number of IoT devices has grown exponentially. As more and more IoT devices are connected to the Internet of Things, a large amount of data is generated at the edge of the wireless network. In order to utilize these valuable data resources, traditional machine learning uploads them to a central server for training. However, this process may lead to the leakage of device privacy data and cause the adverse effect of data abuse. Therefore, a federated edge learning system deployed at the edge of the network is proposed to overcome the above problems, which allows edge devices participating in model training to collaboratively train global models without exposing local data. Unfortunately, the deployment of federated edge learning currently faces two practical challenges: limited spectrum resources and data heterogeneity. Specifically, due to the constraints of concurrent processing capabilities and spectrum resources, edge base stations cannot simultaneously support the transmission services of massive IoT devices, and can only allow a small number of devices to perform a single round of centralized access, which greatly limits the degree of device participation in global model training and has a negative impact on model convergence performance. In addition, the local data sets of large-scale distributed devices are usually statistically heterogeneous and highly redundant, which can easily lead to slow model convergence or even model drift, especially when only a limited number of devices participate in each round of training. Existing research considers increasing the number of devices participating in each round of training through device-to-device (D2D) communication to improve model convergence performance while alleviating base station access constraints. However, due to the influence of unreliable communication environment, there are potential packet loss and device availability constraints during model transmission. The imbalance of local data sets between devices will also affect the device D2D association and scheduling strategy in each training round. Therefore, it is necessary to design a secure and reliable D2D association scheduling scheme that guarantees learning performance in a federated edge learning system under complex communication environment.

[0003] Therefore, how to solve the above-mentioned problems has become a difficult problem that needs to be solved urgently. Summary of the invention

[0004] The purpose of the present invention is to provide a joint method for D2D pairing and device scheduling for federated edge learning, aiming to improve the problem of device access quantity limitation caused by limited base station spectrum resources, increase the number of devices participating in each round of training by establishing and accessing D2D device pairs, thereby reducing the impact of model deviation caused by data heterogeneity and improving model robustness and convergence speed.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A joint approach to D2D pairing and device scheduling for federated edge learning,

[0007] The steps of the joint method for D2D pairing and device scheduling for federated edge learning are as follows:

[0008] S1: Federated learning is performed using a federated edge learning system. The federated edge learning system is a global model formed by collaborative training of an edge base station with limited wireless resources and several devices. The device set is C = {1, 2, ..., C}, C = |C|, and the base station can establish connections with N devices at the same time; the device set that directly communicates with the base station in the tth round is S t , the two devices constitute a D2D device pair set P t , wherein each of the D2D device pairs includes a device B communicating with the base station i,t and said B i,t Communication equipment D i,t , for i∈S t ∪P t , scheduling strategy a i,t ∈[0,1],a i,t =1 means entity i participated in training at training round t, otherwise it did not participate;

[0009] The D2D device communication and the direct communication from the device to the base station are both transmitted in quasi-static Rayleigh fading using orthogonal frequencies; each device encapsulates the model parameters into a single data packet for transmission; the base station verifies the integrity of the data packet through a cyclic redundancy check, and discards the message if it is damaged or lost;

[0010] Before each round of federated learning, the packet loss rate of direct communication between all devices and the base station and the communication packet loss rate of all feasible D2D associated device pairs are calculated respectively, and the availability constraint index is introduced to indicate whether the local model scheduling strategy of the device can be uploaded successfully;

[0011] S2: Considering the negative impact of device data heterogeneity on model training, a virtual queue is introduced for each device in each round of device selection to implement fairness constraints;

[0012] S3: Data quality, packet loss rate, and virtual queues are comprehensively considered as the measurement criteria and the basis for device pairing and selection, and the optimization problem is modeled as a constrained maximum weighted independent edge set problem;

[0013] S4: D2D devices are paired and selected in each round of training through the bipartite graph maximum weight matching algorithm. The D2D device pairs and direct communication devices selected to participate in the training upload the local updated model to the base station for global aggregation, and broadcast the new global model to the selected devices for the next round of training until the model converges.

[0014] Preferably, in S1, the packet loss rate of direct communication between all devices and the base station and the communication packet loss rate of all feasible D2D associated device pairs are as follows:

[0015] The packet loss rate of direct communication between device i and base station is expressed as:

[0016]

[0017] Among them, P i,B 、h i,B is the transmission power and channel gain of device i in direct communication with the base station, W B is the allocated communication bandwidth, N 0 is the noise power spectral density, τ is the system-specific threshold parameter;

[0018] The packet loss rate of D2D device communication is:

[0019]

[0020] Among them, P i,D 、h i,D is the transmission power and channel gain of device i for D2D communication, W D is the dedicated bandwidth for D2D communication, τ′ is the threshold parameter related to the reverse coding gain of D2D communication;

[0021] Define the model transmission index C from device to base station i,B ∈{0,1}, when C i,B =1, it means the local model of device i is uploaded successfully, otherwise C i,B =0, the indicator can be expressed as:

[0022]

[0023] For a D2D device pair i∈P t , introduce the index C i,D ∈{0,1}, when C i,D =1 indicates that the D2D model transmission is successful, otherwise C i,D =0, the indicator can be expressed as:

[0024]

[0025] Preferably, in S2, considering that the data heterogeneity of the device will have a negative impact on the efficiency of model training, the fairness constraint is introduced:

[0026]

[0027] Where T represents the total number of training rounds, c i ∈[0,1) represents the ratio of the minimum number of communication rounds required to select device i to the total number of training rounds; in order to solve the above fairness constraints, a virtual queue D is introduced for device i∈C. i ,Right now:

[0028] D i (t)=[D i (t-1)+c i -a i,t-1 ] + .

[0029] Among them, D i (t) represents the queue length at the beginning of round t, [x] + = max{x,0}, the global device is represented by D(t) = [D 1 (t),...,D C (t)], D(1)=0.

[0030] Preferably, in S3, by optimizing the D2D association and scheduling strategy, an optimization problem with the goal of minimizing the global loss is formulated, including the following steps:

[0031] F1: Federated learning rate satisfies Given a device scheduling matrix a t and the optimal global federated learning model w * , the convergence upper bound is described as:

[0032]

[0033] Among them, in order to meet the above global training requirements, the local loss function of device training satisfies ζ 1 and 2 is a non-negative constant, the global loss function L(w) is quadratically continuously differentiable and satisfies L and μ are non-negative constants for Lipschitz continuity and strong convexity, respectively, and

[0034] F2: To minimize the convergence upper bound, we only need to minimize the interval Rewrite A as Rewrite B as Furthermore, the optimization problem is expressed as:

[0035]

[0036] Where T is the total number of training rounds, N is the maximum number of devices that the base station can access at any given time;

[0037] F3: The optimization problem is modeled as a graph theory problem, where devices are regarded as vertices and potential matches between devices are regarded as edges. Therefore, the optimization problem can be simplified to a constrained maximum weighted independent edge set problem, by finding a set of edges that do not share vertices to maximize the total weight.

[0038] Preferably, in S3, all devices participating in the training use standard gradient descent for local training; by introducing the scheduling strategy and considering the packet loss rate, the update rule of the global model w can be described as:

[0039]

[0040] Among them, K i,B , K i,D and K j Respectively represent device B i,t , D i,t and the size of j’s local dataset, w i,t , w j,t denote the local model parameters of D2D device pair i and direct communication device j from round t, respectively.

[0041] Preferably, in S4, the steps of solving the above maximum weighted independent edge set problem are as follows:

[0042] Step 1): Convert device i∈C to node u i ∈U to construct the general graph G(U) and define the vertex weight of device i:

[0043] V i∈St =βK i (1-q i,B )+(1-β)D i (t)

[0044] Where β∈[0,1]; without loss of generality, the edge weight of D2D device pair i is defined as:

[0045] V i∈Pt =β(1-q i,B )(K i,B +K i,D (1-q i,D ) 2 )+(1-β)(D i,B (t)+D i,D (t))

[0046] Among them, D i,B (t) and D i,D (t) respectively represent equipment B i,t and D i,t Virtual queues;

[0047] Step 2): Perform symmetric operations on the nodes contained in U to obtain a new node set, denoted as V, and construct a bipartite graph G(U,V). i ∈U,v j ∈V, establish edge e(i,j), use the bipartite graph maximum weight matching algorithm to achieve the maximum matching M, and sort M in descending order of edge weight, and take the Top-N elements to meet the constraint on the number of base station access devices.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] (1) This method takes into account the problem of packet loss during model transmission due to factors such as poor channel conditions and exhausted computing power, and introduces a device-associative scheduling strategy to enhance the resilience of the federated edge learning system to fluctuating communication conditions.

[0050] (2) This method takes into account the negative impact of the heterogeneous characteristics of local data sets on mobile devices on the efficiency and performance of model training. By introducing fairness constraints, it ensures that each device has the opportunity to participate in training, making the data for global model learning more comprehensive.

[0051] (3) This method takes into account the limited number of base station accesses in the edge environment, and aims to increase the number of base station accesses by introducing D2D communication to improve the performance of edge federated learning. Secondly, the global loss is minimized under the condition of satisfying the receiving capacity limit of the edge base station, that is, D2D pairing and device scheduling are used to achieve superior learning performance in complex communication environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a system model framework diagram of the present invention;

[0053] Figure 2 It is the overall flow chart of the present invention; DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] like Figure 1As shown, the present invention describes a device D2D association scheduling method for federated edge learning. The method first proposes a federated edge learning architecture based on D2D communication to enhance the federated learning mechanism. The association scheduling of D2D device pairs is performed by collaboratively considering factors such as transmission packet loss rate and data heterogeneity before each round of training, so as to maximize the reliability of the overall device participation in training.

[0056] Embodiment 1

[0057] like Figure 2 As shown, the present invention discloses a D2D pairing and device scheduling joint method for federated edge learning. The federated edge learning system includes an edge base station with limited wireless resources and a global model formed by collaborative training of a large number of devices. The device set is C = {1, 2, ..., C}, C = |C|, and the base station can establish connections with up to N devices at the same time. S t and P t They represent the device set that directly communicates with the base station in the tth round and the set of D2D device pairs consisting of two devices, where each D2D device pair contains device B that communicates with the base station. i,t and B i,t Communication equipment D i,t , for i∈S t ∪P t , scheduling strategy a i,t ∈[0,1],a i,t =1 means entity i participates in training at training round t, otherwise it does not participate. The specific steps are as follows:

[0058] Step 1: Before each round of federated learning, the packet loss rate of direct communication between all devices and the base station and the communication packet loss rate of all feasible D2D associated device pairs are calculated respectively, and the availability constraint indicator is introduced to indicate whether the local model scheduling strategy of the device can be uploaded successfully.

[0059] Step 2: Considering the negative impact of device data heterogeneity on model training, a virtual queue is introduced for each device in each round of device selection to implement fairness constraints.

[0060] Step 3: Take data quality, packet loss rate, and virtual queue into comprehensive consideration as the measurement criteria and the basis for device pairing and selection, and model the optimization problem as a constrained maximum weighted independent edge set problem.

[0061] Step 4: D2D devices are paired and selected through the bipartite graph maximum weight matching algorithm. The D2D device pairs and direct communication devices selected for training upload the local updated models to the base station for global aggregation, and broadcast the new global model to the selected devices for the next round of training until the model converges.

[0062] In step 1 of this embodiment, both D2D communication and direct communication from device to base station are transmitted in quasi-static Rayleigh fading using orthogonal frequencies, and there is no interference between devices. Each device encapsulates the model parameters into a single data packet for transmission, and the base station verifies the integrity of the data packet through a cyclic redundancy check. If it is damaged or lost, the message is discarded. The packet loss rate of direct communication between device i and the base station is expressed as:

[0063]

[0064] Among them, P i,B 、h i,B is the transmission power and channel gain of device i in direct communication with the base station, W B is the allocated communication bandwidth, N 0 is the noise power spectral density, and τ is the system-specific threshold parameter.

[0065] Similar to direct communication between devices and base stations, the packet loss rate of D2D device communication is:

[0066]

[0067] Among them, P i,D 、h i,D is the transmission power and channel gain of device i for D2D communication, W D is the dedicated bandwidth for D2D communication, and τ′ is a threshold parameter related to the reverse coding gain of D2D communication.

[0068] Furthermore, the model transmission index C from the device to the base station is defined i,B ∈{0,1}, when C i,B =1, it means the local model of device i is uploaded successfully, otherwise C i,B =0. This indicator can be expressed as:

[0069]

[0070] Similarly, for a D2D device pair i∈P t , introduce the index C i,D ∈{0,1}, when C i,D =1 indicates that the D2D model transmission is successful, otherwise C i,D =0. This indicator can be expressed as:

[0071]

[0072] In step 2 of this embodiment, considering that the data heterogeneity of the device will have a negative impact on the efficiency of model training, a fairness constraint is introduced

[0073]

[0074] Where T represents the total number of training rounds, c i ∈[0,1) represents the ratio of the minimum number of communication rounds required to select device i to the total number of training rounds. The normalized product of the local dataset size and the upload success rate is used as a measure to describe the importance of the device, so we set where ν i Represents the weight of device i. In order to solve the above fairness constraints, a virtual queue D is introduced for device i∈C. i ,Right now

[0075] D i (t)=[D i (t-1)+c i -a i,t-1 ] + .

[0076] Among them, D i (t) represents the queue length at the beginning of round t, [x] + = max{x,0}, the global device is represented by D(t) = [D 1 (t),...,D C (t)], D(1)=0.

[0077] In step 3 of this embodiment, the D2D association scheduling strategy is optimized to formulate an optimization problem with the goal of minimizing global loss. The steps include:

[0078] Step 1-1): Assume that the federated learning rate satisfies Given a device scheduling matrix a t and the optimal global federated learning model w * , the convergence upper bound can be described as

[0079]

[0080] Among them, in order to meet the above global training requirements, the local loss function of device training satisfies ζ 1 and 2 is a non-negative constant, the global loss function L(w) is quadratically continuously differentiable and satisfies L and μ are non-negative constants for Lipschitz continuity and strong convexity, respectively, and

[0081] Step 1-2): To minimize the above convergence upper bound, we only need to minimize the interval

[0082] Rewrite A as Rewrite B as Furthermore, the optimization problem can be expressed as:

[0083]

[0084] Where T is the total number of training rounds, and N is the maximum number of devices that the base station can access at any given time.

[0085] Step 1-3): Model the problem as a graph theory problem. Devices are considered as vertices and potential matches between devices are considered as edges. Therefore, the optimization problem can be simplified to a constrained maximum weighted independent edge set problem, by finding a set of edges that do not share vertices to maximize the total weight.

[0086] In step 3 of this embodiment, during the training process, all devices participating in the training use the standard gradient descent method for local training. By introducing the scheduling strategy, the update rule of the global model w can be described as

[0087]

[0088] Among them, K i,B , K i,D and K j Respectively represent device B i,t , D i,t and the size of j’s local dataset, w i,t , w j,t denote the local model parameters of D2D device pair i and direct communication device j from round t, respectively.

[0089] In step 4 of this embodiment, the steps for solving the above maximum weighted independent edge set problem can be summarized as follows:

[0090] Step 2-1): Convert device i∈C to node u i ∈U to construct a general graph G(U) and define the vertex weights of device i

[0091] V i∈St =βK i (1-q i,B )+(1-β)D i (t)

[0092] Among them, β∈[0,1], K i represents the size of the local data set of device i. Without loss of generality, the edge weight of D2D device pair i is defined as

[0093] V i∈Pt =β(1-q i,B )(K i,B +K i,D (1-q i,D ) 2 )+(1-β)(D i,B(t)+D i,D (t))

[0094] Among them, D i,B (t) and D i,D (t) respectively represent equipment B i,t and D i,t Virtual queue.

[0095] Step 2-2): Perform symmetric operations on the nodes contained in U to obtain a new node set, denoted as V, and construct a bipartite graph G(U,V). For u i ∈U,v j ∈V, establish edge e(i,j), use KM algorithm to achieve the maximum matching M, sort M in descending order of edge weight, and take the Top-N elements to meet the constraint on the number of base station access devices.

[0096] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A joint method for D2D pairing and device scheduling for federated edge learning, characterized by: The steps of the joint method for D2D pairing and device scheduling for federated edge learning are as follows: S1: Federated learning is performed using a federated edge learning system. The federated edge learning system is a global model formed by collaborative training of an edge base station with limited wireless resources and several devices. The device set is C = {1, 2, ..., C}, C = |C|, and the base station can establish connections with N devices at the same time; the device set that directly communicates with the base station in the tth round is S t , the two devices constitute a D2D device pair set P t , wherein each of the D2D device pairs includes a device B communicating with the base station i,t and said B i,t Communication equipment D i,t , for i∈S t ∪P t , scheduling strategy a i,t ∈[0,1],a i,t =1 means entity i participated in training at training round t, otherwise it did not participate; The D2D device communication and the direct communication from the device to the base station are both transmitted in quasi-static Rayleigh fading using orthogonal frequencies; each device encapsulates the model parameters into a single data packet for transmission; the base station verifies the integrity of the data packet through a cyclic redundancy check, and discards the message if it is damaged or lost; Before each round of federated learning, the packet loss rate of direct communication between all devices and the base station and the communication packet loss rate of all feasible D2D associated device pairs are calculated respectively, and the availability constraint index is introduced to indicate whether the local model scheduling strategy of the device can be uploaded successfully; S2: Considering the negative impact of device data heterogeneity on model training, a virtual queue is introduced for each device in each round of device selection to implement fairness constraints; S3: Data quality, packet loss rate, and virtual queues are comprehensively considered as the measurement criteria and the basis for device pairing and selection, and the optimization problem is modeled as a constrained maximum weighted independent edge set problem; S4: D2D devices are paired and selected in each round of training through the bipartite graph maximum weight matching algorithm. The D2D device pairs and direct communication devices selected to participate in the training upload the local updated model to the base station for global aggregation, and broadcast the new global model to the selected devices for the next round of training until the model converges.

2. A joint method for D2D pairing and device scheduling for federated edge learning according to claim 1, characterized in that: In S1, the packet loss rate of direct communication between all devices and the base station and the communication packet loss rate of all feasible D2D associated device pairs are as follows: The packet loss rate of direct communication between device i and base station is expressed as: Among them, P i,B 、h i,B is the transmission power and channel gain of device i in direct communication with the base station, W B is the allocated communication bandwidth, N0 is the noise power spectrum density, and τ is the system-specific threshold parameter; The packet loss rate of D2D device communication is: Among them, P i,D 、h i,D is the transmission power and channel gain of device i for D2D communication, W D is the dedicated bandwidth for D2D communication, τ′ is the threshold parameter related to the reverse coding gain of D2D communication; Define the model transmission index C from device to base station i,B ∈{0,1}, when C i,B =1, it means the local model of device i is uploaded successfully, otherwise C i,B =0, the indicator can be expressed as: For a D2D device pair i∈P t , introduce the indicator C i,D ∈{0,1}, when C i,D =1 indicates that the D2D model transmission is successful, otherwise C i,D =0, the indicator can be expressed as:

3. A joint method for D2D pairing and device scheduling for federated edge learning according to claim 1, characterized in that: In S2, considering that the data heterogeneity of the device will have a negative impact on the efficiency of model training, the fairness constraint is introduced: Where T represents the total number of training rounds, c i ∈[0,1) represents the ratio of the minimum number of communication rounds required to select device i to the total number of training rounds; in order to solve the above fairness constraints, a virtual queue D is introduced for device i∈C. i ,Right now: D i (t)=[D i (t-1)+c i -a i,t-1 ] + . Among them, D i (t) represents the queue length at the beginning of round t, [x] + = max{x,0}, the global device is represented by D(t) = [D1(t),...,D C (t)], D(1)=0.

4. A joint method for D2D pairing and device scheduling for federated edge learning according to claim 1, characterized in that: In S3, by optimizing the D2D association and scheduling strategy, an optimization problem is formulated with the goal of minimizing the global loss, including the following steps: F1: Federated learning rate satisfies Given a device scheduling matrix a t and the optimal global federated learning model w * , the convergence upper bound is described as: Among them, in order to meet the above global training requirements, the local loss function of device training satisfies ζ1 and ζ2 are non-negative constants, and the global loss function L(w) is quadratically continuously differentiable, satisfying L and μ are non-negative constants for Lipschitz continuity and strong convexity, respectively, and F2: To minimize the convergence upper bound, we only need to minimize the interval Rewrite A as Rewrite B as Furthermore, the optimization problem is expressed as: Where T is the total number of training rounds, N is the maximum number of devices that the base station can access at any given time; F3: The optimization problem is modeled as a graph theory problem, where devices are regarded as vertices and potential matches between devices are regarded as edges. Therefore, the optimization problem can be simplified to a constrained maximum weighted independent edge set problem, by finding a set of edges that do not share vertices to maximize the total weight.

5. A joint method for D2D pairing and device scheduling for federated edge learning according to claim 1, characterized in that: In S3, all devices participating in training use standard gradient descent for local training; By introducing the scheduling strategy and considering the packet loss rate, the update rule of the global model w can be described as: Among them, K i,B , K i,D and K j Respectively represent device B i,t , D i,t and the size of j’s local dataset, w i,t , w j,t denote the local model parameters of D2D device pair i and direct communication device j from round t, respectively.

6. A joint method for D2D pairing and device scheduling for federated edge learning according to claim 1, characterized in that: In S4, the steps to solve the above maximum weighted independent edge set problem are as follows: Step 1): Convert device i∈C to node u i ∈U to construct the general graph G(U) and define the vertex weight of device i: Where β∈[0,1]; without loss of generality, the edge weight of D2D device pair i is defined as: Among them, D i,B (t) and D i,D (t) respectively represent equipment B i,t and D i,t Virtual queues; Step 2): Perform symmetric operations on the nodes contained in U to obtain a new node set, denoted as V, and construct a bipartite graph G(U,V). i ∈U,v j ∈V, establish edge e(i,j), use the bipartite graph maximum weight matching algorithm to achieve the maximum matching M, and sort M in descending order of edge weight, and take the Top-N elements to meet the constraint on the number of base station access devices.

Images