Computationally efficient federated learning on wireless server with privacy

Abstract

According to an example aspect of the present disclosure, there is provided a federated learning user equipment configured to perform transmitting, to a cellular base station over an air interface, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning user equipment, decrypting, using a first decryption key, second parameter data received by the federated learning user equipment from the cellular base station, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, and updating, as part of federated learning, the local model parameters of the local model of the federated learning user equipment with the global model parameters.

Description

COMPUTATIONALLY EFFICIENT FEDERATED LEARNING ON WIRELESS SERVER WITH PRIVACYRELATED APPLICATIONS

[0001] This patent application claims the benefit of priority of India Provisional Patent Application No. 202441077046 filed October 10, 2024, which is hereby incorporated by reference as if reproduced in its entirety.

[0002] FIELD

[0003] The present disclosure relates to federated learning, and in particular a computationally efficient federated learning method and system.

[0004] BACKGROUND

[0005] Federated learning is a sub-field of machine learning in which multiple entities, often referred to as clients, collaboratively train a machine learning model (e.g., an artificial neural network) while maintaining their data decentralized. An alternative term for federated learning is collaborative learning. When the clients are wireless devices, they may send, via a communication network, information on their local models (i.e., may send the gradients or values of the parameters of their local models) to a federated learning server, which combines the gradients or values of the parameters of the local models to obtain a global model update which is then delivered back to the clients (e.g., wireless devices). The clients (e.g., wireless devices) may then update the parameter of their local models using the global model update (e.g., the combined gradient or values of the parameters of the local models).

[0006] Sharing local parameters over a communication channel of a communication network for a federated learning iteration may lead to leakage of private information of clients, wherefore protecting this private information is of interest. For example, if the communication channel is a wireless channel, eavesdropping the channel, or infecting a radio node participating in maintaining the channel with malware, presents a risk of leakage of the information on a client's local model, which can be used to obtain private information of the client.

[0007] To safeguard against leakage of information of the local models of clients (e.g., wireless devices), secure aggregation and secure enclaves may be utilized.

[0008] SUMMARY

[0009] According to some aspects, there is provided the subject-matter of the independent claims. Some embodiments are defined in the dependent claims. The scope of protection sought for various embodiments of the invention is set out by the independent claims. The embodiments, examples and features, if any, described in this specification that do not fall under the scope of the independent claims are to be interpreted as examples useful for understanding various embodiments of the present disclosure.

[0010] According to a first aspect of the present disclosure, there is provided a federated learning client device comprising at least one processor and at least one memory storing instructions that, when executed by the atleast one processor, cause the federated learning client device to perform operations, the operations comprising at least transmitting, to an access node over a communication channel, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning client device, decrypting, using a first decryption key, second parameter data received by the federated learning client device from the access node, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, and updating, as part of federated learning, the local model parameters of the local model of the federated learning client device with the global model parameters.

[0011] According to a second aspect of the present disclosure, there is provided an apparatus for a network node comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to perform operations, wherein the operations comprise at least obtaining, from an access node, a sum of plural parameter data originating in plural federated learning client devices, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific one of the plural federated learning client devices, obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and providing the second parameter data to the plural federated learning client devices via the access node.

[0012] According to a third aspect of the present disclosure, there is provided an access node comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the access node to perform operations, wherein the operations comprise at least receiving, over one or more communication channel, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device, summing the received plural parameter data together to obtain a sum of the plural parameter data, and providing the sum of the plural parameter data to a network node, receiving second parameter data from the network node and delivering the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.

[0013] According to a fourth aspect of the present disclosure, there is provided a method in a federated learning client device, comprising transmitting, to an access node over a communication channel, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning client device, decrypting, using a first decryption key, second parameter data received by the federated learning client device from the access node, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, and updating, as part of federated learning, the local model parameters of the local model of the federated learning client device with the global model parameters.

[0014] According to a fifth aspect of the present disclosure, there is provided a method, comprising obtaining, from an access node, a sum of plural parameter data originating in plural federated learning client devices, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific one of the plural federated learning client devices, obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and providing the second parameter data to the plural federated learning client devices via the access node.

[0015] According to a sixth aspect of the present disclosure, there is provided a method of an access node, comprising receiving, over one or more communication channel, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device, summing the received plural parameter data together to obtain a sum of the plural parameter data, and providing the sum of the plural parameter data to a network node, receiving second parameter data from the network node and delivering the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.

[0016] According to a seventh aspect of the present disclosure, there is provided a non-transitory computer readable medium having stored thereon a set of computer readable instructions that, when executed by at least one processor, cause a federated learning client device to at least transmit, to an access node over a communication channel, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning client device, decrypt, using a first decryption key, second parameter data received by the federated learning client device from the access node, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, and updating, as part of federated learning, the local model parameters of the local model of the federated learning client device with the global model parameters.

[0017] According to an eighth aspect of the present disclosure, there is provided a non-transitory computer readable medium having stored thereon a set of computer readable instructions that, when executed by at least one processor, cause an apparatus to at least obtain, from an access node, a sum of plural parameter data originating in plural federated learning client devices, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific one of the plural federated learning client devices, obtain second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and provide the second parameter data to the plural federated learning client devices via the access node.

[0018] According to a ninth aspect of the present disclosure, there is provided a non-transitory computer readable medium having stored thereon a set of computer readable instructions that, when executed by at least one processor, cause an access node to at least receive, over one or more communication channel, from pluralfederated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device, sum the received plural parameter data together to obtain a sum of the plural parameter data, and provide the sum of the plural parameter data to a network node, receive second parameter data from the network node and deliver the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.

[0019] According to a tenth aspect of the present disclosure, there is provided a user equipment comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the user equipment to perform operations, the operations comprising at least transmitting, to a cellular base station over an air interface, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the user equipment, decrypting, using a first decryption key, second parameter data received by the federated learning user equipment from the cellular base station, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, and updating, as part of federated learning, the local model parameters of the local model of the user equipment with the global model parameters.

[0020] According to an eleventh aspect of the present disclosure, there is provided a cellular network node comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the cellular network node to perform operations, wherein the operations comprise at least obtaining, from a cellular base station, a sum of plurality parameter data originating in a plurality of user equipment, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific user equipment of the plurality of user equipment, obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm and channel coded, to the sum of the plurality of parameter data, and providing the second parameter data to the plurality of user equipment via the cellular base station.

[0021] According to a twelfth aspect of the present disclosure, there is provided a cellular base station comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the cellular base station to perform operations, wherein the operations comprise at least receiving, over one or more air interface, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning user equipment, summing the received plural parameter data together to obtain a sum of the plural parameter data, and providing the sum of the plural parameter data to a network node, receiving second parameter data from the network node and delivering the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.

[0022] According to a thirteenth aspect of the present disclosure, there is provided a method of a user equipment, comprising transmitting, to a cellular base station over an air interface, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the user equipment, decrypting, using a first decryption key, second parameter data received by the user equipment from the cellular base station, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, and updating, as part of federated learning, the local model parameters of the local model of the user equipment with the global model parameters.

[0023] According to a fourteenth aspect of the present disclosure, there is provided a method, comprising obtaining, from a cellular base station, a sum of plural parameter data originating in a plurality of user equipment, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific user equipment of the plurality of user equipment, obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and providing the second parameter data to the plural federated learning client devices via the cellular base station.

[0024] According to a fifteenth aspect of the present disclosure, there is provided a method of a cellular base station, comprising receiving, over one or more air interface, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective user equipment of the plurality of user equipment, summing the received plural parameter data together to obtain a sum of the plural parameter data, and providing the sum of the plural parameter data to a network node, receiving second parameter data from the network node and delivering the second parameter data to each of the plurality of user equipment as an update to the local models of the plurality of user equipment.

[0025] According to a sixteenth aspect of the present disclosure, there is provided a non-transitory computer readable medium having stored thereon a set of computer readable instructions that, when executed by at least one processor, cause a federated learning user equipment to at least transmit, to a cellular base station over a air interface, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning user equipment, decrypt, using a first decryption key, second parameter data received by the federated learning user equipment from the cellular base station, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, and updating, as part of federated learning, the local model parameters of the local model of the federated learning user equipment with the global model parameters.

[0026] According to a seventeenth aspect of the present disclosure, there is provided a non-transitory computer readable medium having stored thereon a set of computer readable instructions that, when executed by at least one processor, cause an cellular network node to at least obtain, from a cellular base station, a sum of plural parameter data originating in plural federated learning client devices, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local modelparameters of a local model of a specific one of the plural federated learning client devices, obtain second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and provide the second parameter data to the plural federated learning client devices via the cellular base station.

[0027] According to an eighteenth aspect of the present disclosure, there is provided a non-transitory computer readable medium having stored thereon a set of computer readable instructions that, when executed by at least one processor, cause a cellular base station to at least receive, over one or more air interface, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning user equipment, sum the received plural parameter data together to obtain a sum of the plural parameter data, and provide the sum of the plural parameter data to a network node, receive second parameter data from the network node and deliver the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.

[0028] According to a nineteenth aspect of the present disclosure, there is provided a federated learning client device, comprising means for transmitting, to an access node over a communication channel, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning client device, decrypting, using a first decryption key, second parameter data received by the federated learning client device from the access node, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, and updating, as part of federated learning, the local model parameters of the local model of the federated learning client device with the global model parameters.

[0029] According to a twentieth aspect of the present disclosure, there is provided an apparatus, comprising means for obtaining, from an access node, a sum of plural parameter data originating in plural federated learning client devices, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific one of the plural federated learning client devices, obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and providing the second parameter data to the plural federated learning client devices via the access node.

[0030] According to a twenty-first aspect of the present disclosure, there is provided an access node comprising means for receiving, over one or more communication channel, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device, summing the received plural parameter data together to obtain a sum of the plural parameter data, and providing the sum of the plural parameter data to a network node, receiving second parameter data from the network node and delivering the second parameter data to each ofthe plural federated learning client devices as an update to the local models of the plural federated learning client devices.

[0031] According to a twenty-second aspect of the present disclosure, there is provided a federated learning user equipment comprising means for transmitting, to a cellular base station over an air interface, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning user equipment, decrypting, using a first decryption key, second parameter data received by the federated learning user equipment from the cellular base station, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, and updating, as part of federated learning, the local model parameters of the local model of the federated learning user equipment with the global model parameters.

[0032] According to a twenty-third aspect of the present disclosure, there is provided an apparatus, comprising means for obtaining, from a cellular base station, a sum of plural parameter data originating in multiple user equipment, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific user equipment of the multiple user equipment, obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and providing the second parameter data to the multiple user equipment via the cellular base station.

[0033] According to a twenty-fourth aspect of the present disclosure, there is provided a cellular base station comprising means for receiving, over one or more air interface, from a plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective user equipment, summing the received plural parameter data together to obtain a sum of the plural parameter data, and providing the sum of the plural parameter data to a network node, receiving second parameter data from the network node and delivering the second parameter data to each of the plural federated learning client devices as an update to the local models of the plurality of user equipment.

[0034] According to a twenty-fifth aspect of the present disclosure, there is provided a network node comprising at least one processor, a trusted execution environment and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to perform operations, wherein the operations comprise at least obtaining, from a federated learning server, homomorphically encrypted second parameter data comprising a sum of randomly generated noise encrypted using the homomorphic encryption algorithm and a sum of plural parameter data originating in a plurality of federated learning client devices, decrypting, in the trusted execution environment of the network node, the second parameter data by reversing the homomorphic encryption and encrypting, in the trusted execution environment, the decrypted second parameter data with a second encryption algorithm, and providing the second parameter data as encrypted using the second encryption algorithm to the plurality of federated learning client devices via an access node.

[0035] According to a twenty-sixth aspect of the present disclosure, there is provided a network node comprising at least one processor, a trusted execution environment and at least one memory storing instructions that, whenexecuted by the at least one processor, cause the apparatus to perform operations, wherein the operations comprise at least obtaining, from a federated learning server, homomorphically encrypted second parameter data comprising a sum of randomly generated noise encrypted using the homomorphic encryption algorithm and a sum of plural parameter data originating in multiple user equipment, decrypting, in the trusted execution environment, the second parameter data by reversing the homomorphic encryption and encrypting, in the trusted execution environment, the decrypted second parameter data with a second encryption algorithm, and providing the second parameter data as encrypted using the second encryption algorithm to the multiple federated learning client devices via a cellular base station.

[0036] According to a twenty-seventh aspect of the present disclosure, there is provided a system comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the system to perform operations, wherein the operations comprise at least receiving, over one or more communication channel, from a plurality of federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device of the plurality of federated learning client devices, summing the received plural parameter data together to obtain a sum of the plural parameter data, obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and delivering the second parameter data to each of the plurality federated learning client devices as an update to the local models of the plurality federated learning client devices.

[0037] According to a twenty-eighth aspect of the present disclosure, there is provided a system comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the system to perform operations, wherein the operations comprise at least receiving, over one or more air interface, from multiple user equipment, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective user equipment of the multiple user equipment, summing the received plural parameter data together to obtain a sum of the plural parameter data, obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and delivering the second parameter data to each of the multiple user equipment as an update to the local models of the multiple user equipment.BRIEF DESCRIPTION OF THE DRAWINGS

[0038] FIGURE 1 is a block diagram of an example federated learning system in accordance with at least some embodiments of the present disclosure;

[0039] FIGURE 2A is a diagram showing signalling sent by and operations performed by wireless devices, a base station, and a server of a federated learning system in accordance with a first federated learning procedure;

[0040] FIGURE 2B is a diagram showing signalling sent by and operations performed by wireless devices, a base station, and a server of a federated learning system in accordance with a second federated learning procedure;

[0041] FIGURE 2C illustrates signalling sent by and operations performed by wireless devices, a base station, and a server of a federated learning system in accordance with at least some embodiments of the present disclosure;

[0042] FIGURE 2D illustrates signalling sent by and operations performed by wireless devices, a base station, and a server of a federated learning system in accordance with at least some embodiments of the present disclosure;

[0043] FIGURE 3A is a block diagram of an example wireless device capable of participating in federated learning in accordance at least some embodiments of the present disclosure;

[0044] FIGURE 3B is a block diagram of an example server capable of participating in federated learning in accordance at least some embodiments of the present disclosure;

[0045] FIGURE 3C is a block diagram of an example device capable of participating in federated learning in accordance at least some embodiments of the present disclosure;

[0046] FIGURE 4 is a first flowchart showing a method of a wireless device of a federated learning system in accordance with at least some embodiments of the present disclosure;

[0047] FIGURE 5 is a second flowchart showing a method of an apparatus of a federated learning system in accordance with at least some embodiments of the present disclosure

[0048] FIGURE 6 is a third flowchart showing a method of another apparatus of a federated learning system in accordance with at least some embodiments of the present disclosure, and

[0049] FIGURE 7 is a fourth flow graph of a method in accordance with at least some embodiments of the present disclosure.

[0050] DETAILED DESCRIPTION

[0051] Herein are disclosed methods, apparatuses and system for computationally efficient privacy-preserving federated learning. In detail, during federated learning, a client (e.g., a wireless device) encrypts the local model parameters of a local model of the client device using homomorphic encryption and adds noise to prevent both a network-side server and peer nodes from discovering private information of client devices. The federated learning client device (e.g., wireless device) may also encode the local model parameters using a linear channel coding algorithm and then transmit the encrypted and encoded local model parameters to an access node of a communication network. Linearity of both a channel coding in use, and of summation in homomorphic encryption, enable a computationally light process in a server, conserving processor cycles, time and energy. Thus, technical effects enabled by various embodiments herein disclosed include ensuring privacy of data of a client privacy when the client is participating in federated learning, and conservation of processor cycles, time and energy used in computation of the update to the parameter models at a central server controlling the federated learning. Local models of the client devices are updated using global model parameters, the globalmodel parameters being aggregated from local model update parameter data of multiple federated learning client devices. During each iteration of federated learning global model parameters are generated by the central server from the local model parameters received from the multiple client devices and the model parameters of the local model at each respective client device is updated with the global model parameters.

[0052] FIGURE 1 illustrates an example federated learning system in accordance with at least some embodiments of the present disclosure. FIGURE 1 illustrates an example federating learning system which includes a cellular communication network that connects clients (e.g., wireless devices which are otherwise referred to herein as federated learning client devices) that include local models that participate in federated learning with a server that includes a central model. However, the present disclosure is not limited to federated learning systems that use cellular technology, rather, it extends also to non-cellular and non-wireless embodiments. The federated learning system shown in FIGURE 1 includes base stations 130, 135 in communication with wireless devices, such as wireless device 110, which are configured to act as client devices when participating in federated learning. Wireless devices 110 are configured to communicate with the base stations 130, 135 of a communication network using a wireless communication interface. A radio link connects base station 130 with wireless device 110 via the wireless communication interface. The radio link may be bidirectional, comprising an uplink, UL, to convey information from wireless device 110 toward base station 130, and a downlink, DL, to convey information from the base station 130 toward wireless device 110. Although the cellular communication network shown in FIGURE 1 only includes two base stations for ease of illustration, a cellular communication network may comprise hundreds or thousands of base stations. The base stations may be distributed in that a particular base station may comprise a centralized unit, CU, and one or more distributed unit, DU. A base station may operate in accordance with a radio access technology, such as the radio access technology for NR and in such cases, the base station is an example of an access node (e.g., a radio access network node).

[0053] Base station 130 is further coupled communicatively with a core network 140, which may comprise, for example, an evolved packet core, EPC, comprising a mobility management entity, MME, a 5G core network comprising virtualized network functions, such as an access and mobility management function, AMF, a 5G unified data repository, UDR, a session management function, SMF, etc, or a 6G core network. The core network 140 may be coupled with a data network 150, which may comprise the Internet or a corporate network, for example. The communication network may communicate with further networks via data network 150. The cellular communication network may be configured to operate in accordance with any suitable radio access technology such as long term evolution, LTE, fifth generation, 5G, which is also known as New Radio, NR, or a sixth generation, 6G radio access technology as defined by the 3rdgeneration partnership project, 3GPP, standards. To obtain interoperation, wireless devices attaching to a communication network (e.g, a radio access network and / or a core network of a cellular communication network) are configured to support (e.g., operate in accordance with) a same standard as the cellular communication network.

[0054] Base station 130 provides and controls, in the example of FIGURE 1 , cells 130A and 130B, of which wireless devices 110 illustrated in FIGURE 1 attached with cell 130A, and base station 135 controls, in theexample of FIGURE 1 , cells 135A and 135B. The number of cells may be in excess of what is illustrated in FIGURE 1 . It is also possible that a base station has a single cell. While illustrated as a sector-shape, cells of a same base station may be omnidirectional and operate on different frequencies, for example.

[0055] Wireless devices configured to perform operations according to embodiments herein described may comprise user equipment which communicate with base stations (e.g., radio access network nodes) of a cellular communication network, such as the cellular communication network as shown in FIGURE 1 , or the wireless devices may comprise, for example, wireless local area network, WLAN, wireless devices which communicate with access points of a WLAN, worldwide interoperability for microwave access, WiMAX, device which communicate with access points of a WiMAX network or Bluetooth™ wireless devices which communicate with access points of Bluetooth™ network or with other Bluetooth™ devices. User equipment configured to communicate with base stations (e.g., radio access network nodes) of a cellular network may comprise smartphones, tablet computers, laptop computers, items of smart jewellery or, internet of things (loT) devices, for example, machine-type user equipment such as wireless utility meters or connected car connectivity devices. By wireless device it is herein not referred to devices which are comprised in a radio access network, such as base stations or wireless access points. Whereas the present disclosure is expressed primarily in terms of wireless devices, wireless devices are an example of federated learning client devices which are usable with the teachings of the present disclosure. In some embodiments the federated learning client devices are not wireless devices but may communicate with an access point of a communication network using a wire-line connection.

[0056] Federated learning is a useful in wireless communication networks to train a machine learning model, such as collision protection model, or an MDT model. An MDT model may be used to maintain a coverage map of a cellular communication network (e.g., a radio access network of a cellular communication network), which may be used, for example, to predict when a wireless device will request handover from one base station to another base station (or one cell to another cell). Predicting a handover may trigger reducing a requested data rate at the wireless device, and directing, by the network, a wireless device to measure a signal strength of signals transmitted in neighbouring cells of a cell that is serving the wireless device. A mobility event, such as a handover, may comprise a switch from one beam to another beam of the same cell, or a switch from one cell to another cell. To support mobility procedures, UEs, including UE 110, are configured to measure signal strengths of adjacent beams and / or cells to generate measurement results, and report these measurement results to the network, which may then take a decision concerning a mobility event, such as a beam change or a cell switch. A machine learning model (e.g., an artificial neural network), trained using federated learning may be used by a wireless device for scheduling radio transmissions, for selecting a cell for connecting to, for identifying pedestrians in image data and for blocking spam messages and spam calls. For example, where the wireless device is a connected car connectivity device (e.g., a wireless device of a connected car), a trained local model (e.g., an artificial neural network) may be trained using FL to detect pedestrians in image data generated by an onboard camera, to assist driving the car safely.

[0057] Federated learning server 13S is co-located with base station 130 in the example of FIGURE 1. Alternatively to being co-located with base station 130, federated learning server 13S may be located in another location, for example it may be in the core network, or in an other part of the radio-access network wherein base stations 130 and 135 are comprised.

[0058] Implementing a simple and usable federated learning system is an active research topic. As described herein above, some federated learning solutions present risk of leakage of private information at, for example, a wireless channel or nodes infected with malware. Further, other federated learning clients may infer some private information from the global model parameter updates. In herein described federated learning solutions, the local model updating, such as an artificial intelligence, Al, local model updating, is facilitated by a server located, for example, at a same site with an access node of a communication network. Each wireless device sends their local model parameter updates to the server to be aggregated and provided back to each participating wireless device as global model parameters. The local model parameters, or their updates, may be gradients or values of the local parameters of the local models. Computation power in the server is limited. The client devices prefer for their gradients to be protected, at least to some extent, against discovery by other federated learning wireless devices and the server may also be considered to be curious concerning the local model parameter updates. The local model parameter updates are homomorphically encrypted and channel coded using a linear channel code, such that the network side, such as a server in the network, can add the local model parameters of plural wireless devices together without reversing the homomorphic encryption, and in some embodiments also without channel decoding. Likewise, noise may be added without opening the homomorphic encryption, such that the noise protects against leakage of the private information to other client devices and the homomorphic encryption protects against leakage of the private information from the clients to the network or the server.

[0059] In a federated learning system as herein described, linear properties of channel coding schemes used to encode the local parameter updates and of summation in homomorphic encryption schemes are relied on to reduce a number of decoding and decryption operations performed by a federated learning server and / or at an access node. The access node may be a cellular base station, a WLAN access point, a WiMAX access point or a Bluetooth access node, for example. Further, noise is added to the summed local model parameters (e.g., the summed gradients or values) at the server before providing the summed local model parameters to the local models for updating the local model parameters to enhance privacy of the client devices. The noise may be added by randomly selecting a set of already encoded, encrypted messages (from e.g. a Gaussian quantized noise distribution) and adding them to the encoded, encrypted summed local model parameters. This allows for gradient perturbation with noise without the need for decryption and decoding of each of the gradients. Alternatively, noise may be generated, channel encoded and encrypted as needed, for example if the encryption key for homomorphic encryption is dynamically shared and not pre-distributed.

[0060] Advantages obtained by the federated learning system described herein include a reduction in computation that needs to be performed at a federated learning server , namely:0061] Here the base-line column corresponds to federated learning with differential privacy, Case 1 corresponds to the first federated learning procedure described herein below, and Case 2 corresponds to the second federated learning procedure described herein below. N is the number or clients (e.g., wireless devices), and DeCO and DeCR are the computational complexity of channel decoding and decryption algorithms, respectively. Furthermore, CO and CR are the computational complexity of encoding and encryption, respectively. Typically CO « DeCO. These substantial computation savings lead to more efficient use of energy at the server of the federated learning system shown in FIGURE 1. In Case 1 , the server performs one encoding and one encryption operation and in Case 2, the server performs one decoding, one encoding and one encryption operation per (e.g., for each) wireless device. Homomorphic encryption, HE, used is lightweight since only linear homomorphic summation is needed and no other operations are required to be performed on data encrypted using this HE. The privacy and security of this federated learning solution is automatically postquantum safe as differential privacy provides an information-theoretic secrecy level via one time pad key, and HE, depending on the algorithm used, may be post-quantum safe. Examples of suitable HE algorithms include the Paillier cryptosystem, the Brakerski-Gentry-Vaikuntanathan, BGV, scheme, and the Gentry-Sahai-Waters, GSW, scheme. The BGV and GSW schemes are post-quantum safe HE algorithms.

[0062] Concerning mathematical notation used herein, HE(-) is a homomorphic encryption operation, which allows for summation and a key for it is shared among all or selected federated learning client devices, such as the wireless devices shown in FIGURE 1. The key sharing process for the federated learning procedure described herein may be any known to a person skilled in the art and hence are not described in detail. The corresponding decryption is denoted by

[0063] A linear channel coding operation, such as low-density parity-check code, LDPC, is denoted with C(-) and the corresponding channel decoding operation is shown by C-1(-). ndpdenotes noise added in the herein disclosed federated learning solution, which may be sampled from an appropriate distribution, such as a Gaussian distribution or a coloured noise distribution.

[0064] A first federated learning procedure will now be described, wherein the i_th federated learning client device sends its local model parameters, referred to herein as gradients gtfirst encrypted and then encoded as Xi'.

[0065] Xi= C(HE(gi)).

[0066] An access node, such as, for example, a radio access network node such as a base station of FIGURE 1, receives, over an e.g. wireless interface and on a communication channel, noisy versions ofas:

[0067] rt= C(HE(g ) + n

[0068] where, n, is noise incurred over the communication channel and in the receiver of the access node. Then the access node, without channel decoding or decrypting rbsums the received from each federated learning client device as follows:

[0071] Here the last two phases rely on the linearity of channel coding and summation in the used homomorphic encryption. Then the access node, or a server to which rgis sent from the access node, adds random noise ndp, which may be uniformly Gaussian noise with given variance, to rgto increase the privacy of the federated learning solution:

[0074] Then the access node, or the server to which rgis sent from the access node, sends pflback to the federated learning client devices. At the i-th federated learning client device, decoding and decrypting the pgmay proceed as:

[0077] dg— Xi i + ndp

[0078] Here nuiis the noise introduced by the communication channel, such as a wireless communication channel, and the receiver of the i-th federated learning client device. In the decoding process, the two noises are removed. In case channel decoding fails, then a negative acknowledgement, NACK, will be sent to the access node or server which rgis sent from the access node. A potential disadvantage of this first federated learning procedure is that if the channel decoding fails at the federated learning client device, it may not be clear if the problem is the n, or nui. This means that the NACK which is sent by the federated learning client device to the access node, or the server which rgis sent from the access node, and the access node, or the server which rgis sent from the access node, can retransmit dg. However, if the problem was too much n, then the access node, or the server which rgis sent from the access node has to send a request to all the federated learning client device's to retransmit their local model parameters (e.g., gradients or values of the local modelparameters). Alternatively, the access node may determine, which ones of the local model parameters (e.g., gradients or values of the local model parameters of the local models of the federated client devices) the access node can channel decode and request retransmission from only the federated learning client device and / or federated learning client devices the gradients of which the access node cannot channel decode.

[0079] As a variant of the first federated learning procedure described above, the HE may be reversed by at least one further node of the communication network, or a device comprising a trusted execution environment, and the decrypted gtmay be encrypted with another encryption scheme prior to sending it to the federated learning client devices. A further node could be, for example, a network node or a network function (NF) of a core network of a cellular communication network, such as a 5G core network (5GC), or an access node, such as a 5G next generation base station (gNB). The decryption of the HE and the encryption with the other encryption scheme may be performed by a single further node, or the decryption may be performed by a first further node and the encryption by a second further node. A first further node could be, for example, a network analytics entity, such as a network data analytics function (NWDAF) of a 5GC that includes a model training logical function (MTLF), for example. A second further node could be, for example, an access management entity, such as an access and mobility management function (AMF) of a 5GC, or a base station such as a gNB of a 5G RAN. An example of the other encryption scheme used to encrypt the decrypted gtis a non-access stratum (NAS) encryption scheme performed by an AMF, or an access stratum (AS) encryption scheme performed by a base station. Thus the decrypting of the global model parameters in the federated learning client device may comprise homomorphic decryption of the homomorphic encryption used to encrypt the local model parameters in the federated client devices, or the decrypting may comprise decrypting of the other encryption scheme.

[0080] FIGURE 2A is a diagram showing signal sent by and operations performed by wireless devices, a base station, and a server of a federated learning system, for example the federated learning system of FIGURE 1, in accordance with a first federated learning procedure. In FIGURE 2A, the wireless devices comprise federated learning client devices and the federated learning server, denoted SRV and corresponding to federated learning server 13S of FIGURE 1, may be co-located with the BS. The BS is a radio access network node. One of federated learning client devices WD1 and WD2 may correspond to wireless device 110 of FIGURE 1.

[0081] At 210 and 220, WD1 and WD2 provide (e.g., send or signal) their homorphically encrypted local model parameters xl =to the BS. xl and %2 are received by the BS as= C(HE gi)) + Tii as described herein above. At 230, the BS sums Trover / to obtain rg=which is provided from the BS to the server (denoted as SRV) at 240. At 250 the server (SRV) adds channel encoded and homomorphically encoded noise ndpto the received rgto obtain pg= a C(HE g ) +nt + as discussed herein above. pgis transmitted to the BS at 270, and the BS sends pgonwardto WD1 and WD2 at 285 and 280, respectively. At 285 and 280 the BS sends pgonward to WD1 and WD2 by broadcasting a system information block comprising pg, or transmitting messages pgaddressed to each of WD1 and WD2 which comprise pg, for example. In case the number of wireless devices is very high,broadcasting is more efficient than transmitting RRC messages to each respective wireless device. In general, and not only relating to FIGURE 2A, broadcasting may comprise transmission of a system information block, of broadcasting via Bluetooth, WLAN or another wireless link. The messages addressed to individual client devices, that is, unicast messages, may be RRC messages, or Bluetooth or WLAN unicast messages, for example.

[0082] The wireless devices channel decode pgand reverse its homomorphic encryption at 292 and 294, and at 296 and 298 the thus obtained global model parametersgt+ ndpare used to update local model parameters of the respective local models of the wireless devices. In some embodiments, the homomorphic encryption has been reversed in the network and the £ ' i gi + ndphave been encrypted using an other encryption scheme, as described herein above.

[0083] Alternatively to the first federated learning procedure, a second federated learning procedure may be performed by a federated learning system, such as the federated learning system of FIGURE 1. The second federated learning procedure differs from the first federated learning procedure in that in the second federated learning procedure the access node reverses the channel coding of the received local model parameters (e.g., gradients or values of the local model parameters) after receiving them from the federated learning client devices. An advantage of second federated learning procedure is that noise n, is removed by the access node using the channel coding. This increases the likelihood that the federated learning client devices will be able to decode the global model parameters sent by the access node, and thus reduce the number of needed retransmissions of pg. However, a limitation of the second federated learning procedure is that the computational load at the access node is increased. The second federated learning procedure is more advantageous e.g. if the air interface (e.g., the radio interface or radio link between a federated learning client device and the access node) is very noisy, or of computational capacity in the access node is not particularly restricted. In some embodiments, the access node is configured to select to use the first federated learning procedure when its computational load status is in excess of a computational load threshold, and to select to use the second federated learning procedure when computational load status is less than the computational load threshold.

[0084] FIGURE 2B illustrates signalling sent by and operations performed by wireless devices, a base station, and a server of a federated learning system in accordance with the second federated learning procedure

[0085] At 210, 220 and 230 correspond to those in FIGURE 2A. At 23A the BS will attempt to channel decode rgto obtain dg= 'liHE gi). If the channel decoding of rgdoes not succeed (e.g., is unsuccessful), the BS will decode the individually and request retransmissions from wireless devices which sent which the BS fail to decode at 23B. Once the decoding succeeds (i.e., is successful), with or without re-transmissions, dgis sent to the server (SRV) at 240, which adds homomorphically encrypted noise, for example gaussian noise, to dgas HE(ndp) at 25A. The homomorphically encrypted noise is not channel encoded, since the parameters (e.g., the gradients or the values of the parameter) of the local model are not either. The homomorphicallyencrypted noise is added to dgat 26A to obtain p'g= dg+ HE(ndp) = £ ' i HE(gi) + HE(ndp). p'gis sent to the BS at 27A, the BS channel encodes p'gto obtain pg= C(p'g).

[0086] As in FIGURE 2A, pgis transmitted to the wireless devices at 280 and 285. Again as in FIGURE 2A, the wireless devices decode and decrypt pgto obtain the global model parameters (e.g., the gradients or values of the parameters of the global model) which are used to update each wireless device's local model parameters. In case decoding fails at 292 or 294, a NACK is returned (e.g., sent or transmitted) to the BS, and the BS responds to the NACK by re-transmitting pgto the wireless device. There is no need for re-transmission of the local model parameters (e.g., gradients or values of the parameters of the local model), unlike in FIGURE 2A, as noise n, was removed already at 23A.

[0087] The wireless client devices may be configured to send their local model parameters (e.g., gradients or values) xi = c(HE gi)) to an access node of the cellular communication network such that gtis not encrypted using packet data convergence protocol, PDCP, encryption. In normal use, one of the services of the PDCP layer is encryption of data to be transported over an air interface of the cellular system. The use of homomorphic encryption renders PDCP encryption superfluous as PCDP encryption would not increase the encryption security level. Thus both in the first federated learning procedure and in the second federated learning procedure, xi are transmitted without PDCP encryption. To inform the radio access network node of the identity of the transmitting wireless device (e.g., federated learning client device) in the absence of PDCP encryption, the wireless client device may be configured to transmit, along with the local model parameters (e.g., the gradients or values of the local model parameters), an encrypted version of the identity of wireless device (generally referred to as wireless device identity), which may be a radio network temporary identity, RNTI, for example. More generally, the local model parameters may be transmitted without encrypting the parameter data using encryption on any protocol stack layer used for transferring the parameter data to the access node. Examples of protocol stack layers include the PDCP layer, the RRC layer, a medium access control, MAC, layer, 802.11 frame layer or logical link control and adaptation protocol layer. The wireless device identity may be encrypted using a separate form of encryption other than the homomorphic encryption, which enables the access node to decrypt the wireless device identity. In some embodiments, the access node in the federated learning system is a cellular base station, the client devices are user equipments and the server is a cellular network node, such as a radio access network node or a core network node.

[0088] The wireless devices (e.g., the federated learning client devices) may be provided, by the access node or the server, a first encryption key usable in performing the homomorphic encryption of the local model parameters (e.g., the gradient or values of the local model parameters). Likewise the wireless devices (e.g., the federated learning client devices) may be provided, by the access node or the server, a second encryption key the wireless devices (e.g., federated learning client device) use to encrypt the identity of the wireless device (e.g., the federated learning client device), such as a UE identity in the case the federated learning system comprises a cellular network, for example. An example of UE identity is an RNTI.

[0089] FIGURE 20 illustrates signalling in accordance with at least some embodiments of the present disclosure. The vertical axes correspond to those in FIGURES 2A and 2B. FIGURE 20 relates to wireless device identification in a cellular embodiment, where the client devices are cellular UEs.

[0090] At C10 and C20, WD1 and WD2 authenticate and register themselves with the cellular network. The WDs also indicate to the network, that they support HE and that they support federated learning using HE. The SRV may be involved, as illustrated, or alternatively an authentication server may take the place of the SRV as it relates to C10 and C20. At C30 and 040, client devices WD1 and WD2 are provided a key for use in homomorphic encryption as a first encryption key and a public key of the BS as a second encryption key. The client devices WD1 and WD2 may also be indicated, which HE algorithm is to be used in encryption of the local model parameters. The wireless devices may be provided the HE algorithm at 030 and 040.

[0091] At 040 and 045, the wireless devices provide to the BS their local modelencrypted with the first encryption key and their identifiers encrypted using the second encryption key. PDCP encryption is not used for the local model parameters. In phase 050 the BS retrieves and decrypts the WD identifiers, and discards local model parameters provided by client devices not in scope of the federated learning process. Phase 060 comprises phases 230 - 260 of FIGURE 2A. 280 - 298 correspond to similarly numbered phases of FIGURE 2A.

[0092] 280 and 285 may take place either using broadcasting, meaning a radio transmission not addressed to a specific node, or via messages, such as RRC messages, addressed to the client devices WD1 , WD2 individually.

[0093] FIGURE 2D illustrates signalling in accordance with at least some embodiments of the present disclosure. FIGURE 2D relates to an loT embodiment where the client devices are wireless loT UEs.

[0094] At D10, the UEs (UE1 and UE2) store their respective UE identifiers. At D15, the server (denoted SRV), or another network node, determines that a public BS key is to be provided to all UEs in a system information block that is broadcast by the the BS or multiple BSs. At D20 and D30, the BS delivers (e.g., sends) the BS public key to the UEs, for example, by broadcasting the BS public keys. Subsequently, the UEs encrypt their UE ID and their local model parameters, using for the local model parameter encrypting a default HE algorithm or an HE algorithm indicated or provided by the BS. At D40 and D45, the encrypted UE ID and the HE-encrypted local model parameters are provided to the BS, wherein the UE ID is encrypted using the key disseminated at D20 and D30. In phase D50, the BS retrieves the UE IDs using its private key, and discards possible local model parameters or other messages received from UEs not in scope of the federated learning solution. D60 corresponds to phases 230 - 260 of FIGURE 2A, and in D70 and D85 the pgis provided to the UEs. In case broadcasting is used, D70 and D85 may form a single operation.

[0095] At D92 and D94 the UEs decode and decrypt the received pg, and at D96 and D98 the decrypted global parameters are used to update the respective local parameters of the respective local models in the UEs.

[0096] FIGURE 3A illustrates an example federated learning client device, such as the wireless device 110 shown in FIGURE 1 , according to some embodiments of the present disclosure. Comprised in device 300 is processor 310, which may comprise, for example, a single- or multi-core processor wherein a single-coreprocessor comprises one processing core and a multi-core processor comprises more than one processing core. Processor 310 may comprise, in general, a control device. Processor 310 may comprise more than one processor. When processor 310 comprises more than one processor, device 300 may be a distributed device comprising multiple physical units wherein processing takes place in more than one physical unit. Processor 310 may be a control device. A processing core may comprise, for example, a Cortex-A8 processing core manufactured by ARM Holdings or a Zen processing core designed by Advanced Micro Devices Corporation. A processing core or processor may be, or may comprise, at least one qubit. Processor 310 may comprise at least one Qualcomm Snapdragon and / or Intel Atom processor. Processor 310 may comprise a hardware accelerator, a tensor processing unit, a graphics processing unit, and / or a central processing unit. Processor 31 may comprise at least one application-specific integrated circuit, ASIC. Processor 310 may comprise at least one field-programmable gate array, FPGA. Processor 310, optionally together with memory and computer instructions, may be means for performing operations of a method, such as transmitting, decrypting, encrypting, receiving, updating, obtaining and providing. Processor 310 may be configured, at least in part by computer instructions, to perform operations.

[0097] A processor may comprise circuitry, or be constituted as circuitry or circuitries, the circuitry or circuitries being configured to perform phases of methods in accordance with embodiments described herein. As used in this application, the term "circuitry” may refer to one or more or all of the following: (a) hardware-only circuit implementations, such as implementations in only analogue and / or digital circuitry, and (b) combinations of hardware circuits and software, such as, as applicable: (I) a combination of analogue and / or digital hardware circuit(s) with software / firmware and (ii) any portions of hardware processor(s) with software (including digital signal processor(s)), software, and memory(ies) that work together to cause an apparatus, such as a client device or server, to perform various functions) and (c) hardware circuit(s) and or processor(s), such as a microprocessor(s) or a portion of a microprocessor(s), that requires software (e.g., firmware) for operation, but the software may not be present when it is not needed for operation.

[0098] This definition of circuitry applies to all uses of this term in this application, including in any claims. As a further example, as used in this application, the term circuitry also covers an implementation of merely a hardware circuit or processor (or multiple processors) or portion of a hardware circuit or processor and its (or their) accompanying software and / or firmware. The term circuitry also covers, for example and if applicable to the particular claim element, a baseband integrated circuit or processor integrated circuit for a mobile device or a similar integrated circuit in server, a cellular network device, or other computing or network device.

[0099] Device 300 may comprise memory 320. Memory 320 may comprise random-access memory and / or permanent memory. Memory 320 may comprise at least one RAM chip. Memory 320 may be a computer readable medium. Memory 320 may comprise solid-state, magnetic, optical and / or holographic memory, for example. Memory 320 may be at least in part accessible to processor 310. Memory 320 may be at least in part comprised in processor 310. Memory 320 may be means for storing information. Memory 320 may comprise computer executable instructions, which when executed by processor 310, causes device 300 to perform the methods described herein. When computer instructions configured to cause processor 310 to perform certainactions or operations are stored in memory 320, and device 300 overall is configured to run under the direction of processor 310 using computer instructions from memory 320, processor 310 and / or its at least one processing core may be considered to be configured to perform said certain actions or operations. Memory 320 may be at least in part external to device 300 but accessible to device 300. Memory 320 may be transitory or non-transitory. The term "non-transitory”, as used herein, is a limitation of the medium itself (that is, tangible, not a signal) as opposed to a limitation on data storage persistency (for example, RAM vs. ROM).

[0100] Device 300 may comprise a transmitter 330. Device 300 may comprise a receiver 340. Transmitter 330 and receiver 340 may be configured to transmit and receive, respectively, information in accordance with at least one cellular or non-cellular standard. Transmitter 330 may comprise more than one transmitter. Receiver 340 may comprise more than one receiver. Transmitter 330 and / or receiver 340 may be configured to send and receive wireless signals (e.g., radio signals) and device 300 may in accordance with global system for mobile communication, GSM, wideband code division multiple access, WCDMA, 5G, long term evolution, LTE, IS-95, wireless local area network, WLAN, Ethernet and / or worldwide interoperability for microwave access, WiMAX, standards, for example.

[0101] Device 300 may comprise a near-field communication, NFC, transceiver 350. NFC transceiver 350 may support at least one NFC technology, such as NFC, Bluetooth™, Bluetooth™ Low Energy, BLE, or similar technologies.

[0102] Device 300 may comprise user interface, Ul, 360. Ul 360 may comprise at least one of a display, a keyboard, a touchscreen, a vibrator arranged to signal to a user by causing device 300 to vibrate, a speaker or a microphone. A user may be able to operate device 300 via Ul 360, for example to accept incoming telephone calls, to originate telephone calls or video calls, to browse the Internet, to manage digital files stored in memory 320 or on a cloud accessible via transmitter 330 and receiver 340, or via NFC transceiver 350, and / or to play games.

[0103] Device 300 may comprise or be arranged to accept a user identity module 370. User identity module 370 may comprise, for example, a subscriber identity module, SIM, card installable in device 300. A user identity module 370 may comprise information identifying a subscription of a user of device 300. A user identity module 370 may comprise cryptographic information usable to verify the identity of a user of device 300 and / or to facilitate encryption of communicated information and billing of the user of device 300 for communication effected via device 300.

[0104] Processor 310 may comprise a transmitter arranged to output information from processor 310, via electrical leads internal to device 300, to other devices comprised in device 300. The transmitter may comprise a serial bus transmitter arranged to, for example, output information via at least one electrical lead to memory 320 for storage therein. Alternatively to a serial bus, the transmitter may comprise a parallel bus transmitter. Likewise processor 310 may comprise a receiver arranged to receive information in processor 310, via electrical leads internal to device 300, from other devices comprised in device 300. Such a receiver may comprise a serial bus receiver arranged to, for example, receive information via at least one electrical lead from receiver 340 for processing in processor 310. Alternatively to a serial bus, the receiver may comprise a parallel bus receiver.

[0105] Device 300 may comprise further components not illustrated in FIGURE 3A. For example, where device 300 comprises a smartphone, a table, or laptop, it may comprise at least one digital camera. A device 300 that comprise a smartphone, tablet, or laptop may comprise a back-facing camera and a front-facing camera, wherein the back-facing camera may be intended for digital photography and the front-facing camera for video telephony. Device 300 may comprise a fingerprint sensor arranged to authenticate, at least in part, a user of device 300. In some embodiments, device 300 lacks at least one of the components described above. For example, devices 300 may not include a NFC transceiver 350 and / or user identity module 370.

[0106] Processor 310, memory 320, transmitter 330, receiver 340, NFC transceiver 350, Ul 360 and / or user identity module 370 may be interconnected by electrical leads internal to device 300 in a multitude of different ways. For example, each of the aforementioned devices may be separately connected to a master bus internal to device 300, to allow for the devices to exchange information. However, as the skilled person will appreciate, this is only one example and depending on the embodiment various ways of interconnecting at least two of the aforementioned devices may be selected without departing from the scope of the present invention.

[0107] FIGURE 3B is a block diagram of an example server capable of participating in federated learning in accordance at least some embodiments of the present disclosure. Like numbering denotes like structure as in FIGURE 3A. Device 30B may comprise a server, such as a federated learning server as described herein. Device 30B comprises transceiver 35B, such as a wire-line transceiver, configured to communicate, for example with an access node. In some embodiments, the server may be a single physical computing device. In some embodiments, the server may be a distributed computing system ("system”) comprising multiple computing devices that operate together. In some embodiments, the server may be a virtual machine running on a cloud computing system or cloud computing platform. In some embodiments, the server may be a cloud computing system or cloud computing platform.

[0108] FIGURE 3C is a block diagram of an example device capable of participating in federated learning in accordance at least some embodiments of the present disclosure. Like numbering denotes like structure as in FIGURE 3A and FIGURE 3B. Device 30C is a node with a trusted execution environment 32C. Trusted execution environment 32C may comprise, for example, TrustZone™ or Open Mobile Terminal Platform trusted execution environment.

[0109] FIGURE 4 is a first flowchart showing a method in accordance with at least some embodiments of the present disclosure. The operations of the method illustrated in FIGURE 4 may be performed by a federated learning client device, for example, or by a control device configured to control the federated learning client device, when installed therein

[0110] At 410, a federated learning client device, or a control device to control the federated learning client device, transmits to an access node over a communication channel, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning client device. At 420, the federated learning client device, or the control device to control the federated learning client device decrypts, using a first decryption key, second parameter data received by the federated learning client device from the access node, the second parameterdata comprising a sum of randomly generated noise and global model parameters of a global model. At 430 the federated learning client device, or the control device to control the federated learning client device updates, as part of federated learning, the local model parameters of the local model of the federated learning client device with the global model parameters.

[0111] FIGURE 5 is a second flowchart showing a method in accordance with at least some embodiments of the present disclosure. The operations of the method illustrated in FIGURE 5 may be performed by a federated learning server, for example, or by a control device configured to control the federated learning server, when installed therein.

[0112] At 510, a federated learning server, or a control device configured to control the federated learning server, obtains, from an access node, a sum of plural parameter data originating in plural federated learning client devices, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific one of the plural federated learning client devices. At 520, the federated learning server, or the control device configured to control the federated learning server, obtains second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data. At 530 the federated learning server, or the control device configured to control the federated learning server, provides the second parameter data to the plural federated learning client devices via the access node.

[0113] FIGURE 6 is a third flowchart showing a method in accordance with at least some embodiments of the present disclosure. The operations of the method illustrated in FIGURE 6 may be performed by an access node, for example, or by a control device configured to control an access node, when installed therein.

[0114] At operation 610, an access node, or a control device configured to control an access node, receives, communication channels from a plurality if federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device. At 620, the access node, or the control device configured to control the access node, sums the received plural parameter data together to obtain a sum of the plural parameter data, and providing the sum of the plural parameter data to a network node, and receives second parameter data from the network node. At 630 the access node, or a control device configured to control the access node delivers the second parameter data to each of the plurality federated learning client devices as an update to the local models of the plurality of federated learning client devices.

[0115] FIGURE 7 is a fourth flowchart of a method in accordance with at least some embodiments of the present disclosure The operations of the method illustrated in FIGURE 7 may be performed in a federated learning client device, for example, or in a control device configured to control a federated learning client device, when installed therein.

[0116] At operation 710, a federated learning client device, or a control device configured to control a federated learning client device, transmits to a cellular base station over an air interface, parameter data encrypted usinga homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning user equipment. At operation 720, the federated learning client device, or the control device configured to control a federated learning client device, decrypts, using a first decryption key, second parameter data received by the federated learning user equipment from the cellular base station, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model. At operation 730, the federated learning client device, or the control device configured to control the federated learning client device, updates, as part of federated learning, the local model parameters of the local model of the federated learning client device with the global model parameters.

[0117] FIGURE 8 is a fourth flowchart showing a method in accordance with at least some embodiments of the present disclosure. The operations of the method illustrated in FIGURE 8 may be performed in a cellular network node, for example, or in a control device configured to control a cellular network node, when installed therein.

[0118] At operation 810, the cellular network, or the control device configured to control a cellular network node, obtains, from a cellular base station, a sum of plural parameter data originating in a plurality of federated learning client devices, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific federated learning client device of the plurality of federated learning client devices. At operation 820 the cellular network, or control device configured to control a cellular network node obtains second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data. At operation 830, the cellular network or the control device configured to control a cellular network node, provides the second parameter data to the plurality of federated learning client device via the cellular base station.

[0119] FIGURE 9 is a fourth chart showing a method in accordance with at least some embodiments of the present disclosure. The operations of the method illustrated in FIGURE 9 may be performed in a cellular base station, for example, or in a control device configured to control a cellular base station, when installed therein.

[0120] At operation 910, a cellular base station, or a control device configure to control a cellular base station, receives, over an air interface from a plurality of federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device of the plurality of federated learning client devices. At operation 920, the cellular base station, or the control device configure to control the cellular base station, sums the received plural parameter data together to obtain a sum of the plural parameter data. Finally, at operation 930, the cellular base station, or the control device configure to control the cellular base station, provides the sum of the plural parameter data to a network node, receives second parameter data from the network node and delivers the second parameter data to each of the plural federated learning client devices as an update to the local models of the plurality of federated learning client devices.

[0121] It is to be understood that the embodiments of the invention disclosed are not limited to the particularstructures, process steps, or materials disclosed herein, but are extended to equivalents thereof as would be recognized by those ordinarily skilled in the relevant arts. It should also be understood that terminology employed herein is used for the purpose of describing particular embodiments only and is not intended to be limiting.

[0122] Reference throughout this specification to one embodiment or an embodiment means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. Thus, appearances of the phrases "in one embodiment” or "in an embodiment” in various places throughout this specification are not necessarily all referring to the same embodiment. Where reference is made to a numerical value using a term such as, for example, about or substantially, the exact numerical value is also disclosed.

[0123] As used herein, a plurality of items, structural elements, compositional elements, and / or materials may be presented in a common list for convenience. However, these lists should be construed as though each member of the list is individually identified as a separate and unique member. Thus, no individual member of such list should be construed as a de facto equivalent of any other member of the same list solely based on their presentation in a common group without indications to the contrary. In addition, various embodiments and example of the present invention may be referred to herein along with alternatives for the various components thereof. It is understood that such embodiments, examples, and alternatives are not to be construed as de facto equivalents of one another, but are to be considered as separate and autonomous representations of the present invention.

[0124] Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the preceding description, numerous specific details are provided, such as examples of lengths, widths, shapes, etc., to provide a thorough understanding of embodiments of the invention. One skilled in the relevant art will recognize, however, that the invention can be practiced without one or more of the specific details, or with other methods, components, materials, etc. In other instances, well- known structures, materials, or operations are not shown or described in detail to avoid obscuring aspects of the invention.

[0125] While the forgoing examples are illustrative of the principles of the present invention in one or more particular applications, it will be apparent to those of ordinary skill in the art that numerous modifications in form, usage and details of implementation can be made without the exercise of inventive faculty, and without departing from the principles and concepts of the invention. Accordingly, it is not intended that the invention be limited, except as by the claims set forth below.

[0126] The verbs "to comprise” and "to include” are used in this document as open limitations that neither exclude nor require the existence of also un-recited features. The features recited in depending claims are mutually freely combinable unless otherwise explicitly stated. Furthermore, it is to be understood that the use of "a" or "an", that is, a singular form, throughout this document does not exclude a plurality.

[0127] As used herein, "at least one of the following: ” and "at least one of ” and similar wording, where the list of two or more elements are joined by "and”or "or”, mean at least any one of the elements, or at least any two or more of the elements, or at least all the elements.INDUSTRIAL APPLICABILITY

[0128] At least some embodiments of the present disclosure find industrial application in model updating in wireless settings.ACRONYMS LISTHE homomorphic encryptionMDT minimization of drive testsRNTI radio network temporary identityUE user equipmentTECHNICAL CLAUSES:Clause 1 . A federated learning client device comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the federated learning client device to perform operations, the operations comprising at least:- transmitting, to an access node over a communication channel, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning client device;- decrypting, using a first decryption key, second parameter data received by the federated learning client device from the access node, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, and- updating, as part of federated learning, the local model parameters of the local model of the federated learning client device with the global model parameters.Clause 2. The federated learning client device according to Clause 1 , wherein the operations further comprise encrypting, using a second encryption key, an identifier of the federated learning client device, and providing the encrypted identifier of the federated learning client device to the access node.Clause 3. The federated learning client device according to Clause 2, wherein the transmitting comprises transmitting the parameter data to the access node without encrypting the parameter data encryption on any protocol stack layer used for transferring the parameter data.Clause 4. The federated learning client device according to any of Clauses 2 - 3, wherein the operations further comprise receiving one or both of the first encryption key and the second encryption key from the access node.Clause 5. The federated learning client device according to any of Clauses 1 - 4, wherein the access node is a base station of a radio access network.Clause 6. The federated learning client device according to any of Clauses 1 - 5, wherein the operations comprise receiving the second parameter data in a broadcasted message, or in a message addressed to the federated learning client device.Clause 7. The federated learning client device according to Clause 7, wherein the message addressed to the federated learning client device is a radio resource control message and the broadcasted message is a system information block.Clause 8. The federated learning client device according to any of Clauses 1 - 6, wherein the operations further comprise, responsive to failure to decode the second parameter data, transmitting a negative acknowledgement to the access node, and responsive to an invitation from the access node, re-transmitting the first parameter data to the access node.Clause 9. An apparatus for a network node comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to perform operations, wherein the operations comprise at least:- obtaining, from an access node, a sum of plural parameter data originating in plural federated learning client devices, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific one of the plural federated learning client devices;- obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and- providing the second parameter data to the plural federated learning client devices via the access node.Clause 10. The apparatus according to Clause 9, wherein the sum of plural parameter data is channel coded with a linear channel coding code, and wherein the randomly generated noise is channel coded with the linear channel coding code.Clause 11 . The apparatus according to Clause 9, wherein the sum of plural parameter data is not channel coded, and wherein the randomly generated noise is not channel coded.Clause 12. The apparatus according to any of Clauses 9 - 11 , wherein the providing of the second parameter data comprises directing the access node to broadcast a message comprising the second parameter data ordirecting the access node to transmit, addressed to each of the plural federated learning client devices, a message comprising the second parameter data.Clause 13. An access node comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the access node to perform operations, wherein the operations comprise at least:- receiving, over one or more communication channel, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device;- summing the received plural parameter data together to obtain a sum of the plural parameter data, and- providing the sum of the plural parameter data to a network node, receiving second parameter data from the network node and delivering the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.Clause 14. The access node according to Clause 13, wherein the providing comprises providing the sum of the plural parameter data to the network node without channel decoding the plural parameter data or the sum of the plural parameter data.Clause 15. The access node according to Clause 13 or 14, wherein the operations further comprise channel decoding, responsive to receiving a negative acknowledgement concerning the second parameter data from one of the federated learning client devices, the sum of the plural parameter data, and providing the channel decoded sum of the plural parameter data to the network node.Clause 16. The access node according to Clause 13, wherein the operations further comprise channel decoding each of the received plural parameter data before the summing.Clause 17. The access node according to any of Clauses 13 - 16, wherein the each of the received parameter data is not encrypted on any protocol stack layer used for transferring the parameter data.Clause 18. A system comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the system to perform operations, wherein the operations comprise at least:- receiving, over one or more communication channel, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm witha first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device;- summing the received plural parameter data together to obtain a sum of the plural parameter data;- obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and- delivering the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.Clause 19. A method in a federated learning client device, comprising:- transmitting, to an access node over a communication channel, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning client device;- decrypting, using a first decryption key, second parameter data received by the federated learning client device from the access node, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, and- updating, as part of federated learning, the local model parameters of the local model of the federated learning client device with the global model parameters.Clause 20. The method according to Clause 19, wherein the method further comprises encrypting, using a second encryption key, an identifier of the federated learning client device, and providing the encrypted identifier of the federated learning client device to the access node.Clause 21. The method according to Clause 20, wherein the transmitting comprises transmitting the parameter data to the access node without encrypting the parameter data encryption on any protocol stack layer used for transferring the parameter data.Clause 22. The method according to any of Clauses 20 - 21 , wherein the method further comprises receiving one or both of the first encryption key and the second encryption key from the access node.Clause 23. The method according to any of Clauses 19 - 22, wherein the access node is a base station of a radio access network.Clause 24. The method according to any of Clauses 19 - 23, wherein the method comprises receiving the second parameter data in a broadcasted message, or in a message addressed to the federated learning client device.Clause 25. The method according to Clause 24, wherein the message addressed to the federated learning client device is a radio resource control message and the broadcasted message is a system information block.Clause 26. The method according to any of Clauses 19 - 25, wherein the method further comprises, responsive to failure to decode the second parameter data, transmitting a negative acknowledgement to the access node, and responsive to an invitation from the access node, re-transmitting the first parameter data to the access node.Clause 27. A method, comprising:- obtaining, from an access node, a sum of plural parameter data originating in plural federated learning client devices, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific one of the plural federated learning client devices;- obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and- providing the second parameter data to the plural federated learning client devices via the access node.Clause 28. The method according to Clause 27, wherein the sum of plural parameter data is channel coded with a linear channel coding code, and wherein the randomly generated noise is channel coded with the linear channel coding code.Clause 29. The method according to Clause 28, wherein the sum of plural parameter data is not channel coded, and wherein the randomly generated noise is not channel coded.Clause 30. The method according to any of Clauses 27 - 29, wherein the providing of the second parameter data comprises directing the access node to broadcast a message comprising the second parameter data or directing the access node to transmit, addressed to each of the plural federated learning client devices, a message comprising the second parameter data.Clause 31. A method in an access node, comprising:- receiving, over one or more communication channel, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device;- summing the received plural parameter data together to obtain a sum of the plural parameter data, andproviding the sum of the plural parameter data to a network node, receiving second parameter data from the network node and delivering the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.Clause 32. The method according to Clause 31, wherein the providing comprises providing the sum of the plural parameter data to the network node without channel decoding the plural parameter data or the sum of the plural parameter data.Clause 32. The method according to Clause 31 or 32, wherein the method further comprises channel decoding, responsive to receiving a negative acknowledgement concerning the second parameter data from one of the federated learning client devices, the sum of the plural parameter data, and providing the channel decoded sum of the plural parameter data to the network node.Clause 33. The method according to Clause 31 , wherein the method further comprises channel decoding each of the received plural parameter data before the summing.Clause 34. The method according to any of Clauses 31 - 33, wherein the each of the received parameter data is not encrypted on any protocol stack layer used for transferring the parameter data.Clause 35. A non-transitory computer readable medium having stored thereon a set of computer readable instructions that, when executed by at least one processor, cause a federated learning client device to at least:- transmit, to an access node over a communication channel, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning client device;- decrypt, using a first decryption key, second parameter data received by the federated learning client device from the access node, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, and- updating, as part of federated learning, the local model parameters of the local model of the federated learning client device with the global model parameters.Clause 36. A non-transitory computer readable medium having stored thereon a set of computer readable instructions that, when executed by at least one processor, cause an apparatus to at least:- obtain, from an access node, a sum of plural parameter data originating in plural federated learning client devices, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific one of the plural federated learning client devices;- obtain second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and- provide the second parameter data to the plural federated learning client devices via the access node.Clause 37. A non-transitory computer readable medium having stored thereon a set of computer readable instructions that, when executed by at least one processor, cause an access node to at least:- receive, over one or more communication channel, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device;- sum the received plural parameter data together to obtain a sum of the plural parameter data, and- provide the sum of the plural parameter data to a network node, receive second parameter data from the network node and deliver the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.Clause 38. A federated learning user equipment comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the federated learning user equipment to perform operations, the operations comprising at least:- transmitting, to a cellular base station over a air interface, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning user equipment;- decrypting, using a first decryption key, second parameter data received by the federated learning user equipment from the cellular base station, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, and- updating, as part of federated learning, the local model parameters of the local model of the federated learning user equipment with the global model parameters.Clause 39. The federated learning user equipment according to Clause 28, wherein the operations further comprise encrypting, using a second encryption key, an identifier of the federated learning user equipment, and providing the encrypted identifier of the federated learning user equipment to the cellular base station.Clause 40. The federated learning user equipment according to Clause 39, wherein the transmitting comprises transmitting the parameter data to the cellular base station without encrypting the parameter data encryption on any protocol stack layer used for transferring the parameter data.Clause 41 . The federated learning user equipment according to any of Clauses 39 - 40, wherein the operations further comprise receiving one or both of the first encryption key and the second encryption key from the cellular base station.Clause 42. The federated learning user equipment according to any of Clauses 39 - 41 , wherein the cellular base station is a base station of a radio access network.Clause 43. The federated learning user equipment according to any of Clauses 39 - 42, wherein the operations comprise receiving the second parameter data in a broadcasted message, or in a message addressed to the federated learning user equipment.Clause 44. The federated learning user equipment according to Clause 43, wherein the message addressed to the federated learning user equipment is a radio resource control message and the broadcasted message is a system information block.Clause 45. The federated learning user equipment according to any of Clauses 39 - 44, wherein the operations further comprise, responsive to failure to decode the second parameter data, transmitting a negative acknowledgement to the cellular base station, and responsive to an invitation from the cellular base station, retransmitting the first parameter data to the cellular base station.Clause 46. A cellular network node for a cellular base station comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the cellular network node to perform operations, wherein the operations comprise at least:- obtaining, from a cellular base station, a sum of plural parameter data originating in plural federated learning client devices, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific one of the plural federated learning client devices;- obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and- providing the second parameter data to the plural federated learning client devices via the cellular base station.Clause 47. The cellular network node according to Clause 46, wherein the sum of plural parameter data is channel coded with a linear channel coding code, and wherein the randomly generated noise is channel coded with the linear channel coding code.Clause 48. The cellular network node according to Clause 46, wherein the sum of plural parameter data is not channel coded, and wherein the randomly generated noise is not channel coded.Clause 49. The cellular network node according to any of Clauses 46 - 48, wherein the providing of the second parameter data comprises directing the cellular base station to broadcast a message comprising the second parameter data or directing the cellular base station to transmit, addressed to each of the plural federated learning client devices, a message comprising the second parameter data.Clause 50. A cellular base station comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the cellular base station to perform operations, wherein the operations comprise at least:- receiving, over one or more air interface, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device;- summing the received plural parameter data together to obtain a sum of the plural parameter data, and- providing the sum of the plural parameter data to a network node, receiving second parameter data from the network node and delivering the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.Clause 51. The cellular base station according to Clause 50, wherein the providing comprises providing the sum of the plural parameter data to the network node without channel decoding the plural parameter data or the sum of the plural parameter data.Clause 52. The cellular base station according to Clause 50 or 51 , wherein the operations further comprise channel decoding, responsive to receiving a negative acknowledgement concerning the second parameter data from one of the federated learning client devices, the sum of the plural parameter data, and providing the channel decoded sum of the plural parameter data to the network node.Clause 53. The cellular base station according to Clause 50, wherein the operations further comprise channel decoding each of the received plural parameter data before the summing.Clause 54. The cellular base station according to any of Clauses 50 - 53, wherein each of the received parameter data is not encrypted on any protocol stack layer used for transferring the parameter data.Clause 55. A system comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the system to perform operations, wherein the operations comprise at least:- receiving, over one or more air interface, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device;- summing the received plural parameter data together to obtain a sum of the plural parameter data;- obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and- delivering the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.Clause 56. A method in a federated learning user equipment, comprising:- transmitting, to a cellular base station over a air interface, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning user equipment;- decrypting, using a first decryption key, second parameter data received by the federated learning user equipment from the cellular base station, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, and- updating, as part of federated learning, the local model parameters of the local model of the federated learning user equipment with the global model parameters.Clause 57. The method according to Clause 56, wherein the method further comprises encrypting, using a second encryption key, an identifier of the federated learning user equipment, and providing the encrypted identifier of the federated learning user equipment to the cellular base station.Clause 58. The method according to Clause 57, wherein the transmitting comprises transmitting the parameter data to the cellular base station without encrypting the parameter data encryption on any protocol stack layer used for transferring the parameter data.Clause 59. The method according to any of Clauses 56 - 58, wherein the method further comprises receiving one or both of the first encryption key and the second encryption key from the cellular base station.Clause 60. The method according to any of Clauses 56 - 59, wherein the cellular base station is a base station of a radio access network.Clause 61. The method according to any of Clauses 56 - 60, wherein the method comprises receiving the second parameter data in a broadcasted message, or in a message addressed to the federated learning user equipment.Clause 62. The method according to Clause 61 , wherein the message addressed to the federated learning user equipment is a radio resource control message and the broadcasted message is a system information block.Clause 63. The method according to any of Clauses 56 - 62, wherein the method further comprises, responsive to failure to decode the second parameter data, transmitting a negative acknowledgement to the cellular base station, and responsive to an invitation from the cellular base station, re-transmitting the first parameter data to the cellular base station.Clause 64. A method, comprising:- obtaining, from a cellular base station, a sum of plural parameter data originating in plural federated learning client devices, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific one of the plural federated learning client devices;- obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and- providing the second parameter data to the plural federated learning client devices via the cellular base station.Clause 65. The method according to Clause 64, wherein the sum of plural parameter data is channel coded with a linear channel coding code, and wherein the randomly generated noise is channel coded with the linear channel coding code.Clause 66. The method according to Clause 65, wherein the sum of plural parameter data is not channel coded, and wherein the randomly generated noise is not channel coded.Clause 67. The method according to any of Clauses 64 - 66, wherein the providing of the second parameter data comprises directing the cellular base station to broadcast a message comprising the second parameter data or directing the cellular base station to transmit, addressed to each of the plural federated learning client devices, a message comprising the second parameter data.Clause 68. A method in a cellular base station, comprising:- receiving, over one or more air interface, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryptionkey, each received parameter data comprising local model parameters of a local model of a respective federated learning user equipment;- summing the received plural parameter data together to obtain a sum of the plural parameter data, and- providing the sum of the plural parameter data to a network node, receiving second parameter data from the network node and delivering the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.Clause 69. The method according to Clause 68, wherein the providing comprises providing the sum of the plural parameter data to the network node without channel decoding the plural parameter data or the sum of the plural parameter data.Clause 70. The method according to Clause 68 or 69 wherein the method further comprises channel decoding, responsive to receiving a negative acknowledgement concerning the second parameter data from one of the federated learning client devices, the sum of the plural parameter data, and providing the channel decoded sum of the plural parameter data to the network node.Clause 71. The method according to Clause 68, wherein the method further comprises channel decoding each of the received plural parameter data before the summing.Clause 72. The method according to any of Clauses 68 - 71, wherein the each of the received parameter data is not encrypted on any protocol stack layer used for transferring the parameter data.Clause 73. A non-transitory computer readable medium having stored thereon a set of computer readable instructions that, when executed by at least one processor, cause a federated learning user equipment to at least:- transmit, to a cellular base station over a air interface, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning user equipment;- decrypt, using a first decryption key, second parameter data received by the federated learning user equipment from the cellular base station, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, and- updating, as part of federated learning, the local model parameters of the local model of the federated learning user equipment with the global model parameters.Clause 74. A non-transitory computer readable medium having stored thereon a set of computer readable instructions that, when executed by at least one processor, cause an cellular network node to at least:- obtain, from a cellular base station, a sum of plural parameter data originating in plural federated learning client devices, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific one of the plural federated learning client devices;- obtain second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and- provide the second parameter data to the plural federated learning client devices via the cellular base station.Clause 75. A non-transitory computer readable medium having stored thereon a set of computer readable instructions that, when executed by at least one processor, cause a cellular base station to at least:- receive, over one or more air interface, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning user equipment;- sum the received plural parameter data together to obtain a sum of the plural parameter data, and- provide the sum of the plural parameter data to a network node, receive second parameter data from the network node and deliver the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.Clause 76. A network node comprising at least one processor, a trusted execution environment and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to perform operations, wherein the operations comprise at least:- obtaining, from a federated learning server, homomorphically encrypted second parameter data comprising a sum of randomly generated noise encrypted using the homomorphic encryption algorithm and a sum of plural parameter data originating in plural federated learning client devices;- decrypting, in the trusted execution environment, the second parameter data by reversing the homomorphic encryption and encrypting, in the trusted execution environment, the decrypted second parameter data with a second encryption algorithm, and- providing the second parameter data as encrypted using the second encryption algorithm to the plural federated learning client devices via an access node.Clause 77. An network node comprising at least one processor, a trusted execution environment and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to perform operations, wherein the operations comprise at least:- obtaining, from a federated learning server, homomorphically encrypted second parameter data comprising a sum of randomly generated noise encrypted using the homomorphic encryption algorithm and a sum of plural parameter data originating in plural federated learning client devices;- decrypting, in the trusted execution environment, the second parameter data by reversing the homomorphic encryption and encrypting, in the trusted execution environment, the decrypted second parameter data with a second encryption algorithm, and- providing the second parameter data as encrypted using the second encryption algorithm to the plural federated learning client devices via a cellular base stationClause 78. A system comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the system to perform operations, wherein the operations comprise at least:- receiving, over one or more communication channel, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device;- summing the received plural parameter data together to obtain a sum of the plural parameter data;- obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and- delivering the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.Clause 79. A system comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the system to perform operations, wherein the operations comprise at least:- receiving, over one or more air interface, from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device;- summing the received plural parameter data together to obtain a sum of the plural parameter data;- obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data, and- delivering the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.

Claims

39CLAIMS:

1. A user equipment comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the federated learning client device to perform operations, the operations comprising at least: transmitting, to a cellular base station via a radio link, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the user equipment client device; decrypting, using a first decryption key, second parameter data received by the user equipment from the cellular base station, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model; and updating, as part of federated learning, the local model parameters of the local model of the user equipment with the global model parameters.

2. The user equipment according to claim 1, wherein the operations further comprise: encrypting, using a second encryption key, an identifier of the user equipment, and providing the encrypted identifier of the user equipment to the cellular base station.

3. The user equipment according to claim 2, wherein the transmitting comprises transmitting the parameter data to the cellular base station without encrypting the parameter data encryption on any protocol stack layer used for transferring the parameter data.

4. The user equipment according to claim 2 or 3, wherein the operations further comprise: receiving one or both of the first encryption key and the second encryption key from the cellular base station.

5. The user equipment according to any of claims 1 to 4, wherein the cellular base station is a base station of a radio access network.

6. The user equipment according to any of claims 1 to 5, wherein the operations further comprise: receiving the second parameter data in a broadcasted message, or in a message addressed to the user equipment.

7. The user equipment according to claim 6, wherein the message addressed to the federated learning client device is a radio resource control message and the broadcasted message is a system information block.

408. The user equipment according to any of claims 1 to 6, wherein the operations further comprise: responsive to failure to decode the second parameter data, transmitting a negative acknowledgement to the cellular base station, and responsive to an invitation from the cellular base station, re-transmitting the first parameter data to the cellular base station.

9. A cellular network node comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the cellular network node to perform operations, wherein the operations comprise at least: obtaining, from a cellular base station, a sum of plural parameter data originating in a plurality of user equipment, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific user equipment of the plurality user equipment; obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data; and providing the second parameter data to the plural federated learning client device via the cellular base station.

10. The cellular network node according to claim 9, wherein the sum of plural parameter data is channel coded with a linear channel coding code, and wherein the randomly generated noise is channel coded with the linear channel coding code.

11. The cellular network node according to claim 9, wherein the sum of plural parameter data is not channel coded, and wherein the randomly generated noise is not channel coded.

12. The cellular network node according to any of claims 9 - 11 , wherein the providing of the second parameter data comprises directing the cellular base station to broadcast a message comprising the second parameter data or directing the cellular base station to transmit, addressed to each of the plurality of user equipment, a message comprising the second parameter data.

13. A cellular base station comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the cellular base station to perform operations, wherein the operations comprise at least:- receiving, over air interfaces from a plurality of user equipment, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning user equipment;- summing the received plural parameter data together to obtain a sum of the plural parameter data, and41 providing the sum of the plural parameter data to a network node, receiving second parameter data from the network node and delivering the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning user client devices.

14. The cellular base station according to claim 13, wherein the providing comprises providing the sum of the plural parameter data to the network node without channel decoding the plural parameter data or the sum of the plural parameter data.

15. The cellular base station according to claim 13 or 14, wherein the operations further comprise channel decoding, responsive to receiving a negative acknowledgement concerning the second parameter data from one of the federated learning client devices, the sum of the plural parameter data, and providing the channel decoded sum of the plural parameter data to the network node.

16. The cellular base station according to claim 13, wherein the operations further comprise channel decoding each of the received plural parameter data before the summing.

17. The cellular base station according to any of claims 13 to 16, wherein each of the received parameter data is not encrypted on any protocol stack layer used for transferring the parameter data.

18. An apparatus comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the system to perform operations, wherein the operations comprise at least: receiving, over air interfaces from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning client device; summing the received plural parameter data together to obtain a sum of the plural parameter data; obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm, and channel coded, to the sum of plural parameter data; and delivering the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.

19. A method of a federated learning client device, comprising:- transmitting, to a cellular base station over a air interface, parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, the parameter data comprising local model parameters of a local model of the federated learning user equipment;- decrypting, using a first decryption key, second parameter data received by the federated learning user equipment from the cellular base station, the second parameter data comprising a sum of randomly generated noise and global model parameters of a global model, andupdating, as part of federated learning, the local model parameters of the local model of the federated learning user equipment with the global model parameters.

20. The method according to claim 19, wherein the method further comprises encrypting, using a second encryption key, an identifier of the federated learning user equipment, and providing the encrypted identifier of the federated learning user equipment to the cellular base station.

21. The method according to claim 20, wherein the transmitting comprises transmitting the parameter data to the cellular base station without encrypting the parameter data encryption on any protocol stack layer used for transferring the parameter data.

22. The method according to any of claims 20 - 21 , wherein the method further comprises receiving one or both of the first encryption key and the second encryption key from the cellular base station.

23. The method according to any of claims 19 to 22, wherein the cellular base station is a base station of a radio access network.

24. The method according to any of claims 19 - 23, wherein the method comprises receiving the second parameter data in a broadcasted message, or in a message addressed to the federated learning user equipment.

25. The method according to claim 24, wherein the message addressed to the federated learning user equipment is a radio resource control message and the broadcasted message is a system information block.

26. The method according to any of claims 19 - 25, wherein the method further comprises, responsive to failure to decode the second parameter data, transmitting a negative acknowledgement to the cellular base station, and responsive to an invitation from the cellular base station, re-transmitting the first parameter data to the cellular base station.

27. A method, comprising: obtaining, from a cellular base station, a sum of plural parameter data originating in a plurality of user equipment, each parameter data being encrypted using a homomorphic encryption algorithm with an encryption key, each parameter data comprising local model parameters of a local model of a specific one of the plural federated learning client devices; obtaining second parameter data by adding randomly generated noise encrypted using the homomorphic encryption algorithm and channel coded, to the sum of plural parameter data; and providing the second parameter data to the plural federated learning client devices via the cellular base station.

28. The method according to claim 27, wherein the sum of plural parameter data is channel coded with a linear channel coding code, and wherein the randomly generated noise is channel coded with the linear channel coding code.

29. The method according to claim 28, wherein the sum of plural parameter data is not channel coded, and wherein the randomly generated noise is not channel coded.

30. The method according to any of claims 27 - 29, wherein the providing of the second parameter data comprises directing the cellular base station to broadcast a message comprising the second parameter data or directing the cellular base station to transmit, addressed to each of the plural federated learning client devices, a message comprising the second parameter data.

31. A method of a cellular base station, the method comprising: receiving, over air interfaces from plural federated learning client devices, plural parameter data, each parameter data encrypted using a homomorphic encryption algorithm with a first encryption key, each received parameter data comprising local model parameters of a local model of a respective federated learning user equipment; summing the received plural parameter data together to obtain a sum of the plural parameter data; and providing the sum of the plural parameter data to a network node, receiving second parameter data from the network node and delivering the second parameter data to each of the plural federated learning client devices as an update to the local models of the plural federated learning client devices.

32. The method according to claim 31 , wherein the providing comprises providing the sum of the plural parameter data to the network node without channel decoding the plural parameter data or the sum of the plural parameter data.

33. The method according to claim 31 or 32, wherein the method further comprises channel decoding, responsive to receiving a negative acknowledgement concerning the second parameter data from one of the federated learning client devices, the sum of the plural parameter data, and providing the channel decoded sum of the plural parameter data to the network node.

34. The method according to claim 31 , wherein the method further comprises channel decoding each of the received plural parameter data before the summing.

35. The method according to any of claims 31 to 33, wherein the each of the received parameter data is not encrypted on any protocol stack layer used for transferring the parameter data.

36. A non-transitory computer readable medium having stored thereon computer readable instructions that, when executed by at least one processor, cause a federated learning client device to perform the method of any of claims 19 to 26.4437. A non-transitory computer readable medium having stored thereon computer readable instructions that, when executed by at least one processor, cause a cellular network node to perform the method of any of claims 27 to 30.

38. A non-transitory computer readable medium having stored thereon a set of computer readable instructions that, when executed by at least one processor, cause a cellular base station to perform the method of any of claims 31 to 34.

39. A computer program comprising computer readable instructions, wherein when the computer program is executed by at least one processor of a federated learning client device, the federated learning client device is caused to perform the method of any of claims 19 to 26.

40. A computer program comprising computer readable instructions, wherein when the computer program is executed by at least one processor of a cellular network node, the cellular network node is caused to perform the method of any of claims 27 to 30.41 . A computer comprising computer readable instructions, wherein when the computer program is executed by at least one processor of a cellular base station, the cellular base station is caused to perform the method of any of claims 31 to 34.

42. A network node comprising at least one processor, a trusted execution environment and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to perform operations, wherein the operations comprise at least: obtaining, from a federated learning server, homomorphically encrypted second parameter data comprising a sum of randomly generated noise encrypted using the homomorphic encryption algorithm and a sum of plural parameter data originating in plural federated learning client devices; decrypting, in the trusted execution environment, the second parameter data by reversing the homomorphic encryption and encrypting, in the trusted execution environment, the decrypted second parameter data with a second encryption algorithm; and providing the second parameter data as encrypted using the second encryption algorithm to the plural federated learning client devices via a cellular base station.

Images