Federated learning for generating synthetic mobility data

Abstract

The invention relates to a method for generating synthetic data from mobility data, the method comprising the following steps, implemented in equipment that is capable of communicating with a server of a cellular network, the method comprising the following steps: - obtaining (201; 202) mobility data; - determining (203) optimised local parameters for a module for generating synthetic data from the obtained mobility data; - transmitting (204) the optimised local parameters to the server; - receiving (220) optimised global parameters from the server; - updating (221) the module on the basis of the received optimised global parameters; - generating (222) synthetic data by the updated module.

Description

Federated learning for synthetic mobility data generation

[0001] The invention relates to the field of generating synthetic data representative of the mobility of equipment connected to a telecommunications network.

[0002] Its aim is in particular to enable federated learning for the generation of such synthetic data. State of the art

[0003] Entities connected to a telecommunications network, referred to hereinafter as equipment, such as smartphones, vehicles, connected objects accessing a mobile network such as cellular, for example, are capable of generating data representative of their mobility. Each piece of equipment may in particular collect, during its movement in the telecommunications network, data describing the trajectory of the equipment as well as metrics representative of the equipment's environment.

[0004] The trajectory may correspond to a series of geolocated points, through which the equipment passed during its journey, and the metrics are descriptive data of the customer's environment for each of these points. The metrics may in particular be descriptive of network states at each point. For this purpose, the metrics may include a signal-to-noise ratio, or SNR for "Signal to Noise Ratio" in English, a latency, a throughput value, etc.

[0005] Such metrics can have many advantages, for several network actors in particular:- for the network manager, because the metrics can be used to optimize the network;- when the equipment is a vehicle or a terminal on board a vehicle, for other vehicles connected to the network, the metrics can be used for informative purposes;- when the entity is a vehicle or a terminal on board a vehicle, for a transport actor who can use the metrics for information and prevention.

[0006] However, it is not desirable for a device to share its exact routes and metrics, as this data is private and potentially confidential.

[0007] It is known to overcome this problem to generate synthetic data from the collected mobility data. For this purpose, the mobility data is centralized in a central server, which is able to generate synthetic data, which can then be exploited by other actors, the synthetic data not including confidential private data.

[0008] However, such a solution requires each device to transmit the actual mobility data, which is confidential, to the central server, which poses security concerns. Indeed, an attacker can intercept this confidential data, and corrupt (modify) it or misuse it.

[0009] Furthermore, the transmission of real mobility data by each device to the centralized server, then the sending of synthetic data to other network players, causes significant traffic, which is therefore costly and can lead to network congestion.

[0010] The invention offers a solution which does not have the drawbacks of the state of the art.

[0011] To this end, according to a functional aspect, the invention relates to a method for generating synthetic data from mobility data, the method comprising the following steps, implemented in equipment capable of communicating with a server of a cellular network: - obtaining mobility data; - determining optimized local parameters of a module for generating synthetic data from the mobility data obtained; - transmitting the optimized local parameters to the server; - receiving optimized global parameters from the server; - updating the module on the basis of the optimized global parameters received; - generating synthetic data by the updated module.

[0012] Thus, the synthetic data generation module is updated from optimized global parameters, received from a server capable of receiving optimized local parameters from several devices accessing a telecommunications network. A federated learning of the parameters of the synthetic data generation module is thus implemented. Federated learning allows all the devices participating in it to have the same optimized global parameters, regardless of their computing resources: it thus allows on the one hand that the devices participating in the federated learning have optimized global parameters, but also, on the other hand, it allows the synthetic mobility data generated locally to be generated from the same generative model implemented by the respective synthetic data generation modules of the devices.Synthetic mobility data can thus be accumulated and compared, as they come from the same generative model, even when they come from multiple devices, which allows the exploitation of large quantities of synthetic data. In addition, federated learning accelerates the convergence of generative models of synthetic mobility data implemented locally in the devices, compared to a local learning of the generative model by each of the devices.

[0013] According to embodiments, the method may further comprise transmitting the generated synthetic data to another device accessing the cellular network, and / or to the server, and / or to another server of the cellular network.

[0014] Thus, the synthetic data generated can be used by other entities, without these entities accessing confidential data of the user of the equipment, since only synthetic data is transmitted.

[0015] Additionally, the method may further comprise an evaluation of a performance of the updated synthetic data generation module, from the generated synthetic data and the mobility data, and the generated synthetic data are transmitted to the other equipment accessing the cellular network, and / or to the server, and / or to the other server of the cellular network, only if the evaluated performance is greater than a performance threshold.

[0016] This ensures high performance when generating synthetic data. Only high-quality synthetic data, i.e., data generated by a high-performance synthetic data generation module (whose performance evaluation is above the performance threshold), is shared with other entities.

[0017] According to embodiments of the invention, the steps of determining optimized local parameters, transmitting the optimized local parameters, receiving the optimized global parameters and updating the synthetic data generation module constitute a learning cycle, and, at the end of a current learning cycle, a following learning cycle is implemented, new optimized local parameters being determined from the synthetic data generation module updated during the current learning cycle.

[0018] This makes it possible to iterate the learning cycles of federated learning, which allows for continuous improvement of the optimized global parameters, and which facilitates convergence of the synthetic data generation model of each device.

[0019] According to embodiments, the synthetic data generation module may implement a generative adversarial network comprising a generator module and a discriminator module, and the optimized local parameters may be determined by unsupervised learning based on at least a portion of the mobility data used as training mobility data.

[0020] A generative adversarial network has the advantages of simplicity of implementation as well as non-reversibility of the synthetic data generated: the confidentiality associated with mobility data is thus reinforced for the user of the equipment.

[0021] Additionally, the optimized local parameters may include first optimized local parameters of the generator module and / or second optimized local parameters of the discriminator module.

[0022] Thus, federated learning makes it possible to improve the performance of a generative model implemented locally by the equipment.

[0023] Additionally, the generator module and / or the discriminator module can be an artificial neural network.

[0024] This makes it possible to generate synthetic data from complex mobility data, for example including several types of descriptive data of the equipment's environment.

[0025] According to embodiments, the mobility data may comprise at least a first time series of equipment position data and descriptive data of the equipment environment, and the synthetic data may comprise at least a second time series of synthetic position data and synthetic data descriptive of the equipment environment, the synthetic position data of the second time series being different from the position data of the first time series.

[0026] Thus, the synthetic data does not correspond to real trajectories of the equipment, which improves the confidentiality of the synthetic data compared to the obtained mobility data.

[0027] In addition, the descriptive data of the environment may include one or more of the following data: - data representative of a quality of service of the telecommunications network at a given position; - data representative of traffic between the equipment and the telecommunications network for a given position; - descriptive data of a place located at a given position; and / or - descriptive data of meteorological conditions at a given position.

[0028] Thus, synthetic data can be obtained from mobility data enriched with descriptive data of the environment. The invention can thus be used for the generation of synthetic data used in various applications. The list of descriptive data mentioned above is not exhaustive: other descriptive data of the environment can be integrated, according to the invention, into the mobility data.

[0029] According to embodiments, the method may comprise collecting raw mobility data, the mobility data may be obtained from the raw mobility data by:- determining missing data from the raw mobility data and adding the missing data to the raw mobility data to obtain the mobility data; and / or- normalizing the raw mobility data to obtain the mobility data.

[0030] This ensures that the data used as input to the synthetic data generation module is complete and / or standardized, which improves the generated synthetic data.

[0031] According to a hardware aspect, the invention relates to equipment comprising a network interface capable of communicating with a server of a cellular network and a processor configured to:- obtain mobility data;- determine optimized local parameters of a synthetic data generation module from the obtained mobility data;- transmit optimized local parameters to the server via the network interface;- receive optimized global parameters from the server via the network interface;- update the module on the basis of the received optimized global parameters;- generate synthetic data by the updated module.

[0032] According to embodiments, the equipment may further comprise at least one sensor or may execute at least one application, capable of obtaining at least part of the mobility data.

[0033] According to another material aspect, the invention also relates to a computer program capable of being implemented on the defined equipment, the program comprising code instructions which, when the program is executed by a processor, carry out the steps of the defined method.

[0034] Such programs can use any programming language. They can be downloaded from a communications network and / or stored on a computer-readable medium.

[0035] According to another material aspect, the invention relates to a data recording medium on which at least one series of program code instructions has been stored for the execution of the method defined above.

[0036] According to another material aspect, the invention also relates to a system comprising a server and a plurality of equipments defined above, capable of communicating in a cellular network.

[0037] The invention will be better understood upon reading the following description, given by way of example and with reference to the appended drawings in which:illustrates an example of a telecommunications network according to embodiments of the invention;illustrates the steps of a method for generating synthetic mobility data, according to embodiments of the invention;illustrates a module for generating synthetic mobility data according to embodiments of the invention;illustrates an artificial neural network forming a generator module of a generative adversarial network for generating synthetic mobility data, according to embodiments of the invention.

[0038] Illustrates an example environment for implementing the invention according to embodiments.

[0039] A telecommunications network 100 according to the invention comprises at least one first server 110 and at least one first base station 101 associated with the first server.

[0040] The telecommunications network 100 may be a cellular type mobile network, comprising a plurality of base stations, including the first base station 101 shown in the, each base station being capable of covering a respective radio coverage area, called radio zone in the following, that is to say of communicating bidirectionally with equipment located in the radio zone that it covers. In the, the first base station 101 covers a first radio zone 102 in which there are a first equipment 111.1 and a second equipment 111.2.

[0041] For example, the telecommunications network 100 may be a 3G, 4G, 5G or any other cellular network, in particular a later generation network.

[0042] Each base station of the telecommunications network is associated with at least one radio cell covered in the radio area of ​​the base station. For example, the first base station 101 may be associated with at least one radio cell, or even several radio cells (a 2G cell, a 3G cell and a 4G cell for example).

[0043] In the following, it is considered, for illustrative purposes only, that the base station 101 is associated with only one radio cell.

[0044] The first equipment 111.1 and the second equipment 111.2 are mobile equipment in the radio zone 102: it may be portable equipment, such as a Smartphone, embedded in a vehicle, or the vehicle itself, or any other equipment capable of accessing a telecommunications network and being moved. Only two pieces of equipment are shown in the, in order to simplify the understanding of the invention: however, a larger set of equipment, which may comprise several tens or hundreds of pieces of equipment, may simultaneously access the first server 110 via the first base station 101.

[0045] At a given time, a first set of equipment comprises the mobile equipment located in the radio zone 102, capable of accessing the first server 110 and locally implementing a generative model of synthetic mobility data in a synthetic mobility data generation module, described below. The first set of equipment thus varies with time: an equipment can leave the first radio zone 102 to enter another radio zone associated with another base station of the telecommunications network 100, and / or a new equipment can enter the first radio zone 102.

[0046] The first equipment 111.1 comprises: - computing resources, which may comprise a processor 113.1, the processor 113.1 comprising a synthetic mobility data generation module capable of implementing a generative model of synthetic mobility data, as described below;

[0047] - storage resources 112.1 such as a memory of the “Random Access Memory” type, RAM, or a memory of the “Read Only Memory” type, ROM, or any other type of memory (Flash, EEPROM, etc.). Alternatively, the memory 112.1 comprises several memories of the aforementioned types. The memory 112.1 comprises at least one non-volatile memory in which are stored, temporarily or permanently, mobility data, raw or pre-processed, which may include position data from a location module 116.1, optionally enriched with descriptive data of the environment from a set 114.1 of at least one sensor of the first equipment 111.1, and / or an application executed on the first equipment 111.1; - the set 114.1 comprises at least one sensor capable of obtaining raw data descriptive of the environment of the vehicle, also called “metrics” in the following.The term "raw data" means data captured directly by a sensor from the environment of the equipment 111.1 and to which the pre-processing described with reference to has not been applied. According to the invention, in addition to or as a variant of the data collected by the assembly 114.1, the descriptive raw data of the invention may be collected by an application, or several applications, installed and executed on the first equipment 111.1. Note that the acquisition of the descriptive raw data of the environment of the vehicle is optional according to the invention: the first equipment 111.1 may not include the assembly 114.1, which is therefore also optional; - a network interface 115.1 capable of exchanging bidirectionally with the base station 101; - a location module 116.1 capable of determining position data which is a location of the equipment 111.1 at a given time, the location being able to comprise a set of coordinates locating the first equipment 111.1 in the radio zone 102. The location module 116.1 can for example be a satellite geolocation module.

[0048] Thus, the first equipment 111.1 is capable of acquiring raw mobility data comprising a first time series of position data associated respectively with descriptive data of the environment of the first equipment 111.1.

[0049] As described in the following, the synthetic data generation module included in the computing resources 113.1, can implement a generative model, for example resulting from machine learning, the parameters of the generative model being modified by training according to training mobility data which is part or all of the mobility data, raw or pre-processed.

[0050] The second device 111.2 also includes computing resources including a processor 113.2 that includes a synthetic data generation module, storage resources 112.2, an optional set 114.2 of at least one sensor, a network interface 115.2, and a location module 116.2. The resources of the second device 111.2 may differ from the resources of the first device 111.1: the memory 112.1 may have a storage capacity distinct from the storage capacity of the memory 112.2, the computing resources 113.1 may be more powerful, or less powerful, than the computing resources 113.2.

[0051] According to the invention, the first server 110 is a central unit which plays an orchestrator role in a federated learning to parameterize the synthetic data generation models of the first set of equipment, comprising the first equipment 111.1 and the second equipment 111.2. In the example considered here, the server 110 is associated with the first radio cell served by the first base station 101 in the first radio zone 102. In a variant not described, the server 110 can be associated with several first radio cells served by the first base station 101 in the first radio zone 102. In another variant not described, the server 110 can be associated with several radio cells served by several base stations of the telecommunications network 100 in several radio zones.

[0052] Since only one server is shown in the figure, to simplify the description of the invention, the first server 110 is called server 110 in the following.

[0053] The present invention presents the steps of a method for generating synthetic mobility data according to embodiments of the invention.

[0054] In a step 200, the first device 111.1 initializes parameters of the synthetic data generation module. No restrictions are attached to the initial parameters of the synthetic data generation module, the initial parameters being able to be stored before the first device 111.1 is put into operation, during the installation of the synthetic data generation module, or being received from the server 110 via the network interface 115.1.

[0055] In a step 201, the first device 111.1 obtains raw mobility data, the raw mobility data comprising time series in the following format , in which:- t is a time of collection of the raw mobility data;- (x,y) are the position data indicating the geographical position of the first equipment 111.1, at time t;- [ ] is a set of descriptive data of the environment at time t, also called metrics, and which are obtained by the set 114.1 of at least one sensor and / or by one or more applications executed on the first device 111.1. The set comprises n metrics, with n integer greater than or equal to 1.

[0056] Note that according to the invention, the descriptive data of the environment making it possible to enrich the raw mobility data are optional. Thus, according to certain embodiments, the raw mobility data do not include descriptive data of the environment and thus take the form of a time series of position data [t,(x,y)].

[0057] In the following, it is considered, for illustrative purposes, that raw mobility data also includes descriptive environmental data, or metrics.

[0058] The metrics may comprise any of the following data, or any combination obtained from the following data:- data representative of a quality of service of the telecommunications network for the position (x,y) at time t, which may comprise a ratio between a signal level and a noise level, also called SINR for “Signal to Interference + Noise Ratio” in English, an average value of a reference signal received power, also called RSRP, for “Reference Signal Receive Power” in English, and / or a technology among 3G, 4G and 5G technologies;- data representative of traffic between the equipment and the telecommunications network for the position (x,y) at time t;- descriptive data of a location situated at the position (x,y) at which the first equipment 111.1 is located at time t; and / or- descriptive data of weather conditions for the position (x,y) at time t.

[0059] The data listed above are non-exhaustive examples of data that can be integrated into mobility data as metrics.

[0060] Step 201 may in particular be implemented continuously, as the first equipment 111.1 moves in the first radio zone 102, and / or in other radio zones not shown in the figures. The first equipment 111.1 may thus collect raw mobility data for a duration greater than the duration of passage in the first radio zone 102, for example over a period of the order of a day, a week or a month.

[0061] In an optional step 202, the first device 111.1 can process the raw mobility data to obtain mobility data. Such processing can be considered as a pre-processing allowing to improve the raw mobility data before the implementation of the following steps, allowing a federated learning of the parameters of the synthetic data generation module.

[0062] The processing step 202 may comprise one or other of the following sub-steps: - a first sub-step of determining missing data in the raw mobility data. The determination of missing data may be implemented by linear interpolation, or by smoothing which consists of transforming the discrete raw data into functions, the functions then being used to determine the missing values. In the case where the quantity of raw mobility data is large, for example greater than a first threshold, and the proportion of missing data is small, for example less than a second threshold, the missing data may be ignored and the first sub-step is not implemented. The missing data may be determined both for the position data (x,y) but also for all the metrics [ ], where the raw mobility data includes such metrics; and / or

[0063] - a second sub-step of normalization of the raw mobility data. Normalization consists of bringing data of different natures and amplitudes to identical or close intervals. For example, the position data and each metric are brought to the same interval between 0 and 1. Such normalization subsequently allows the improvement of the machine learning implemented locally to determine optimized local parameters of the synthetic mobility data generation module. Indeed, it makes it possible to avoid learning biases linked to the amplitudes, the trained generative model being able to give more weight to the data whose amplitude is the greatest. No restriction is attached to the normalization technique implemented by the first device 111.1, which can be based for example on the MinMaxScaler() or RobustScaler() function, or which can be based on a more complex technique such as mode normalization, called "Mode Normalization" in English.

[0064] Thus, step 202 can make it possible to obtain mobility data which is complete and normalized from the raw mobility data.

[0065] Alternatively, when step 202 is not applied, the raw mobility data is the mobility data mentioned in the remainder of the description.

[0066] In a step 203, the first device 111.1 determines optimized local parameters of the synthetic data generation module, from the mobility data obtained in the previous steps. The synthetic data generation module can in fact implement a generative model of synthetic data which can be trained by machine learning, according to the mobility data obtained in the previous steps. At the start of step 203, the synthetic data generation module is updated from the initial parameters: the training by machine learning from the mobility data obtained allows a modification of the initial parameters to obtain the optimized local parameters.

[0067] “Synthetic mobility data” means a time series of the same format than the mobility data acquired in the previous steps, but distinct from the acquired mobility data. The acquired mobility data are real data, which thus include confidential personal data of the user of the first device 111.1, such as his journeys in particular (the successive positions occupied over a given period). On the contrary, the synthetic mobility data comprise a succession of synthetic position data of which at least a part has not been occupied by the user of the first device 111.1, in association, optionally, with metrics corresponding respectively to the synthetic position data. The synthetic mobility data thus do not make it possible to know the exact journeys made by the user of the first device 111.1.The synthetic mobility data may in particular comprise a time series of position data which are not consecutive, and which thus form a point cloud in the radio zone 102, and not continuous trajectories like the position data of the real mobility data.

[0068] The present invention enables federated learning training of a synthetic data generative model. Once trained, the synthetic data generative model can generate synthetic mobility data, which illustrates the evolutions of the metrics over time and space, without however containing confidential personal information of the user of the first equipment 111.1.

[0069] There are no restrictions on the type of synthetic data generative model implemented by the synthetic data generation module. The synthetic data generative model can be, for example:- a Generative Adversarial Network, also called a GAN;- a Variational Auto-Encoder, or VAE;- a Flow-Based Generative Model; or- a Diffusion Model.

[0070] According to preferred embodiments, the generative model of synthetic mobility data is a GAN. Indeed, a GAN has the advantage of being easier to implement than an auto-encoder, which requires great complexity to obtain convincing results.

[0071] In the remainder of the description, it is considered, for illustrative purposes only, that the generative model of synthetic mobility data is a GAN. However, the invention can be based on any type of generative model.

[0072] This presents the structure of a generative model 300 of the GAN type, according to embodiments of the invention.

[0073] As indicated previously, the GAN 300 can be implemented by the synthetic data generation module included in the computing resources 113.1 of the first equipment 111.1. According to the invention, the equipment of the first set of equipment participating in the federated learning each comprises its own synthetic data generative model, therefore its own GAN 300, implemented by its respective computing resources.

[0074] In accordance with the structure of a GAN, the GAN 300 comprises a generator module 301 trained to generate synthetic mobility data 311 denoted X', from noise data 310, denoted Z. The term "noise data" means any data of the same format as the real mobility data, but generated randomly or semi-randomly according to statistical laws such as the normal law, and not corresponding to any real data. The noise data 310 are therefore in no way correlated with real mobility data 312, denoted X, used as training data, for the unsupervised learning of the parameters of the GAN. The real mobility data 312 used as training data may be all or part of the real mobility data from the previous step 202.

[0075] The GAN 300 further comprises a discriminator module 302, and a loss function module 303. The discriminator module 302 is capable of receiving as input the synthetic mobility data 311 as well as the real training mobility data 312. The discriminator module 302 is configured to classify the received data 311 and 312, into real data or synthetic data.

[0076] According to the principles of a GAN, the generator module 301 is trained so as to minimize the capacity of the discriminator module 302 to correctly classify the real and synthetic mobility data received as input from the discriminator module 302. At the end of one or more classifications carried out by the discriminator module 302, the loss function module 303 evaluates the performance of the GAN 300, and modifies the parameters of the GAN 300 according to the performance thus evaluated. The training process for learning the optimized local parameters during step 203 is thus iterative. After training the GAN 300 with the real training mobility data 312, optimized local parameters are thus obtained, following successive iterations.

[0077] According to embodiments of the invention, the parameters of the GAN 300 are defined by parameters of the generator module 301. In this case, the discriminator module 302 may be defined in advance and its parameters are not trained. Alternatively, the parameters of the GAN 300 comprise both parameters defining the generator module 301 but also parameters defining the discriminator module 302. The discriminator module 302 and the generator module 301 are in this case trained during unsupervised learning on the basis of the actual training mobility data 312.

[0078] The generator module 301 and / or the discriminator module 302 may implement an artificial neural network. In the following, it is considered that the generator module 301 implements a first artificial neural network and the discriminator module 302 implements a second artificial neural network.

[0079] The GAN 300 is trained to obtain the optimized local parameters in step 203 by solving a game theory problem, defined as follows:

[0080] with

[0081] Or represents the distribution of real mobility data 312, represents the distribution of the synthetic mobility data 311 generated by the generator module 301, and D(x) represents the classification made by the discriminator module 302. V(D,G) is thus evaluated by the loss function module 303.

[0082] The objective of the generator module 301 is to learn to deceive the discriminator module 302, that is to say to produce synthetic mobility data X' such that p(X')=p(G(z)) is not differentiable from p(X).

[0083] The discriminator module 302 classifies the received input data into two categories: real or synthetic / fake. Real data is classified into a class 1 and fake data is classified into a class 0.

[0084] The generator module 301 is thus trained to generate synthetic data which appears real, from the point of view of the discriminator module 302.

[0085] Each of the first and second artificial neural networks may be multi-layer neural networks, comprising one or more neurons per layer, which are connected to neurons in subsequent layers.

[0086] This presents the structure of a first artificial neural network forming a generator module 301 of a GAN 300 capable of generating synthetic mobility data, according to embodiments of the invention.

[0087] The first artificial neural network comprises a first layer 400 comprising three neurons 401, 402 and 403, a second layer 410 comprising two neurons 411 and 412, and a third layer 420 comprising a single neuron 421.

[0088] Such an architecture is given for illustrative purposes, and no restriction is attached to the number of layers or the number of neurons per layer, of the first neural network forming the generator module 301.

[0089] The parameters defining the generator module 301 are the weights W= [w 11 1 ;w 12 1 ;w 21 1 ;w 22 1 ;w 31 1 ;w 32 1 ;w 11 2 ;w 12 2 ] and the biases B=[b1 1 ;b2 1 ;b1 2 ], in which :- w 11 1 is the weight assigned to the output of node 401 in node 411 ;- w 121 is the weight assigned to the output of node 401 in node 412 ;- w 21 1 is the weight assigned to the output of node 402 in node 411 ;- w 22 1 is the weight assigned to the output of node 402 in node 412 ;- w 31 1 is the weight assigned to the output of node 403 in node 411 ;- w 32 1 is the weight assigned to the output of node 403 in node 412 ;- w 11 2 is the weight assigned to the output of node 411 in node 421 ;- w 12 2 is the weight assigned to the output of node 412 in node 421 ;- b1 1 is the bias added to node 411 ;- b2 1 is the bias added to node 412 ;- b1 2 is the bias added to node 421.

[0090] Biases are represented by the dotted arrows in the figure. Note that the parameters may not include biases, which are optional, and may only include the respective weights of the nodes in the neural network.

[0091] Thus, the input Z is propagated from the first layer 400 to the second layer 410: let z i , for i varying from 1 to 3 the respective outputs of neurons 401 to 403 of the first layer 400. Then, the output y1 of neuron 411 is equal to , sigma denoting an activation function, such as a ReLU, softmax, or any other activation function.

[0092] Similarly, the output y2 of neuron 412 is equal to .

[0093] The output X' is then equal to .

[0094] Thus, the synthetic mobility data generative model is initialized with parameters W in and B in, and, during the training of step 203, the first device 111.1 determines optimized local parameters W opt and B opt . The performance of the synthetic data generative model configured with W opt and B opt are better than in the initial setting. For this purpose, training can be based on the gradient descent method, applied by the loss function module 303.

[0095] According to embodiments, the discriminator module 302 is a second artificial neural network, which may comprise an input layer capable of receiving as input the synthetic mobility data 311 or the real training mobility data 312, an output layer capable of determining as output a class 0 or 1, and which may or may not comprise one or more intermediate layers between the input layer and the output layer. Again, the parameters defining the discriminator module are a weight vector W and an optional bias vector B, the number of components of which depends on the number of layers and the number of neurons per layer. Thus, the parameters of the GAN 300 may comprise a weight vector W G from the generator module 301, a bias vector B G from the generator module 301, a weight vector W D of the discriminator module 302 and a bias vector B Dof the discriminator module 302. In this case, during step 203, the first device 111.1 determines the optimized local parameters W G,Lopt , W D,Lopt , B G,Lopt , B D,Lopt , the GAN 300 being initialized at the start of step 203 with initial parameters W G,in , W D,in , B G,in , B D,in .

[0096] Note that, as previously stated, the generative model may be different from a GAN 300, and may be, for example, an autoencoder, a diffusion model, or a flow-based generative model. According to each of these alternatives, the generative model may also comprise one or more artificial neural networks, defined by parameters that are optimized to obtain optimized local parameters.

[0097] Referring again to the, step 203 of obtaining the optimized local parameters, by training the GAN 300, in particular on the basis of the real training mobility data, has been described.

[0098] Following step 203, the first device 111.1 transmits the optimized local parameters to the server 110 in a step 204, via the network interface 115.1 of the first device 111.1.

[0099] In a step 210, the server 110 receives the optimized local parameters transmitted by the first equipment 111.1. In a step 211 occurring before or after step 210, the server 110 receives the optimized local parameters by the second equipment 111.2. It is thus understood that the second equipment 111.2 implements steps identical to steps 200 to 204 described with reference to the first equipment 111.1. However, the resources of the second equipment 111.2 and the actual mobility data obtained by the second equipment 111.2 being distinct from those of the first equipment 111.1, the optimized local parameters determined by the second equipment 111.2 and received in step 211 by the server 110 may be different from the optimized local parameters determined by the first equipment 111.1 and received in step 210 by the server 110.

[0100] More generally, the server 110 receives the optimized local parameters of some or all of the equipment of the first set that participate in the federated learning. For example, the server 110 may preselect some of the equipment of the first set so that they locally determine the optimized local parameters according to steps 200 to 204, the other equipment of the first set being exempted from implementing these steps. Such a selection may for example be based on the quantity of real training mobility data available to each equipment of the first set, and / or the computing resources available to each equipment of the first set.

[0101] When the optimized local parameters have been received from some or all of the equipment in the first set, the server 110 implements step 212. In the example considered here, with only two equipment in the first set, the server 110 implements step 212 following the reception of optimized local parameters from the first equipment 111.1 and from the second equipment 111.2.

[0102] In step 212, the server 110 aggregates the optimized local parameters received from some or all of the equipment in the first set to obtain optimized global parameters. No restriction is attached to the function applied to the received optimized local parameters to obtain the optimized global parameters. In the example where the synthetic mobility data generation module is based on the GAN 300 considered above, the optimized global parameters may be vectors W G,Gopt , W D,Gopt , BG,Gopt , B D,Gopt .

[0103] The determination of the optimized global parameters can be based on an average of each of the received optimized local parameters. For example, for the parameter w 11 1 described previously, considering that the parameter w 11 1i was received from each device of index i, i varying between 1 and an integer p corresponding to the number of devices having returned the optimized local parameters (2 in the example described previously), the optimized global parameter w 11,Gopt 1 is equal to .

[0104] As an alternative to the use of the average described above, the determination of the optimized global parameters can be based on other techniques for aggregating the optimized local parameters, for example on a FedMA algorithm, for "Federated Matching Average" in English, or on an RFA algorithm, for "Robust Federated Aggregation" in English.

[0105] After obtaining the optimized global parameters, the server 110 transmits, in a step 213, the optimized global parameters to the first device 111.1, as well as to any other device of the first set (therefore also to the second device 111.2 in the example of the) participating in the federated learning.

[0106] At a step 220, the first device 111.1 receives the optimized global parameters from the server 110, via the network interface 115.1.

[0107] In a step 221, the first device 111.1 updates the synthetic data generation module 300 on the basis of the received optimized global parameters. In the example of the GAN 300, the generator module 301 and the discriminator module 302 can thus be updated on the basis of the vectors W G,Gopt , W D,Gopt , B G,Gopt , B D,Gopt forming the optimized global parameters.

[0108] Following step 221, the first device 111.1 can be selected by the server 110 to participate in a new learning cycle of the federated learning:

[0109] - on the basis of new raw mobility data acquired during step 201; and / or- on the basis of the raw mobility data acquired previously, by locally training, during step 203, the synthetic mobility data generation model whose initialization is this time the update resulting from step 221, in order to obtain new optimized local parameters. The new optimized local parameters are transmitted during step 204 to the server 110, and, following their aggregation with other new optimized local parameters of other equipment of the first set, new optimized global parameters are obtained in step 220. Steps 221 to 223 are repeated on the basis of these new optimized global parameters.

[0110] If the first device 111.1 is not selected for the new federated learning cycle (for example because the server 110 has selected other devices for the new federated learning cycle), it can still receive new optimized global parameters during a new step 220, and can thus update the synthetic data generation module during a new step 221.

[0111] Federated learning may thus comprise several learning cycles which are implemented by iterating steps 203, 204, 220 and 221. No restriction is attached to the number of learning cycles, federated learning being terminated when a stopping criterion is reached, the stopping criterion being able to be:- a predefined number of learning cycles; and / or- when a value of the loss function implemented by the loss function module 303 is lower than a predefined threshold.

[0112] At a step 222 at the end of the federated learning (therefore the last learning cycle implemented), the first device 111.1 generates synthetic mobility data by the synthetic data generation module updated during step 221 (of the last learning cycle), and from part or all of the raw mobility data collected in step 201, and optionally pre-processed during step 202.

[0113] In a step 223, the first equipment 111.1 evaluates a performance of the synthetic data generation module 301 updated during step 221, from the synthetic data generated during step 222 and from part or all of the raw mobility data collected in step 201, and optionally pre-processed during step 202.

[0114] The performance of the synthetic data generation module 301 is evaluated by the loss function implemented by the loss function module 303 described above. The performance can thus be evaluated by the function V(D,G) described above.

[0115] If the evaluated performance is lower than a given performance threshold, the first equipment 111.1 may not transmit the generated synthetic mobility data, which is indicated by a step 230 of absence of transmission on the. In this case, the synthetic data generative model may be improved during a next federated learning, again comprising one or more learning cycles, initiated for example when the first equipment moves into another radio zone of the cellular network or at the end of a given period.

[0116] If the performance evaluated in step 223 is greater than the given performance threshold, then the first equipment 111.1 can transmit, in a step 224, the synthetic mobility data generated during step 222, to another equipment accessing the telecommunications network, for example the second equipment 111.2, and / or the server 110, and / or another server of the telecommunications network. The synthetic mobility data can thus be used by these entities, without disclosing confidential information of the user of the first equipment 111.1.The transmission in step 224 may be immediately subsequent to a positive comparison of the performance evaluated with the performance threshold in step 223, or may further be conditioned by one or more of the following conditions, given for illustrative purposes and listed in a non-exhaustive manner: - the transmission of step 224 is conditioned by the reception of a request for synthetic data from the other equipment accessing the server, from the server 110 and / or from the other server of the network 100; - the transmission of step 224 is implemented periodically, with a transmission at the end of each period defined by a predetermined frequency. In this case, the transmission of step 224 is implemented at the end of a current period.

[0117] Note that the other equipment in the first set implements the same steps 220 to 224 locally, following the transmission of the global parameters optimized by the server 110 during step 213.

[0118] Thus, at a step 225, the first device 111.1 can receive synthetic mobility data from another device, for example from the second device 111.2, which has locally implemented the same steps 220 to 224 and which has therefore also participated in the federated learning.

[0119] As previously indicated, a new federated learning can be initiated when the first equipment moves into a radio zone different from radio zone 102, or at the expiration of a given period, in which case the steps described previously are repeated for the new federated learning.

[0120] The invention thus allows the implementation of federated learning for the configuration of a synthetic mobility data generation module. According to such an approach, the learning is distributed between the different devices participating therein. The devices thus collaborate to construct a global generative model, the collaboration only involving the transfer, to the server 110 common to several devices, of parameters of a local model, the parameters being optimized locally by each device during step 203.

[0121] Federated learning allows all participating devices to have the same optimized global parameters, regardless of their computing resources: it allows the devices in the first set participating in federated learning to use globally optimized parameters, but also, on the other hand, it allows the locally generated synthetic mobility data to be generated from the same generative model. Synthetic mobility data can thus be accumulated and compared, even when they come from several devices, which makes it possible to exploit large quantities of synthetic mobility data.

[0122] Additionally, federated learning accelerates the convergence of generative models of synthetic mobility data implemented locally in devices, compared to only local learning of the generative model by each device.

[0123] Thus, the generated synthetic data can be used by other entities, without these entities accessing confidential data of the user of the equipment generating the synthetic data, since only synthetic data is transmitted.

[0124] Examples of exploitation of synthetic mobility data according to the invention are described below, for illustrative purposes.

[0125] According to a first example, the generated synthetic mobility data can be used for the improvement and information on the conditions of the telecommunications network 100. The equipment of the first set is vehicles or telephones embedded in vehicles, and each equipment comprises an application capable of measuring, periodically, the quality of service of the telecommunications network, for each position occupied by the equipment, the quality of service being integrated into the raw mobility data as a metric.

[0126] Synthetic mobility data are then generated according to the invention, and can be transmitted to a server responsible for managing the network 100, for the purposes of improving the quality of service in the network 100, and / or to other equipment which can then adapt their respective routes according to the network quality indicated in the synthetic mobility data.

[0127] According to a second example, the generated synthetic mobility data are used to improve road traffic. In this second example, the quality of service of the network of the first example is replaced by data representative of a traffic quality (indicating a fluid traffic state, a congested state, works), for each position occupied by a device. The data representative of the traffic quality are thus integrated as metrics in the raw mobility data. Then, synthetic mobility data are generated according to the invention and can be used by a given server to provide information on the traffic at certain positions and / or for traffic planning.

[0128] According to a third example, the synthetic mobility data are used for tourism purposes. The equipment integrates an application capable of providing tourism quality indicators for a route. For example, during a journey, a piece of equipment can collect metrics at given positions along the route, such as climate, cleanliness, quietness, shops, etc. Synthetic mobility data are then generated according to the invention, and can be transmitted to other equipment integrating the same application, to guide them towards routes with the best tourism indicators.

[0129] According to a fourth example, the synthetic mobility data is used for environmental purposes. The devices of the first set may have an application measuring the environmental impact associated with the current journey, the environmental impact being the carbon emissions generated, integrated into the raw mobility data as metrics. The synthetic mobility data is then generated according to the invention, then transmitted to other devices on board vehicles to raise awareness of taking the most environmentally friendly routes.

Claims

Method for generating synthetic data from mobility data, the method comprising the following steps, implemented in equipment (111.1; 111.2) capable of communicating with a server (110) of a cellular network (100): - obtaining (201; 202) mobility data; - determining (203) optimized local parameters of a synthetic data generation module (300) from the obtained mobility data; - transmitting (204) the optimized local parameters to the server; - receiving (220) optimized global parameters from the server; - updating (221) said module on the basis of the received optimized global parameters; - generating (222) synthetic data by said updated module. The method of claim 1, further comprising transmitting (224) the generated synthetic data to another device (111.2) accessing the cellular network, and / or to the server (110), and / or to another server of the cellular network. Method according to claim 2, further comprising an evaluation (223) of a performance of the updated synthetic data generation module, from the generated synthetic data and the mobility data, and wherein the generated synthetic data is transmitted (224) to the other equipment (111.2) accessing the cellular network (100), and / or to the server (110), and / or to the other server of the cellular network, only if the evaluated performance is greater than a performance threshold. Method according to one of the preceding claims, in which the steps of determining (203) optimized local parameters, transmitting (204) the optimized local parameters, receiving (220) the optimized global parameters and updating (221) the synthetic data generation module constitute a learning cycle, and in which, at the end of a current learning cycle, a following learning cycle is implemented, new optimized local parameters being determined from the synthetic data generation module updated during the current learning cycle. Method according to one of claims 1 to 4, wherein the synthetic data generation module (300) implements a generative adversarial network comprising a generator module (301) and a discriminator module (302), and wherein the optimized local parameters are determined by unsupervised learning based on at least part of the mobility data used as training mobility data (312). The method of claim 5, wherein the optimized local parameters comprise first optimized local parameters of the generator module (301) and / or second optimized local parameters of the discriminator module (302). Method according to claim 6, wherein the generator module (301) and / or the discriminator module (302) is an artificial neural network. Method according to one of the preceding claims, in which the mobility data comprises at least a first time series of position data of the equipment and descriptive data of the environment of the equipment (111.1; 111.2), and in which the synthetic data comprises at least a second time series of synthetic position data and synthetic data descriptive of the environment of the equipment, the synthetic position data of the second time series being different from the position data of the first time series. Method according to claim 8, in which the descriptive data of the environment comprises one or more of the following data:- data representative of a quality of service of the telecommunications network at a given position;- data representative of traffic between the equipment and the cellular network for a given position;- data descriptive of a place located at a given position; and / or- data descriptive of meteorological conditions at a given position. A method according to one of the preceding claims, comprising collecting (201) raw mobility data, wherein the mobility data is obtained from the raw mobility data by:- determining (202) missing data from the raw mobility data and adding the missing data to the raw mobility data to obtain the mobility data; and / or- normalizing (202) the raw mobility data to obtain the mobility data. Equipment (111.1; 111.2) comprising a network interface (115.1; 115.2) capable of communicating with a server (110) of a cellular network (100) and a processor (113.1; 113.2) configured to:- obtain mobility data;- determine optimized local parameters of a synthetic data generation module from the obtained mobility data;- transmit optimized local parameters to the server via the network interface;- receive optimized global parameters from the server via the network interface;- update said module on the basis of the received optimized global parameters;- generate synthetic data by said updated module. Equipment according to claim 11, further comprising at least one sensor (114.1; 114.2) or executing at least one application, capable of obtaining at least part of the mobility data. A computer program capable of being implemented in equipment (111.1; 111.2) as defined in claim 11 or 12, the program comprising code instructions which, when executed by a processor (113.1; 113.2), performs the steps of the method defined in one of claims 1 to 10. Data storage medium on which at least one series of program code instructions has been stored for executing the method according to one of claims 1 to 10. System comprising a server (110) and a plurality of equipment (111.1; 111.2) according to claim 11 or 12, capable of communicating in a cellular network (100).

Images