System and method for network topology change in active in-network learning heterarchical intelligent collaboration session
Patent Information
- Application Number
- PCT/CN2025/085402
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-03-27
- Publication Date
- 2026-10-01
Smart Images

Figure CN2025085402_01102026_PF_FP_ABST
Abstract
Description
SYSTEM AND METHOD FOR NETWORK TOPOLOGY CHANGE IN ACTIVE IN-NETWORK LEARNING HETERARCHICAL INTELLIGENT COLLABORATION SESSIONTECHNICAL FIELD
[0001] The present disclosure relates to In-Network Learning (INL) and Heterarchical Intelligent Collaboration (HiC) in distributed communication networks. Specifically, the present disclosure relates to a system and a method of signaling for network topology change in an active INL HiC session in a communications network.BACKGROUND
[0002] In the field of telecommunications and artificial intelligence (AI) , distributed learning and collaborative training have emerged as essential methods for improving the efficiency and adaptability of AI models within communications networks. In-Network Learning (INL) involves training and inference processes distributed across different clients, base stations, and core network nodes. This allows for real-time processing and adaptation of AI models based on data collected from multiple spatially distributed devices.
[0003] However, managing the dynamic nature of network topology presents significant challenges. As the network evolves, new devices may join, existing devices may detach, and the availability of data modalities may change. The topology changes can disrupt the learning process, degrade model performance, and create communication bottlenecks. Therefore, a robust signaling mechanism is required to adapt to these changes without interrupting the ongoing training or inference session.
[0004] Current solutions for distributed learning in communication networks primarily focus on static network configurations. Existing approaches lack efficient signaling methods for handling real-time topology changes in active INL sessions. Specifically, existing systems do not support real-time adaptation to the addition or removal of nodes during an active learning session, there are no established mechanisms for handling changes in data modality availability without retraining the entire model. Further, existing approaches fail to account for variations in computational resources and data relevance across different nodes, leading to suboptimal model performance.
[0005] Therefore, in light of the foregoing discussion, there exists a need to overcome the aforementioned drawbacks.SUMMARY
[0006] The present disclosure provides a system, a method, and a computer program of signaling for network topology change in an active in-network learning (INL) heterarchical intelligent collaboration (HiC) session in a communications network. The present disclosure provides a solution to the existing problem of how to maintain continuity in the active INL HiC sessions when network topology changes occur in telecommunication networks. In modern telecommunication infrastructure, data and computational resources are inherently distributed across various nodes, including user equipment, base stations, and core network components. The distribution creates challenges for AI learning algorithms, which must operate efficiently despite spatial separation of resources. An objective of the present disclosure is to provide a solution that overcomes at least partially the problems encountered in the prior art and provides an improved system and an improved method for signaling network topology changes in the active INL HiC session. The disclosure introduces a comprehensive signaling protocol that operates within the established HiC architecture to enable seamless adaptation to topology changes. The signaling protocol works through a structured two-phase approach that efficiently manages communication between HiC agents, an HiC Controller, and Application Function / Network Function (AF / NF) nodes to update neural network patterns and parameters throughout the distributed learning system.
[0007] One or more objectives of the present disclosure are achieved by the solutions provided in the enclosed independent claims. Advantageous implementations of the present disclosure are further defined in the dependent claims.
[0008] In one aspect, the present disclosure provides a method of signaling for network topology change in an active In-Network Learning (INL) Heterarchical Intelligent Collaboration (HiC) session in a communications network. The method includes a first phase of signaling and a second phase of signaling. During the first phase of signaling, the method includes receiving topology change information from any of a plurality of HiC agents, where an HiC agent is located within a respective node of the communications network. During the first phase of the signaling, the method further includes sending an update message related to patterns and parameters related to the received topology change information, wherein the patterns correspond to the active INL HiC session and includes information concerning a neural network model, an optimisation method, related hyperparameters and a communication path. During the second phase of signaling, the method includes sending a required pattern related to the received topology change information to the HiC agent of the node which sent the topology change information. Further, during the second phase of signaling, the method further includes sending an updated pattern related to the received topology change information to a parent node of the node having the HiC agent which sent the topology change information.
[0009] The method provides the ability to dynamically adjust the INL architecture in response to topology changes, ensuring seamless learning and inference operations without requiring a full retraining of the AI model. By implementing a structured two-phase signaling mechanism, the system may quickly detect and respond to the addition or removal of nodes and changes in data modality availability. In the first phase, the method enhances the network’s responsiveness by allowing HiC agents to report topology changes in real-time, enabling the HiC Controller to assess the impact of these changes and generate update messages accordingly. The dynamic changes ensure that modifications in network structure, such as the addition of new devices or detachment of existing ones, do not disrupt the ongoing learning process. The incorporation of patterns and parameters related to the neural network model, optimization methods, hyperparameters, and communication paths allows for an intelligent adaptation strategy that maintains model accuracy and efficiency despite dynamic network conditions. In the second phase, the system ensures minimal disruption by providing updated patterns to both the node initiating the topology change and its parent node. The targeted signaling approach prevents unnecessary network-wide updates, thereby reducing communication overhead and computational burden.
[0010] Additionally, by adapting the learning model’s structure dynamically, the method ensures that distributed AI processing remains optimal even as network conditions evolve. This contributes to improved resource utilization, as computational resources are reallocated based on real-time changes, leading to better efficiency in AI model execution. Furthermore, the proposed signaling approach enhances network reliability by enabling continuous operation even in fluctuating environments. Traditional methods require manual reconfiguration or reinitialization of AI models when network topology changes, leading to downtime and performance degradation. In contrast, the automated signaling framework allows for a seamless transition, maintaining the accuracy and consistency of AI learning processes. By efficiently propagating updates through the network hierarchy, the method ensures that all affected nodes are synchronized with the latest topology changes, thereby maintaining robust and efficient in-network learning.
[0011] In an implementation form, one of the HiC agents is in a user equipment (UE) device of the communications network.
[0012] Placing the HiC agents in the UE device enables localized processing, reducing reliance on centralized servers and minimizing latency. The reduction in reliance enhances real-time adaptability to network changes, optimizes resource utilization at the edge, and improves overall system efficiency by distributing computational load across network nodes.
[0013] In an implementation form, one of the HiC agents is in a base station (BS) device of the communications network.
[0014] Placing the HiC agents in the BS device enables efficient data aggregation and low-latency processing for in-network learning. AI model updates and distributed learning tasks are optimized by leveraging the computational resources and network proximity of the BS. Communication overhead is reduced by handling data fusion and task coordination at an intermediate network level.
[0015] In an implementation form, the topology change information relates to a new node being added to the active INL HiC session, with an existing modality.
[0016] The addition of the new node with an existing modality in an active INL HiC session enhances network scalability and efficiency by allowing seamless integration without retraining the AI model. The structured signaling mechanism ensures minimal latency, optimized resource allocation, and real-time synchronization with parent nodes. The real-time synchronization improves workload distribution, prevents bottlenecks, and enhances fault tolerance by providing redundancy in data sources. Overall, the approach maintains AI model accuracy while adapting dynamically to network topology changes.
[0017] In an implementation form, the topology change information relates to a node detaching from the active INL HiC session.
[0018] When the node detaches from the active INL HiC session, the proposed signaling mechanism ensures seamless adaptation by updating network parameters in real-time without disrupting ongoing AI learning. The signaling mechanism enhances resource allocation by redistributing computational tasks among remaining nodes, preventing performance degradation. Additionally, the system maintains model accuracy by dynamically adjusting communication paths and fusion processes, ensuring efficient learning despite node departures. The enhances network resilience, minimizing downtime and maintaining continuous AI operations.
[0019] In an implementation form, the topology change information relates to a new data modality becoming available for the active INL HiC session.
[0020] When the new data modality becomes available in an active INL HiC session, the proposed signaling mechanism enables the seamless integration of the new modality without disrupting ongoing AI learning. The seamless integration enhances the system’s adaptability by allowing the AI model to incorporate additional data types, leading to improved learning accuracy and robustness. The structured signaling process ensures that the HiC Controller efficiently updates communication paths, fusion parameters, and optimization settings, enabling smooth coordination with existing nodes. By dynamically adjusting the network topology, the system optimizes resource allocation, prevents bottlenecks, and enhances multi-modal data fusion, ultimately improving model performance and decision-making capabilities.
[0021] In another implementation form, the topology change information relates to an existing data modality becoming unavailable for the active INL HiC session.
[0022] When the existing data modality becomes unavailable in an active INL HiC session, the proposed signaling mechanism ensures seamless adaptation by dynamically reconfiguring the network and redistributing learning tasks. The dynamic reconfiguration of the network prevents AI model degradation by adjusting fusion processes and optimizing available data sources to maintain accuracy. The maintenance of accuracy enhances system resilience, minimizing disruptions and ensuring continuous AI operations despite changes in data availability.
[0023] In another implementation form, the topology change information includes computational resource capabilities for the new node.
[0024] Including computational resource capabilities for the new node enables efficient workload distribution and optimal utilization of network resources in the INL HiC session. The workload distribution allows the system to assign tasks based on the node’s processing power, preventing bottlenecks and improving overall performance. Additionally, it enhances scalability by dynamically balancing computational loads, ensuring seamless integration of new nodes without disrupting ongoing AI learning.
[0025] In an implementation form, the topology change information includes a local Relevance Indicator (RI) for the new node.
[0026] Including the local RIs for the new node allows the INL HiC session to assess the quality and importance of the node’s data and computational contribution. The assessment of the node’s data enables intelligent task allocation, ensuring that higher-quality data sources are prioritized for learning, improving model accuracy. Additionally, it optimizes resource utilization by dynamically adjusting communication and fusion processes based on the node’s relevance, enhancing overall network efficiency.
[0027] In an implementation form, the topology change information includes a detach notification.
[0028] Including the detach notification in the topology change information ensures real-time network updates, allowing the active INL HiC session to quickly reconfigure and redistribute tasks. The reconfiguration and redistribution of tasks prevents disruptions in AI learning by dynamically adjusting fusion processes and communication paths. Additionally, it enhances system resilience by maintaining optimal performance despite node departures, ensuring continuous and efficient network operations.
[0029] In an implementation form, the topology change information relates to a new node joining the network which has the new data modality.
[0030] When the new node joins the network with the new data modality, the proposed signaling mechanism enables seamless integration, enhancing the diversity and robustness of the INL HiC session. The diversity and robustness of the INL HiC session allow the AI models to leverage additional data types, improving learning accuracy and adaptability. Efficient network updates ensure optimized resource allocation and fusion processes, maintaining high performance while expanding the system’s capabilities.
[0031] In an implementation form, the topology change information relates to an existing node detaching from the active INL HiC session.
[0032] When the existing node detaches from the active INL HiC session, the proposed signaling mechanism ensures seamless adaptation by dynamically updating communication paths and redistributing learning tasks. The dynamically updating communication paths and redistributing learning tasks prevent disruptions in AI model training and inference, maintaining overall system performance. Additionally, it enhances network resilience by optimizing resource allocation and ensuring continuous learning despite node departures.
[0033] In an implementation form, in the first phase of signaling, the receiving step is performed by an HiC Controller (HicC) node.
[0034] Performing the receiving step at the HiC Controller (HicC) node ensures centralized coordination and efficient management of topology changes in the INL HiC session. The efficient management of topology changes enables real-time processing of network updates, optimizing communication and resource allocation. Additionally, it enhances system stability by ensuring all topology changes are systematically handled, minimizing disruptions to AI learning and inference processes.
[0035] In an implementation form, the first phase of signaling, the HiC controller node forwards the received information to an Application Function / Network Function (AF / NF) node.
[0036] Forwarding the received information from the HiC Controller to the AF / NF node ensures efficient decision-making and policy enforcement for topology changes. Efficient decision-making enhances network adaptability by enabling optimized resource management and seamless integration of new nodes or modalities. Additionally, it improves system reliability by ensuring structured processing and validation of topology updates before implementation.
[0037] In an implementation form, the AF / NF node sends the update message to the HiC controller node.
[0038] When the AF / NF node sends the update message to the HiC Controller node, it ensures accurate validation and synchronization of topology changes within the INL HiC session. This facilitates efficient decision-making, optimizing network resource allocation and communication pathways. Additionally, it enhances system reliability by maintaining consistency in AI learning processes while adapting to dynamic network conditions.
[0039] In another aspect, the present disclosure provides a system comprising means adapted for carrying out all the steps of the method.
[0040] The system achieves all the advantages and technical effects of the method of the present disclosure.
[0041] It is to be appreciated that all the aforementioned implementation forms can be combined.
[0042] It has to be noted that all devices, elements, circuitry, units, and means described in the present application could be implemented in the software or hardware elements or any kind of combination thereof. All steps which are performed by the various entities described in the present application as well as the functionalities described to be performed by the various entities are intended to mean that the respective entity is adapted to or configured to perform the respective steps and functionalities. Even if, in the following description of specific embodiments, a specific functionality or step to be performed by external entities is not reflected in the description of a specific detailed element of that entity which performs that specific step or functionality, it should be clear for a skilled person that these methods and functionalities can be implemented in respective software or hardware elements, or any kind of combination thereof. It will be appreciated that features of the present disclosure are susceptible to being combined in various combinations without departing from the scope of the present disclosure as defined by the appended claims.
[0043] Additional aspects, advantages, features, and objects of the present disclosure would be made apparent from the drawings and the detailed description of the illustrative implementations construed in conjunction with the appended claims that follow.BRIEF DESCRIPTION OF THE DRAWINGS
[0044] The summary above, as well as the following detailed description of illustrative embodiments, is better understood when read in conjunction with the appended drawings. For the purpose of illustrating the present disclosure, exemplary constructions of the disclosure are shown in the drawings. However, the present disclosure is not limited to specific methods and instrumentalities disclosed herein. Moreover, those in the art will understand that the drawings are not to scale. Wherever possible, like elements have been indicated by identical numbers.
[0045] Embodiments of the present disclosure will now be described, by way of example only, with reference to the following diagrams wherein:
[0046] FIG. 1 is a block diagram that depicts a system configured for network topology change in an active in-network learning (INL) heterarchical intelligent collaboration (HiC) session in a communication network, in accordance with an embodiment of the present disclosure;
[0047] FIG. 2 is a flowchart depicting a method of signaling for network topology change in the active in-network learning (INL) heterarchical intelligent collaboration (HiC) session in the communications network, in accordance with an embodiment of the present disclosure;
[0048] FIG. 3 is an exemplary diagram depicting signaling flow when a new device with existing data modality joins in the active in-network learning (INL) heterarchical intelligent collaboration (HiC) session in the communications network, in accordance with an embodiment of the present disclosure;
[0049] FIG. 4 is an exemplary diagram depicting signaling flow when an existing node detaches from the active in-network learning (INL) heterarchical intelligent collaboration (HiC) session in the communications network, in accordance with another embodiment of the present disclosure;
[0050] FIG. 5 is an exemplary diagram depicting signaling flow when a new device joins the network or becomes available having a new data modality, in accordance with another embodiment of the present disclosure; and
[0051] FIG. 6 is an exemplary diagram depicting signaling flow when existing data modality becomes unavailable, in accordance with another embodiment of the present disclosure.
[0052] In the accompanying drawings, an underlined number is employed to represent an item over which the underlined number is positioned or an item to which the underlined number is adjacent. A non-underlined number relates to an item identified by a line linking the non-underlined number to the item. When a number is non-underlined and accompanied by an associated arrow, the non-underlined number is used to identify a general item at which the arrow is pointing.DETAILED DESCRIPTION OF EMBODIMENTS
[0053] The following detailed description illustrates embodiments of the present disclosure and ways in which they can be implemented. Although some modes of carrying out the present disclosure have been disclosed, those skilled in the art would recognize that other embodiments for carrying out or practicing the present disclosure are also possible.
[0054] FIG. 1 is a block diagram that depicts a system configured for network topology change in an active in-network learning (INL) heterarchical intelligent collaboration (HiC) session in a communication network, in accordance with an embodiment of the present disclosure. With reference to FIG. 1, there is shown a block diagram that includes a system 100. The system 100 includes a user equipment (UE) device 102 communicably connected to a base station (BS) device 104 via a communication network 106. The system 100 further includes a fusion centre 108 connected to the BS device 104 and a HiC Controller 110. The HiC Controller 110 is coupled to the BS device 104 and the fusion centre (FC) 108. Further, the HiC Controller 110 is connected to core network function nodes 112.
[0055] The In-Network Learning (INL) enables distributed and collaborative processing by leveraging spatially dispersed clients. The INL approach enhances network resource utilization while preserving data privacy and reducing communication overhead. In the context of communication networks, the INL approach is structured into three distinct levels to handle AI learning tasks efficiently. The first level consists of devices that function as Feature Extractors (FE) , responsible for processing raw data and extracting relevant features before transmitting them further in the network. The devices minimize data transmission requirements by sending only essential information rather than full raw datasets. The second level comprises base stations that act as Local Intermediate Fusioners (LIF) , aggregating and refining the extracted features from multiple devices. The intermediate processing reduces the burden on central processing units and improves response times by performing preliminary inference tasks at the edge. Finally, the third level operates at the core network, serving as the FC 108, where the final aggregation, inference, and decision-making occur. The FC 108 ensures a global understanding of the distributed data, consolidates intermediate results and enhances AI model accuracy through comprehensive learning. By structuring INL into the heterarchical levels, communication networks achieve the balance between computational efficiency, reduced latency, and privacy preservation. The multi-tiered approach allows for adaptive AI learning while ensuring that relevant tasks are executed at the most suitable network nodes.
[0056] The system 100 provides a comprehensive signaling mechanism for network topology change framework for distributed intelligent networks, designed to optimize collaboration between network elements using the HiC architecture. The HiC technology (e.g., disclosed in Patent Application Number PCT / CN2022 / 135723; WO 2024 / 113188 A1) includes multiple nodes engaging in real-time data exchange and distributed decision-making, often under varying network conditions. The present disclosure is an improvement of the HiC technology discussed in Patent Application Number PCT / CN2022 / 135723; WO 2024 / 113, 288 A1.
[0057] HiC is a framework that operates within INL, enabling intelligent coordination, early inference, and efficient decision-making among nodes. While HiC utilizes INL’s three-level structure, it focuses on optimizing collaboration, signaling, and dynamic task allocation across these levels . The HiC is a standard architecture for intelligent collaboration between network elements (NEs) at different layers, including collaboration organization management, protocol interfaces, and interaction processes. The HiC efficiently organizes the collaboration between AI tasks on NEs / terminals in terms of space and time, supports the transmission of diversified intelligent representations between the NEs, and unifies the transmission format. In addition, collaboration is controllable, and collaboration patterns are scalable, and various collaborative learning modes are flexibly supported.
[0058] HiC is an advanced distributed intelligence framework designed for enhancing intelligent tasks across multiple nodes within a communication network. HiC enables the collaborative execution of AI-driven tasks by leveraging a network of hierarchical and self-organizing nodes. The HiC framework enhances the efficiency of AI model training and inference by ensuring seamless coordination among terminals, access network nodes, and core network nodes. The HiC framework consists of two primary types of nodes: the HiC Controller node and the HiC agent node. The HiC Controller node functions as the managing entity responsible for configuring, enhancing, and allocating tasks, while the HiC agent node executes the assigned intelligent collaboration tasks. The HiC framework supports different organizational models for intelligent collaboration. The HiC framework can function in a heterarchical mode, where the HiC Controller 110 manages and configures multiple HiC agent nodes, ensuring efficient execution of distributed tasks. In contrast, the HiC framework can also operate in a self-organizing mode, where HiC agent nodes negotiate among themselves to determine task execution without a central controlling entity. The flexibility allows the HiC framework to be scalable and adaptable, making it suitable for small-scale and large-scale distributed learning environments. In scenarios involving a large number of nodes, a combination of hierarchical and self-organizing modes can be used to balance efficiency and complexity.
[0059] One of the key advantages of HiC framework is its ability to dynamically allocate computational resources across multiple nodes. For example, after completing an intelligent collaboration task, the node can initiate a subsequent task based on the results of the previous one. If the first task involves AI model training, the trained model can be used for further inference or adaptive retraining at another node without needing to restart from an initial state. The approach improves node resource utilization and enhances communication efficiency, as each node can progressively refine its AI models without redundant processing.
[0060] The HiC framework also includes mechanisms for task authorization and configuration. Before the HiC agent node executes an intelligent collaborative task, it must first obtain authorization from the HiC-Controller 110. The authorization ensures that tasks are properly assigned and executed in a coordinated manner. The HiC Controller 110 determines which nodes will participate in a given task, assigns necessary computational resources, and ensures that data samples are sufficiently diverse to meet AI model training requirements. Once the HiC agent nodes are configured, they can execute their assigned tasks and report back to the HiC Controller 110 for further coordination. In addition, HiC framework incorporates a robust signaling mechanism for monitoring and optimizing ongoing tasks. The nodes participating in an intelligent collaborative task continuously exchange status updates with the HiC Controller 110 or with other HiC agent nodes, depending on the organizational structure. If the HiC agent node detects inefficiencies, such as insufficient data diversity or network congestion, it can trigger adaptive reconfiguration of the task execution plan. This allows HiC framework to dynamically respond to changes in network conditions, computational capacity, and task complexity, ensuring sustained performance across all participating nodes.
[0061] Overall, HiC represents a highly scalable, adaptable, and intelligent framework for distributed AI processing in communication networks. Its combination of hierarchical control and self-organizing capabilities allows for efficient task execution, reduced processing latency, and optimal resource utilization. By dynamically coordinating AI-driven tasks, HiC significantly improves the efficiency of AI applications in telecommunications, edge computing, and autonomous system.
[0062] The system 100 specifically addresses the challenges of maintaining continuity in active INL sessions when network topology changes occur. The system 100 implements a structured two-phase signaling protocol that enables seamless adaptation. Through the HiC Controller 110, the system 100 orchestrates the exchange of topology change information, computational resource capabilities, and relevance indicators between HiC agents located in network nodes. The information is processed to generate appropriate update messages containing patterns and parameters related to neural network models, optimization methods, hyperparameters, and communication paths. The system 100 then distributes these updates to ensure all affected nodes can adjust their operations accordingly without requiring retraining or reestablishment of the INL HiC session, thereby significantly improving efficiency and resource utilization in telecommunications networks employing distributed artificial intelligence.
[0063] The system 100, utilizing the HiC architecture, introduces an approach to managing network topology changes in distributed INL sessions. The system 100 follows a sophisticated two-phase signaling protocol that enables seamless adaptation to dynamic network environments while maintaining the continuity of collaborative learning processes.
[0064] In the first phase, the system 100 facilitates comprehensive topology change information exchange, where HiC agents within network nodes report relevant details to the HiC Controller 110, including computational resource capabilities, available data modalities, and local Relevance Indicators (RI) . The HiC Controller 110 then coordinates with the AF / NF nodes to validate and process these changes. The approach of the HiC Controller 110 ensures a structured and intelligent response to network dynamics, whether it involves adding new devices, detaching existing nodes, introducing new data modalities, or managing the unavailability of existing modalities. The second phase focuses on precise configuration and parameter updates across the distributed network. The HiC strategically distributes updated patterns to relevant network nodes, including the new or modified device, its parent Local Intermediate Fusioner (LIF) nodes, and the FC. The updates encompass critical information such as localized loss functions, model and optimization parameters, communication patterns, and fusion-related parameters. By dynamically adjusting the parameters, the system 100 maintains the integrity of the neural network model without requiring complete retraining. The present topology changes scenarios while preserving the efficiency of distributed learning. Unlike existing solutions that would typically interrupt or restart learning processes, the system 100 enables continuous adaptation. The system 100 manages the complex interactions between various network nodes-including FE devices, Local Intermediate Fusioner (LIF) base stations, and Fusion Center (FC) nodes-ensuring that each topological change is seamlessly integrated into the ongoing learning session.
[0065] By implementing this adaptive signaling mechanism, the system 100 dramatically improves the robustness of distributed neural network training in telecommunications networks. The system 100 addresses the inherent challenges of spatially distributed computational resources, bandwidth constraints, and privacy considerations, providing a flexible framework for intelligent collaboration across dynamically changing network environments. The approach of the system 100 represents an advancement in distributed machine learning, offering telecommunications networks a sophisticated method to maintain learning continuity, optimize resource utilization, and adapt to real-time network changes with unprecedented efficiency and reliability.
[0066] The UE device 102 refers to a mobile or edge computing device that participates in INL within the HiC framework. In some implementations, the UE device 102 may be a smartphone, IoT sensor, autonomous vehicle, or any computing-enabled terminal capable of collecting local data, executing learning tasks, and sending periodic context reports to the network. The UE device 102 functions as a distributed HiC agent, assisting in dynamic network topology change by providing real-time environmental data, computational resource status, and communication metrics to the HiC Controller 110.
[0067] The BS device 104 refers to a cellular base station that facilitates communication between UE devices 102 and the core network while also serving as a processing and coordination node in the HiC framework. The BS device 104 aggregates topology change information from multiple UEs, performs preliminary processing of learning tasks, and assists in network topology change on network conditions. Additionally, the BS device 104 helps in the management of task distribution, communication path selection, and network resource allocation to enhance the overall efficiency of the system 100.
[0068] The communication network 106 includes a medium (e.g., a communication channel) through which the UE device 102 communicates with the BS device 104. The communication network 106 may be wired or wireless. Examples of the communication network 106 may include, but are not limited to, a Local Area Network (LAN) , a wireless personal area network (WPAN) , a Wireless Local Area Network (WLAN) , a wireless wide area network (WWAN) , a cloud network, a Long-Term Evolution (LTE) network, a plain old telephone service (POTS) , a Metropolitan Area Network (MAN) , and / or Internet.
[0069] The FC 108 refers to a centralized or edge-based processing unit that plays a role in data aggregation, model fusion, and higher-level decision-making in the HiC network. The FC 108 processes task results from multiple UEs and BSs, integrates learning insights and refines the global learning model used across the network. It ensures that data from distributed nodes is synthesized, validated, and used effectively for real-time adaptation and optimization in the INL HiC session.
[0070] The HiC Controller 110 is the central intelligence node responsible for managing and coordinating network topology changes in distributed INL sessions. Functioning as the core orchestration mechanism within the HiC architecture, the HiC Controller 110 receives and processes critical topology change information from various network nodes, including UE device 102, the BS device 104, and Fusion Centers (FC) . When network topology undergoes dynamic changes-such as device addition, detachment, or modifications in data modalities-the HiC Controller 110 initiates a sophisticated two-phase signaling mechanism.
[0071] The core network function nodes 112 refer to Application Function (AF) and Network Function (NF) nodes within the 5G / 6G service-based architecture that manage network intelligence, policy control, and resource orchestration. The core network function nodes 112 provide key network insights such as congestion levels, mobility patterns, security policies, and service prioritization, enabling the HiC framework to adapt learning strategies accordingly. The core network function nodes 112 also facilitate low-latency coordination between distributed HiC agents and core network services, ensuring efficient execution of in-network learning tasks while maintaining optimal network performance.
[0072] There is provided the system 100, during the first phase of signaling, the system 100 is configured to receive topology change information from any of the plurality of HiC agents. The HiC agent is located within a respective node of the communications network. The process begins when an HiC agent detects the topology change within its respective node. In accordance with an embodiment, the topology change information relates to a new node being added to the active INL HiC session, with an existing modality. The modality refers to the type of data a node processes, such as images, audio, sensor readings, or text. When the new node joins with the existing modality, it means that the type of data it provides is already present in the network and used by other nodes for AI learning and inference. The addition of the new node with the existing modality requires seamless integration into the active INL HiC session without disrupting the ongoing AI learning process. The seamless integration is achieved through a structured signaling mechanism, where the HiC Agent within the new node transmits topology change information to the HiC Controller 110. The HiC Controller 110 then processes the information and updates the network to accommodate the new node efficiently. The topology change information refers to the structured data exchanged between network components to notify, process, and implement modifications in the network topology. The modifications include the addition of new nodes, removal of existing nodes, changes in data modalities, or updates in computational resources. The topology change information ensures that the network dynamically adapts to the new node's presence while maintaining optimal AI model performance and resource efficiency. The process of adding a new node with an existing modality follows a structured sequence of events, ensuring that the INL HiC session remains stable and efficient. During topology change the new node becomes available, the node detects its presence in the network and initiates a request to join the active INL HiC session. The HiC Agent within the new node compiles topology change information. The information is transmitted to the HiC Controller 110 as a join request message. The HiC Controller 110 receives the topology change information from the new node’s HiC Agent. The HiC Controller 110 validates the request by checking network policies, security measures, and resource availability. Since the modality is already present in the network, the HiC Controller 110 does not need to create a new AI processing pathway but instead assigns the new node to an appropriate LIF that already handles the same modality. The HiC Controller 110 sends an update message to the AF / NF node, requesting approval for the topology modification. Once approved, the HiC Controller 110 updates the network topology by assigning the new node the parent LIF for data processing, updating fusion processes to include the new node’s contributions and modifying communication links to ensure efficient data flow. The HiC Controller 110 sends final configuration parameters to the new node, including AI model settings, optimization parameters and communication pathways.
[0073] Advantageously, since the node uses the existing modality, the AI model does not need major reconfiguration, ensuring minimal disruption. The HiC Controller 110 dynamically assigns workloads to the new node, preventing overloading of other nodes. The structured signaling mechanism ensures that the new node is detected, validated, and integrated into the network within milliseconds. The additional node provides redundancy, ensuring that AI learning can continue efficiently even if other nodes fail. The updated topology maintains low latency and high efficiency by leveraging existing fusion pathways and AI model structures.
[0074] The topology change information ensures the seamless integration of the new node with the existing modality in an active INL HiC session. The process begins with detection and notification by the HiC Agent, followed by validation and processing by the HiC Controller 110, leading to final integration and synchronization with the network. The structured signaling and real-time updates allow the network to remain scalable, adaptive, and efficient, ensuring uninterrupted AI learning despite dynamic topology changes. The HiC agent continuously monitors node status, data modalities, computational resources, relevance indicators (RI) , and communication links to ensure it can quickly identify any modifications in network structure. Upon detecting the change, the HiC agent generates a topology change notification message, which contains key information such as the node ID, type of topology change, available computational resources, relevance indicator, and updated communication parameters. The message is then transmitted to the system 100 using a structured signaling protocol to ensure low latency and secure communication. Once the system 100 receives the topology change information from one or more HiC agents, it performs several operations to analyze and process the received data. First the authentication and validation is done, where the system 100 ensures that the topology change message is genuine and originates from a legitimate HiC agent. The authentication and validation prevent unauthorized modifications that could compromise the AI learning process. After validation, the system 100 analyzes the impact of the topology change on the ongoing INL session. If the new node is added, the system 100 determines how to integrate it efficiently into the existing learning hierarchy. In accordance with an embodiment, the topology change information further relates to a node detaching from the active INL HiC session. If the node detaches, the system 100 identifies any dependencies and redistributes AI processing tasks among the remaining nodes to ensure uninterrupted learning.
[0075] In accordance with an embodiment, the topology change information further relates to a new data modality becoming available for the active INL HiC session. Similarly, if the new data modality is introduced, the system 100 updates fusion mechanisms to incorporate the new data, while if the data modality is removed, it adjusts the AI model accordingly to compensate for the missing information. If the node reports changes in computational resource capabilities, the system 100 optimizes task allocation across the network to maintain balanced processing efficiency. In accordance with an embodiment, the topology change information relates to an existing data modality becoming unavailable for the active INL HiC session.
[0076] In accordance with an embodiment, the topology change information includes computational resource capabilities for the new node. the topology change information includes computational resource capabilities for the new node to ensure efficient integration into the INL HiC session. When the new node joins the network, it is essential to evaluate its computational power, memory, processing speed, and bandwidth capacity to optimize task allocation and network efficiency. The computational resource capabilities refer to the hardware and software processing capabilities available within the new node, which influence its role in AI training, inference, and data fusion. The capabilities include parameters such as Central Processing Unit (CPU) and Graphics Processing Unit (GPU) performance, which determine the node’s ability to handle complex machine learning computations; memory availability, which impacts data storage and processing efficiency; network bandwidth, which affects the speed and stability of data exchange between nodes; and energy consumption constraints, which are crucial for power-sensitive devices such as IoT sensors or edge computing nodes. The system 100 evaluates these computational resource capabilities to assign appropriate AI learning tasks to the new node, ensuring that it does not become overloaded or underutilized. If the new node has high processing power, it may be assigned advanced AI model training tasks, while lower-capability nodes may focus on feature extraction or data pre-processing. Additionally, the system 100 dynamically adjusts network parameters based on the new node’s computational profile, optimizing communication pathways, learning model distribution, and workload balancing. By including computational resource capabilities in the topology change information, the system ensures that new nodes integrate efficiently without disrupting ongoing AI learning, leading to optimized performance, reduced latency, and enhanced scalability of the INL HiC session.
[0077] In accordance with an embodiment, the topology change information further includes a local Relevance Indicator (RI) for the new node. The local RI is a metric that quantifies the quality, reliability, and importance of the data and computational resources provided by the new node in relation to the ongoing AI training and inference processes. The local RI helps the system 100 prioritize and optimize task allocation, ensuring that high-quality data sources are given greater weight in the AI learning model. The local RI is determined based on multiple factors, including data accuracy, completeness, and redundancy, where nodes providing highly relevant or unique data receive a higher RI value. Additionally, computational efficiency is taken into account, as nodes with greater processing power and lower latency are more effective in contributing to real-time learning. The network proximity and connectivity of the node also influence the RI, as nodes with stable and high-bandwidth connections are preferable for seamless data exchange and collaboration. By incorporating the local RI in the topology change information, the system 100 ensures dynamic optimization of AI model training, where nodes with higher relevance contribute more significantly to the fusion process, feature extraction, and decision-making algorithms. This results in improved learning accuracy, reduced computational overhead, and enhanced overall network efficiency. Furthermore, the RI allows for adaptive learning, where AI models can dynamically adjust their focus based on the most relevant and high-quality data sources, ensuring robust and scalable AI processing within the HiC session.
[0078] In accordance with an embodiment, the topology change information includes a detach notification. which is a critical signal used to inform the system that a node is leaving the In-INL HiC session. The detach notification serves as a formal indication that the node is no longer available to participate in AI training, inference, or data exchange within the network. This can occur due to various reasons, such as network disconnection, device shutdown, mobility of user equipment (UE) , resource constraints, or administrative removal. When the node initiates detachment, its HiC Agent generates a detach notification, which includes key information such as the node ID, reason for detachment, last known status, and impact assessment on the network topology. This notification is then transmitted to the HiC Controller 110 , which is responsible for processing the topology update. Upon receiving the detach notification, the HiC Controller 110 validates the request, ensuring it aligns with network policies, and then forwards it to the AF / NF node for further evaluation. Once the detachment is confirmed, the system 100 updates the network topology by reallocating computational tasks, reassigning parent-child communication links, and modifying AI model fusion parameters to maintain learning efficiency. If necessary, replacement nodes may be activated to compensate for the loss of data or computational power from the detached node. By including a detach notification in the topology change information, the system ensures that network disruptions are minimized, AI learning remains stable, and resource allocation is dynamically adjusted to maintain optimal performance within the active HiC session.
[0079] In accordance with an embodiment, the topology change information relates to a new node joining the network which has the new data modality. The data modality refers to the type or format of data processed by a node, such as image, audio, text, sensor data, or video streams. When a new node joins with a new modality, the system 100 must ensure that the data can be seamlessly integrated into the existing AI learning process. The process begins when the HiC Agent within the new node detects its availability and transmits a join request to the HiC Controller 110. The request may include key topology change information, such as the node’s ID, computational resource capabilities, RI, and the newly available data modality. Upon receiving this information, the HiC Controller 110 validates the request, ensuring that the new modality is compatible with the ongoing learning tasks. Since the modality is new to the network, the HiC Controller 110 must update the AI fusion mechanisms at the LIF and the FC nodes to incorporate the additional data stream. The system assigns a parent LIF to the new node and updates communication pathways to facilitate data exchange. Additionally, the AI model may require reconfiguration to accommodate the new modality, including updating feature extraction, multi-modal fusion, and inference algorithms. Once the system finalizes these updates, it sends an update message to the new node, providing it with the necessary patterns, optimization parameters, and communication settings. The new node then begins contributing its data modality to the AI learning process, enhancing the network’s ability to perform more complex and comprehensive learning tasks. By efficiently handling the addition of new data modalities, the system ensures that AI models remain adaptable, scalable, and capable of learning from diverse and evolving datasets within the active HiC session.
[0080] In accordance with an embodiment, the topology change information relates to an existing node detaching from the active INL HiC session. The node may detach for various reasons, such as loss of network connectivity, hardware failure, energy constraints, system shutdown, mobility of the device, or administrative removal due to reconfiguration or load balancing requirements. The detachment of a node affects the hierarchical structure of the INL HiC session, requiring the system to update communication links, redistribute computational workloads, and adjust AI learning processes to maintain performance and efficiency.
[0081] The detachment process begins when the HiC Agent within the affected node detects that it can no longer participate in the network. The HiC Agent generates a detach notification, which contains essential topology change information, including the node ID, reason for detachment, computational capabilities before detachment, Relevance Indicator (RI) , and a summary of its contributions to the AI model. The detach notification is then sent to the HiC Controller (HicC) , which is responsible for managing topology updates. The HicC processes the request by verifying the detachment reason and assessing its impact on the overall learning framework. Since detaching a node may lead to a loss of data, computational power, or a break in communication pathways, the HiC Controller 110 must take corrective actions to prevent performance degradation. Once the detachment is confirmed, the HicC updates the network topology by reassigning the node’s parent-child relationships to ensure that remaining nodes continue to operate efficiently. If the detached node was a data contributor, the system checks if similar data is available from other nodes or if an alternative learning pathway needs to be established. If the node was responsible for feature extraction, data fusion, or inference, the system reallocates these tasks to other nodes with similar capabilities. Additionally, the HiC Controller 110 notifies other connected nodes, such as LIFs and FCs, to update their AI processing pipelines and communication links.
[0082] To maintain the efficiency of the active INL HiC session, the system 100 may also activate redundant or standby nodes to replace the detached node, ensuring that AI training and inference continue without disruption. If the detachment causes a critical gap in data availability, the system may modify the AI model’s fusion mechanisms to compensate for the missing information. Finally, the network completes the detachment process by synchronizing all active nodes with the updated topology and sending an acknowledgment message confirming that the system has successfully adapted to the change. By efficiently handling existing node detachment, the system ensures that AI learning remains robust, dynamic, and resilient, even when nodes leave the network. This approach prevents interruptions in distributed AI training and inference, optimizes resource allocation, and maintains seamless collaboration between remaining nodes, allowing the HiC session to continue operating with minimal performance impact.
[0083] During the first phase of signaling, the system 100 is further configured to send an update message related to patterns and parameters related to the received topology change information. The patterns correspond to the active INL HiC session and includes information concerning a neural network model, an optimisation method, related hyperparameters and a communication path. The patterns in the update message refer to predefined configurations that dictate how nodes interact, process data, and contribute to AI learning. The patterns include data processing patterns, which define how data is collected, transformed, and fused within the network; learning patterns, which specify the role of the node in the AI training hierarchy, such as Feature Extractor, Local Intermediate Fusioner, or Fusion Center; communication patterns, which define the hierarchical relationships between nodes and govern data exchange protocols; and optimization patterns, which outline the methodologies used for fine-tuning AI models. The parameters in the update message provide specific numerical configurations essential for maintaining network functionality. The specific numerical configurations include computational resource allocation, detailing the distribution of processing power, memory, and bandwidth among nodes; fusion parameters, specifying how multi-modal data is combined at different network levels; latency and throughput requirements, ensuring efficient data transmission and processing; and task assignment parameters, defining how AI training and inference tasks are distributed across the network. In addition to patterns and parameters, the update message contains details concerning the neural network model, ensuring that all nodes remain synchronized in the AI training process. The update includes model architecture, describing the structure of layers and connections in the AI model; model weights and biases, providing necessary adjustments to ensure consistent learning performance across distributed nodes; and training state, indicating whether the AI model is currently in training mode or inference mode. The optimization method used in AI training is also included in the update message, specifying details about the gradient descent algorithm, such as Stochastic Gradient Descent (SGD) or Adam optimizer; learning rate schedules, which define how the model adjusts learning efficiency based on performance; and regularization techniques, including dropout layers or batch normalization methods. The hyperparameters in the update message further refine the AI learning process and include batch size, which controls the number of data samples processed simultaneously; epoch count, determining the number of iterations the model undergoes during training; and momentum and weight decay, which optimize convergence speed and stability.
[0084] Further, the update message also includes information regarding the communication path, which defines how nodes exchange data within the network. The parent-child relationships between nodes are specified, ensuring proper routing of data and learning updates. Additionally, the update message provides information on data transmission frequency, governing how often nodes communicate updates with their assigned parent nodes, and error handling mechanisms, ensuring data integrity through retransmission protocols in case of network failures. Upon detecting the topology change, the system 100 generates and transmits an update message to relevant nodes, including the new or modified node itself, the parent nodes (such as LIF or FCs) , and the Application Function (AF) or Network Function (NF) nodes responsible for network validation and policy enforcement.
[0085] The transmission of the update message follows a structured signaling protocol, ensuring efficient, low-latency communication across the network. The system first encodes and packages the update message, incorporating error-checking mechanisms such as CRC or checksum verification to maintain data integrity. The message is then transmitted to the target nodes using optimized network signaling protocols, including wireless communication frameworks such as 5G NR or edge computing networks. Upon reception, the affected nodes send acknowledgment (ACK) signals to confirm successful reception of the update message. If errors are detected, error-correction protocols are initiated to ensure accurate data transmission. Finally, the newly updated node integrates into the network, adjusting its processing behaviour based on the received patterns and parameters. The HiC Controller 110 ensures that all parent-child relationships are updated, allowing AI training and inference to continue seamlessly without disruptions.
[0086] The structured transmission of the update message provides significant advantages in network adaptation, AI model consistency, and resource optimization. The update message enables the network to seamlessly integrate new or modified nodes, ensuring that AI training remains efficient and uninterrupted. The synchronization of AI models across nodes prevents inconsistencies and enhances overall system reliability. By dynamically adjusting resource allocation, computational workloads, and data fusion processes, the system optimizes network performance while minimizing latency. Furthermore, the scalability of the INL HiC session is enhanced, as the system can efficiently accommodate changes in network topology without requiring a complete retraining of AI models. The structured signaling process ensures real-time updates, maintaining low latency and high efficiency in data transmission and model synchronization.
[0087] During the second phase of signaling, the system 100 is further configured to send a required pattern related to the received topology change information to the HiC agent of the node which sent the topology change information. The required pattern represents a set of predefined instructions, configurations, and optimization parameters that dictate how the node should adjust its role, data processing behaviour, and communication structure in response to the topology change. This ensures that the node aligns with the updated network architecture, maintains synchronization with other participating nodes, and efficiently contributes to the AI learning process within the INL HiC session. Upon receiving the topology change information from an HiC agent, system 100 first analyses and processes the update, determining the optimal pattern to be applied to the reporting node based on factors such as its computational resource capabilities, data modality, RI, and role within the INL HiC hierarchy. If the topology change involves a new node joining, the system identifies its designated parent node (such as a LIF or FC) ) and generates a pattern that defines its role, required communication pathways, and AI model integration parameters. If the topology change involves an existing node detaching, system 100 sends a pattern instructing the reporting node on detachment procedures, task reallocation protocols, and final data synchronization requirements before exiting the network. To ensure seamless adaptation, system 100 encodes the required pattern into an update message, which includes details such as model configurations, learning task assignments, fusion parameters, and network optimization settings. The message is formatted according to a structured signaling protocol that ensures secure, low-latency transmission. The updated message is then transmitted to the HiC agent of the node, which processes the received pattern and applies the necessary changes. If the pattern involves reconfiguration of AI learning models, the HiC agent updates its feature extraction, fusion algorithms, or inference parameters to align with the revised network structure. If the pattern involves communication pathway modifications, the node adjusts its parent-child relationships, transmission frequency, and data exchange methods to maintain network efficiency. Once the required pattern is applied, the HiC agent generates an acknowledgment response, confirming successful implementation of the update. The system 100 then verifies the adaptation process by monitoring performance metrics, data integrity, and network latency to ensure that the node is functioning correctly within the HiC session. If necessary, system 100 may issue further refinements to the required pattern based on real-time feedback from the node. By dynamically sending customized patterns in response to topology changes, system 100 ensures that the INL HiC session remains adaptive, scalable, and resilient to evolving network conditions, facilitating uninterrupted AI training and inference across the distributed system.
[0088] Further, the system 100 is further configured to send an updated pattern related to the received topology change information to a parent node of the node having the HiC agent which sent the topology change information. updated pattern ensures that the parent node dynamically adjusts its operations, communication pathways, and AI learning processes to accommodate the modifications caused by the topology change. The parent node, which may be a Local Intermediate Fusioner (LIF) or a Fusion Center (FC) , plays a role in aggregating data, performing feature fusion, and distributing AI model updates across the hierarchical network structure. Thus, any topology changes at a lower level-such as a new node joining, an existing node detaching, or a data modality being added or removed-requires the parent node to be updated accordingly.
[0089] Upon receiving topology change information, the system 100 first analyses the impact of the modification on the overall network structure and determines the necessary adjustments to be applied at the parent node. The determines the necessary adjustments involves evaluating the communication dependencies, resource distribution, data fusion mechanisms, and AI model updates required to maintain seamless network functionality. Once the analysis is complete, the system 100 generates an updated pattern, which is a structured set of operational instructions, optimization parameters, and reconfigured AI learning protocols designed to adapt the parent node to the new network state.
[0090] The updated pattern sent to the parent node may include key parameters such as revised communication pathways, recalibrated fusion algorithms, modified task distribution logic, and updated learning model configurations. For instance, if a new node with an existing or new data modality joins, the parent node must update its data aggregation structure to incorporate the new data source efficiently. If a node detaches, the parent node must redistribute the computational workload among remaining child nodes to prevent data loss or AI model degradation. Additionally, if a modality becomes unavailable, the parent node must adjust feature extraction and fusion mechanisms to compensate for the missing data. To ensure low-latency and error-free transmission, the system 100 encodes the updated pattern into a structured update message, incorporating error-detection mechanisms to prevent transmission loss. The message is then sent to the parent node via an optimized network signaling protocol, ensuring that the updated configuration is applied in real-time without disrupting the ongoing INL HiC session. Upon receiving the update, the HiC agent within the parent node processes the updated pattern and initiates the necessary modifications, which may involve restructuring AI inference paths, adjusting synchronization intervals with other network elements, or refining computational resource allocation.
[0091] Once the parent node successfully applies the updated pattern, it generates an acknowledgment response and transmits it back to the system 100, confirming successful adaptation to the topology change. To further validate the network's stability, the system 100 continuously monitors performance metrics, AI model accuracy, and communication efficiency, ensuring that the parent node operates optimally in response to the topology update. If additional refinements are needed, system 100 may send subsequent updates to fine-tune the parent node’s operations further. By dynamically sending updated patterns to parent nodes in response to topology changes, the system 100 ensures that the INL HiC session remains resilient, adaptable, and capable of handling evolving network conditions. This structured approach prevents AI model inconsistencies, optimizes data fusion processes, and maintains seamless connectivity between hierarchical network layers, ultimately enhancing the performance and scalability of the distributed learning framework.
[0092] FIG. 2 is a flowchart method of signaling for network topology change in the active In-Network Learning (INL) Heterarchical Intelligent Collaboration (HiC) session in the communications network, in accordance with an embodiment of the present disclosure. FIG. 2 is explained in conjunction with FIG. 1. With reference to FIG. 2, there is shown a flowchart of a method 200 of signaling for network topology change in the active In-Network Learning (INL) Heterarchical Intelligent Collaboration (HiC) session in the communications network. The method 200 includes steps 202 to 208.
[0093] At step 202, the method 200 includes receiving topology change information from any of a plurality of HiC agents, where an HiC agent is located within a respective node of the communications network. The process of receiving topology change information begins with continuous monitoring by each HiC agent embedded within the respective nodes of the communications network. Each HiC agent is responsible for detecting real-time changes in network topology, such as changes in connectivity, computational resource status, and data modality availability. When a topology change event occurs, the HiC agent at the affected node collects critical information regarding the change. This topology change information includes, but is not limited to, the node ID, type of topology change, computational resource capabilities, RI, updated communication links, data modality status, and performance metrics relevant to the active HiC session. The HiC agent compiles this information into a structured topology change message, ensuring that all necessary details are included for further processing.
[0094] Once the topology change message is prepared, the HiC agent transmits the message to the central system, which is responsible for managing topology changes within the active HiC session. The transmission process utilizes a structured signaling protocol optimized for low latency and high reliability. The HiC agent may transmit the topology change message via dedicated network signaling channels, which may include wireless transmission protocols, Ethernet-based connections, or dedicated communication pathways within the hierarchical network architecture. The message is then received by the HiC Controller 110, which serves as the primary node responsible for processing and validating topology change events.
[0095] Upon receiving the topology change information, the HiC Controller 110 first performs message authentication and validation to ensure that the received data is accurate, complete, and originates from a legitimate HiC agent. The system 100 may implement error-checking mechanisms, such as checksum verification, cyclic redundancy checks (CRC) , or cryptographic authentication protocols, to prevent transmission errors or unauthorized modifications. Once the message is validated, the HiC Controller 110 analyzes the impact of the topology change on the overall network structure and AI learning processes. This analysis includes evaluating task redistribution requirements, AI model fusion updates, communication link adjustments, and computational resource realignment to maintain system performance and stability.
[0096] After processing the topology change information, the HiC Controller 110 forwards the validated update to the AF or Network Function NF node for further policy enforcement, resource allocation, and system-wide adjustments. The AF / NF node evaluates the implications of the topology change in terms of network performance, security policies, and load balancing, ensuring that the modification aligns with predefined system requirements. Once the topology change is approved, the AF / NF node sends an update confirmation message back to the HiC Controller 110, authorizing the necessary adjustments. The HiC Controller 110 then initiates real-time network updates based on the received topology change information. The updates may include assigning a new parent node to the affected node, modifying AI learning and inference parameters, updating communication pathways, and optimizing resource allocation across the network. If the topology change involves a node detaching from the network, the HiC Controller 110 ensures that all remaining nodes adjust accordingly, preventing data loss or interruptions in AI learning. If a new node with a new or existing data modality joins, the system integrates the node into the hierarchical structure, assigning the appropriate role and optimizing fusion processes to incorporate the newly available data. Throughout the process, the HiC Controller 110 continuously monitors network performance, AI learning accuracy, and data synchronization efficiency to ensure that the topology change does not negatively impact system operations. If further refinements are required, the system 100 may issue additional updates to fine-tune the network configuration based on real-time performance feedback. The process of receiving topology change information is therefore a multi-step, structured, and automated procedure that enables the INL HiC session to remain scalable, adaptable, and efficient, ensuring that AI training and inference continue uninterrupted despite dynamic changes in network topology.
[0097] At step 204, the method 200 further includes sending an update message related to patterns and parameters related to the received topology change information. The patterns correspond to the active INL HiC session and includes information concerning a neural network model, an optimisation method, related hyperparameters and a communication path. The process of sending the update message begins after the HiC Controller 110 has received, validated, and analysed the topology change information. Once the topology update is assessed, system 100 generates an update message containing the revised patterns and parameters to be applied to the affected nodes. The update message is structured into distinct components, each serving a specific purpose in the reconfiguration process. The first component of the message pertains to the patterns, which define the structural adjustments required within the INL HiC session. The patterns include modifications to data processing workflows, ensuring that AI learning continues smoothly despite changes in network topology. Additionally, learning patterns within the update message specify how the AI model will adjust to new nodes, detached nodes, or modified modalities. Communication patterns define how the affected node should interact with parent and child nodes, updating parameters such as data exchange frequency, transmission hierarchy, and synchronization intervals. Optimization patterns specify fine-tuning techniques required to enhance AI model performance, minimize computational overhead, and maximize data utility within the learning framework.
[0098] In addition to patterns, the update message includes parameters that contain critical configuration settings for AI model adaptation. The neural network model component of the update message specifies the architecture of the AI model, including input layers, hidden layers, activation functions, and weight distribution updates. If a new node with additional computational resources is introduced, the system 100 may reallocate learning tasks across different nodes by updating the AI model configuration accordingly. The update message also includes details regarding the optimization method, specifying which gradient descent algorithms, learning rate schedules, or regularization techniques should be applied to fine-tune the AI model after the topology change. These optimization methods ensure that AI learning continues without degradation in accuracy, efficiency, or convergence speed.
[0099] Furthermore, the update message includes hyperparameters, which are essential for controlling the behaviour of the neural network after the topology change. The hyperparameters define batch size, number of epochs, loss function parameters, momentum values, and weight decay coefficients, ensuring that the AI model remains optimized for distributed learning across all updated nodes. If the new data modality is introduced, the system 100 updates the AI model's fusion parameters to effectively incorporate the new type of data. If an existing data modality is no longer available due to a node detachment, the system 100 adjusts the hyperparameters to compensate for the missing data while maintaining inference accuracy. The final component of the update message pertains to the communication path, which ensures that affected nodes can seamlessly integrate into the updated network structure. The final component of the update message section of the message includes instructions on parent-child relationships, specifying whether the node should communicate with a Local LIF, FC, or other edge nodes. It also contains data transmission policies, specifying the frequency and priority of data exchanges, ensuring that bandwidth and computational resources are utilized efficiently. In cases where a node detaches, the communication path section of the update message redirects data transmission to alternative nodes, ensuring continuous operation of the AI learning process without bottlenecks or data loss.
[0100] Once the update message is compiled, the system 100 transmits it to the affected nodes using a structured signaling protocol designed for low-latency, high-reliability communication. The transmission may occur over wireless networks, dedicated high-speed links, or edge computing channels, depending on the network infrastructure. Upon receiving the update message, the HiC agent within the affected node decodes the instructions, applies the new patterns and parameters, and initiates internal reconfiguration to align with the updated network state. If the update involves AI model modification, the node downloads the revised neural network weights, optimization functions, and hyperparameter configurations, ensuring synchronization with the rest of the network.
[0101] To confirm the successful application of the update message, the HiC agent sends an acknowledgment (ACK) response back to the HiC Controller 110, verifying that the patterns and parameters were correctly implemented. If an issue is detected, the system 100 initiates error-handling protocols, retransmitting the update message or issuing corrective instructions to ensure proper configuration. Finally, after the update is successfully applied across all affected nodes, the system 100 monitors AI performance, data consistency, and network efficiency, ensuring that the INL HiC session remains stable, adaptive, and optimized for real-time AI learning.
[0102] By dynamically sending update messages containing patterns and parameters, the system 100 ensures that network topology changes are seamlessly integrated without disrupting the ongoing AI training and inference processes. The structured approach maintains scalability, adaptability, and resilience, allowing the INL HiC session to operate efficiently even in dynamically changing network environments.
[0103] At step 206, the method 200 further includes sending a required pattern related to the received topology change information to the HiC agent of the node which sent the topology change information The HiC Controller 110, having received and processed the topology change information and subsequent update message from the AF / NF node, now initiates targeted communication to configure the affected node appropriately. The required pattern sent to the originating HiC agent contains comprehensive instructions and parameters necessary for the node to function optimally within the modified network topology.
[0104] The required pattern encompasses multiple essential components carefully calibrated based on the specific type of topology change that has occurred. For new nodes joining with existing data modalities, this pattern includes the localized loss function parameters specific to the node's computational role and data characteristics. The localized loss function parameters are mathematically derived to ensure proper weighting of the node's contribution to the overall distributed neural network model. The required pattern further includes detailed neural network model specifications, such as layer configurations, activation functions, and initialization parameters, all tailored to ensure compatibility with the existing network architecture. Optimization method directives form another critical element of the required pattern, providing instructions regarding gradient descent variants, learning rates, momentum values, and regularization techniques appropriate for the node's computational capabilities and data quality. Hyperparameter configurations are precisely specified, including batch sizes, convergence criteria, and epoch limitations, which have been determined through analysis of the node's RI and computational resource capabilities provided during the first phase of signaling.
[0105] The required pattern also establishes a comprehensive communication framework for the node, defining its parent-child relationships within the hierarchical network structure. This includes identification of the parent LIF or FC to which the node should transmit its processed data, as well as communication timing parameters such as frequency, interval specifications, and synchronization requirements. The timing parameters are important for maintaining the temporal integrity of the distributed learning process across the network. Additionally, the required pattern includes resource allocation directives that specify how the node should distribute its computational resources among various tasks, including feature extraction, data pre-processing, model training, and communication overhead. These resource allocation instructions are specifically optimized based on the computational capabilities information provided by the node during the first phase of signaling.
[0106] For nodes with existing data modalities, the required pattern also includes modality-specific parameters that define how the node's data should be processed, normalized, and integrated with similar data modalities present elsewhere in the network. These parameters ensure that the node's contribution to intra-modality fusion is properly calibrated to maintain the integrity of the distributed learning process. The transmission of this required pattern occurs through secure and reliable communication channels established during the HiC session initialization, employing encryption and error-correction mechanisms to ensure accurate delivery. The pattern is transmitted in a standardized format that can be efficiently parsed and implemented by the receiving HiC agent, minimizing the computational overhead required for configuration. This carefully orchestrated process ensures that the node which initiated the topology change can seamlessly integrate into or adapt within the active INL HiC session without disrupting the ongoing distributed learning process, representing a significant advancement over prior art methods that would typically require re-establishing the entire learning session when network topology changes occur.
[0107] At step 208, the method 200 further includes sending an updated pattern related to the received topology change information to a parent node of the node having the HiC agent which sent the topology change information. Following the processing of topology change information and receipt of the update message from the AF / NF node, the HiC Controller 110 initiates targeted communication with the parent node of the originating node to ensure coordinated adaptation across the hierarchical network structure. The updated pattern sent to the parent node encompasses a comprehensive set of modified instructions and parameters carefully calibrated to accommodate the topology change while preserving the integrity of the distributed learning process. For parent nodes that function as Local Intermediate LIFs, the updated pattern includes revised information about their child nodes, incorporating details about newly added nodes or removing information about detached nodes depending on the nature of the topology change. The child node information comprises computational capabilities, RIs, and available data modalities that are essential for the parent node to effectively coordinate intra-modality fusion operations. The updated pattern includes modified communication path definitions that specify revised data flow routes within the network topology. These path definitions precisely identify which nodes should communicate with each other, at what frequency, and using which protocols, ensuring optimal data exchange despite the altered network structure. For scenarios involving node addition, these paths incorporate the new node into existing data flows, while for node detachment scenarios, the paths are reconfigured to exclude the detached node and reroute communications as necessary. For LIF parent nodes, these include revised joint component parameters that define how the LIF should combine information from its child nodes. The joint parameters are mathematically derived based on the number of child devices having each data modality, their computational capabilities, and their relevance indicators. The trade-off parameter is also recalibrated to ensure proper weighting between joint and local components in the fusion process. For nodes with multiple data modalities, the pattern includes separate parameters for each modality, allowing for modality-specific fusion operations.
[0108] The updated pattern further provides revised parameters related to the parent node's role as a child for higher-level fusion operations, particularly when communicating with the FC. These include updated local component parameters that define how the parent node's fused information should be weighted in higher-level fusion operations. The updated pattern parameters ensure that changes at lower levels of the network hierarchy are properly propagated upward, maintaining the coherence of the distributed learning process across all network levels. For the FC parent nodes, the updated pattern includes revised inter-modality fusion parameters, such as joint component parameters and trade-off parameters, that govern how information from different data modalities is combined to produce the final output. The parameters are recalculated based on the modified network topology to ensure optimal fusion performance despite the changes in available data sources.
[0109] The transmission of the updated pattern occurs through secure and reliable communication channels with robust error detection and correction mechanisms. The updated pattern is encoded in a standardized format that allows for efficient parsing and implementation by the receiving parent node, minimizing processing overhead and ensuring rapid adaptation to the topology change. This comprehensive update process ensures that parent nodes can seamlessly adjust their operations to accommodate changes in their child nodes, maintaining the continuity and effectiveness of the distributed learning process across the network hierarchy and representing a significant advancement over prior art methods that would typically require re-establishing hierarchical relationships when network topology changes occur.
[0110] The steps 202 to 208 are only illustrative, and other alternatives can also be provided where one or more steps are added, one or more steps are removed, or one or more steps are provided in a different sequence without departing from the scope of the claims herein.
[0111] There is provided a computer program comprising instructions that, when executed by a computer system, cause the computer system to implement the method 200. In an example, the instructions are implemented on the computer-readable media, which include, but are not limited to, Electrically Erasable Programmable Read-Only Memory (EEPROM) , Random Access Memory (RAM) , Read-Only Memory (ROM) , Hard Disk Drive (HDD) , Flash memory, a Secure Digital (SD) card, Solid-State Drive (SSD) , a computer-readable storage medium, and / or CPU cache memory.
[0112] FIG. 3 is an exemplary diagram depicting signaling flow when a new device with existing data modality joins in an active INL HiC session, in accordance with an embodiment of the present disclosure. FIG. 3 is explained in conjunction with elements of FIGS. 1 to 2. With reference to FIG. 3, there is shown an exemplary diagram 300 depicting a comprehensive signaling flow for network topology change when a new device (for example, a new UE device with a node 304A) with existing data modality joins the active INL HiC session.
[0113] The exemplary diagram 300 includes a section 302, including a node 304A of an HiC Agent in the new UE device, a node 304 of an HiC Agent in UE device 102, a node 306 of an HiC Agent in Local Intermediate Fusioner (LIF) , and a node 308 of a HiC Agent in the FC 108. Further, the exemplary diagram 300 includes the HiC Controller 110 and a node 310 of application function (AF) or network function (NF) . A communication block 314 depicts process and communication according to HiC. The HiC Controller 110 and the node 310 of AF / NF are interconnected through various communication paths.
[0114] In operation, the signaling flow follows a structured two-phase approach. The first phase begins with the node 304A, which sends information to the HiC Controller 110 via communication path 316A. The information includes an HiC agent capacity of the node 304A, RI, and a data modality report. The data modality report contains essential information including the computational resource capabilities of the new UE device having the node 304A, available data modality type, and local Relevance Indicator (RI) that quantifies data quality. Upon receiving the information, the HiC Controller 110 processes the information and forwards “new device info" to the node 310 of the AF / NF node via a communication path 316B. The node 310 of AF / NF analyzes this information to determine if the new UE device should be accepted into the active INL HiC session. The node 310 then responds with an acknowledgment signal (sent via a communication path 318) indicating acceptance or rejection, followed by sending an update message to the HiC controller 110 containing updated patterns and parameters via a communication path 320. Once the HiC controller 110 receives the update message, the first phase of signaling comes to an end.
[0115] The second phase begins after the HiC Controller 110 receives the update message from the node 310. The HiC Controller 110 then distributes configurations to three different nodes. First, the HiC Controller 110 sends initial configurations to the node 304A via path 322. The initial configurations include the required pattern, model and optimization parameters, and communication patterns necessary for the new UE device to participate in the active learning session. Simultaneously, the HiC Controller 110 sends “HiC updated pattern and updated communication for LIF" to the node 306 via path 324A. The “HiC updated pattern and updated communication for LIF" includes information about the new UE device, updated communication patterns that incorporate the new UE device, and adjusting parameters that are essential for proper intra-modality fusion operations. Additionally, the HiC Controller 110 transmits "HiC updated pattern for FC" to the node 308 via path 324B. The "HiC updated pattern for FC" primarily modifies parameters such as modality-based components (Lc) , joint component (LJC) and trade-off parameter (α) at the FC 108 to ensure proper integration of the data of the new UE device with the node 306 into the higher-level fusion operations.
[0116] A plurality of HiC agents (i.e., the HiC agent in the new UE device with the node 304A, the HiC agent in UE device 102, the HiC agent in the LIF, and the HiC agent in the FC 108 process and communicate according to the received HiC pattern with the help of communication block 314. Further, a status report is sent back to the HiC Controller 110 through path 328, confirming successful integration into the active HiC session. Throughout the process, the HiC Controller 110 is being monitored by a monitoring device 326 to oversee the network's adaptation to the topology change. When the active HiC session needs to be terminated, the HiC Controller 110 initiates the process of ending the active HiC session by sending an "End HiC" signal via a path 330 to the plurality of HiC agents, ensuring proper clean-up and shutdown across all network nodes. The comprehensive signaling flow demonstrates the efficient two-phase approach and seamlessly integrates the new device with existing data modality into the active INL HiC session without disrupting the ongoing distributed learning process.
[0117] FIG. 4 is an exemplary diagram depicting signaling flow when an existing node detaches from an active HiC session, in accordance with another embodiment of the present disclosure. FIG. 4 is explained in conjunction with elements of FIGS. 1 to 3. With reference to FIG. 4, there is shown an exemplary diagram 400 depicting a scenario of signaling when a device detaches from an active INL HiC session. The exemplary diagram 400 includes a section 402, including a node 404 of an HiC Agent in the device, the node 304 of an HiC Agent in UE device 102, the node 306 of an HiC Agent in Local Intermediate Fusioner (LIF) , and the node 308 of the HiC Agent in the FC 108. Further, the exemplary diagram 400 further includes the HiC Controller 110 and the node 310 of AF or NF. The communication block 314 depicts process and communication according to HiC. The HiC Controller 110, and the node 310 of AF / NF, are interconnected through various communication paths.
[0118] In operation, the signaling process for handling the node detachment begins when the device decides to detach from the active INL HiC session. The process commences with the HiC agent of the device sending the detach notification to the HiC Controller 110 via a path 416A. The detach notification informs the HiC Controller 110 that the device is leaving the active HiC session. Upon receiving the detach notification, the HiC Controller 110 forwards a detached device info to the node 310 of AF or NF via a path 416B. The AF / NF processes the detached device info and responds with a message regarding "HiC Update” back to the HiC Controller 110 via a path 418, providing necessary updates for the network to adjust to the topology change. After receiving the necessary updates from the AF / NF, the HiC Controller 110 enters the second phase of signaling, distributing updated patterns to the relevant nodes in the network. First, the HiC Controller 110 sends updated patterns to the node 306 of the HiC Agent in the LIF via a path 422A. The updated patterns include modified parameters and communication patterns that exclude the device that has been detached. Simultaneously, the HiC Controller 110 transmits the HiC updated pattern to the node 308 of the HiC Agent in the FC 108 via path 422B, particularly updating parameters such as parameters such as modality-based components (Lc) , joint component (LJC) and trade-off parameter (α) at the FC 108 to maintain the integrity of the neural network model despite the topology change.
[0119] A plurality of HiC agents (i.e., the Hi agent in the device with the node 404, the HiC agent in UE device 102, the HiC agent in the LIF, and the HiC agent in the FC 108 process and communicate according to the received HiC pattern with the help of the communication block 314. Following the processing, a status report is sent back to the HiC Controller 110 via a path 424, confirming the successful adaptation to the topology change. The HiC Controller 110 enters a monitoring phase with the help of the monitoring device 326 and, when required, sends an end HiC signal via a path 428 to terminate the active HiC session.
[0120] FIG. 5 is an exemplary diagram depicting signaling flow when a new data modality becomes available in an active INL HiC session, in accordance with another embodiment of the present disclosure. FIG. 5 is explained in conjunction with elements of FIGS. 1 to 4. With reference to FIG. 5, there is shown an exemplary diagram 500 depicting a signaling flow for network topology change when a new data modality becomes available for the active INL HiC session. The exemplary diagram 500 includes a section 502 containing multiple network elements: a node 504 of an HiC Agent in a new device with a new data modality, a node 506 of an HiC Agent in the LIF that will also be newly configured to handle the new data modality, and the node 308 of the HiC agent in the FC 108. The exemplary diagram 500 further includes the HiC Controller 110 and the node 310 of AF / NF, all interconnected through various communication paths.
[0121] In operation, the signaling flow follows a structured two-phase approach similar to other topology change scenarios but with specific adaptations for handling a new data modality. The first phase begins with the HiC Agent in the node 504 sending a plurality of indicators to the HiC Controller 110 via a communication path 516A. The plurality of indicators includes the HiC Agent capacity of the node 504, RI, and data modality report. The data modality report contains information including the computational resource capabilities of the node 504, the new data modality type that was previously unavailable in the network, and the local Relevance Indicator (RI) that quantifies data quality. Upon receiving the plurality of indicators, the HiC Controller 110 processes it and forwards "new device info" to the node 310 through a communication path 516B. The node 310 of AF / NF analyzes “new device info” information to determine how to integrate the new data modality into the active INL HiC session. The node 310 of AF / NF then responds with an acknowledgment signal (via a path 520) indicating acceptance or rejection, followed by an update message containing updated patterns and parameters via a communication path 518. As the HiC Controller 110 receives the update message, the first phase of signaling is completed.
[0122] The second phase begins after the HiC Controller 110 receives the update message from the node 310. The HiC Controller 110 then distributes configurations to three different nodes. First, the HiC Controller 110 sends HiC initial configurations to the node 504 via a path 522A. The initial configurations include the required pattern with parameters such as local components (Lk, m) specific to the new data modality, model and optimization parameters, and communication paths necessary for the new device to participate in the active learning session. The HiC Controller 110 sends initial configurations to node 506 of the HiC Agent in the LIF via a path 522B. This is because the node 506 the HiC Agent in the LIF needs to be newly configured to handle the previously unavailable data modality. The initial configurations include model and optimization parameters, information about the new device and its local components, parameters, communication patterns, parameters related to its role as LIF parent node, and parameters for its role as a child for FC.
[0123] Additionally, the HiC Controller 110 transmits a HiC updated pattern for the FC 108 to the node 308 via a path 524. The HiC updated pattern for the FC 108 modifies parameters modality-based components (Lc) , joint component (LJC) and trade-off parameter (α) at the FC 108 and includes updated communication patterns that incorporate the node 506 handling the new data modality, ensuring proper integration of the new data modality into the higher-level fusion operations.
[0124] Following the distribution of the configurations, all HiC Agents begin to "process and communicate according to HiC Pattern" with the help of the communication block 314. The network nodes then send a status report back to the HiC Controller 110 through a path 528, confirming the successful integration of the new data modality. Throughout this process, the monitoring device 326 oversees the network's adaptation to the topology change. When the active HiC session needs to be terminated, the HiC Controller 110 initiates the process by sending an "End HiC" signal via path 530 to all relevant HiC agents, ensuring proper clean-up and shutdown across all network nodes.
[0125] FIG. 6 is an exemplary diagram depicting signaling flow when an existing data modality becomes unavailable in an active INL HiC session, in accordance with another embodiment of the present disclosure. With reference to FIG. 6, there is shown an exemplary diagram 600 illustrating the signaling mechanism for adapting to a scenario where an existing data modality becomes unavailable for the active INL HiC session. The exemplary diagram 600 includes a section 602, which contains network elements: a node 604 of HiC Agent in an UE device, a node 606 of HiC Agent in LIF, and the node 308 of HiC Agent in the FC 108. The exemplary diagram 600 further includes the HiC Controller 110 and the node 310 of Application Function / Network Function (AF / NF) .
[0126] In operation, the signaling flow illustrated in FIG. 6 follows a structured two-phase approach specifically adapted for handling the unavailability of an existing data modality. The first phase begins with the node 604 of the HiC Agent in an UE device sending a detach notification to the HiC Controller 110 via a communication path 616A. The detach notification informs the HiC Controller 110 that a specific data modality is no longer available in the network, typically because a node containing a unique data modality is detaching from the network. Upon receiving the detach notification, the HiC Controller 110 processes it and forwards an information about the detached device to the node 310 of the AF / NF through a communication path 616B. The information of the detached device contains details about the data modality that will become unavailable. The node 310 of AF / NF analyzes this information to determine how the learning process should be adapted to accommodate the loss of this data modality. The node 310 of AF / NF then responds with an update message to the HiC Controller 110 via a path 618, providing updated patterns and parameters for the network to adjust to this change. This completes the first phase of signaling.
[0127] The second phase begins after the HiC Controller 110 receives the update message from the node 310 of the AF / NF. In this phase, the HiC Controller 110 distributes configurations to the affected nodes. First, the HiC Controller 110 sends a detach notification to the node 606 of HiC Agent in LIF via a path 620, confirming acknowledgment of the detachment from the active HiC session with respect to the specific data modality. Additionally, the HiC Controller 110 transmits "HiC updated pattern for FC" to the node 308 of HiC Agent in the FC 108 via a path 622. This update includes modified parameters associated with the now unavailable data modality. The communication pattern is also updated to exclude the detached LIF node from the active HiC session with respect to that specific data modality.
[0128] Following the distribution of these configurations, the HiC Agents in FC 108 begin to process and communicate according to the HiC Pattern with the help of the communication block 610. The network nodes then send a status report back to the HiC Controller 110 through a path 624, confirming successful adaptation to the topology change. Throughout this process, the monitoring device 326 oversees the network's adaptation to the loss of a data modality. When the session needs to be terminated, the HiC Controller 110 initiates this process by sending an "End HiC" signal via path 628 to all relevant HiC agents, ensuring proper clean up and shutdown across all network nodes.
[0129] Modifications to embodiments of the present disclosure described in the foregoing are possible without departing from the scope of the present disclosure as defined by the accompanying claims. Expressions such as "including" , "comprising" , "incorporating" , "have" , "is" used to describe, and claim the present disclosure are intended to be construed in a non-exclusive manner, namely allowing for items, components or elements not explicitly described also to be present. Reference to the singular is also to be construed to relate to the plural. The word "exemplary" is used herein to mean "serving as an example, instance or illustration" . Any embodiment described as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments or to exclude the incorporation of features from other embodiments. The word "optionally" is used herein to mean "is provided in some embodiments and not provided in other embodiments" . It is appreciated that certain features of the present disclosure, which are, for clarity, described in the context of separate embodiments, may also be provided in combination in a single embodiment. Conversely, various features of the invention, which are, for brevity, described in the context of a single embodiment, may also be provided separately or in any suitable combination or as suitable in any other described embodiment of the disclosure.
Claims
1.A method (200) of signaling for network topology change in an active In-Network Learning (INL) Heterarchical Intelligent Collaboration (HiC) session in a communications network, the method (200) comprising steps of:in a first phase of signaling:receiving topology change information from any of a plurality of HiC agents, where an HiC agent is located within a respective node of the communications network; andsending an update message related to patterns and parameters related to the received topology change information, wherein the patterns correspond to the active INL HiC session and includes information concerning a neural network model, an optimisation method, related hyperparameters and a communication path;and in a second phase of signaling:sending a required pattern related to the received topology change information to the HiC agent of the node which sent the topology change information; andsending an updated pattern related to the received topology change information to a parent node of the node having the HiC agent which sent the topology change information.2.The method (200) of claim 1, wherein one of the HiC agents is in a user equipment (UE) device (102) of the communications network.3.The method (200) of claim 1, wherein one of the HiC agents is in a base station (BS) device (104) of the communications network.4.The method (200) of claim 1, wherein the topology change information relates to a new node being added to the active INL HiC session, with an existing modality.5.The method (200) of claim 1, wherein the topology change information relates to a node detaching from the active INL HiC session.6.The method (200) of claim 1, wherein the topology change information relates to a new data modality becoming available for the active INL HiC session.7.The method (200) of claim 1, wherein the topology change information relates to an existing data modality becoming unavailable for the active INL HiC session.8.The method (200) of claim 4, wherein the topology change information includes computational resource capabilities for the new node.9.The method (200) of claim 8, wherein the topology change information includes a local Relevance Indicator (RI) for the new node.10.The method (200) of claim 5, wherein the topology change information includes a detach notification.11.The method (200) of claim 6, wherein the topology change information relates to a new node joining the network which has the new data modality.12.The method (200) of claim 7, wherein the topology change information relates to an existing node detaching from the active INL HiC session.13.The method (200) of claim 1, wherein in the first phase of signaling, the receiving step is performed by an HiC Controller (HicC) node.14.The method (200) of claim 13, wherein in the first phase of signaling, the HicC node forwards the received information to an Application Function / Network Function (AF / NF) node.15.The method (200) of claim 14, wherein the AF / NF node sends the update message to the HicC node.16.A system (100) comprising means adapted for carrying out all the steps of the method (200) according to any preceding method claim.17.A computer program comprising instructions for carrying out all the steps of the method (200) according to any preceding method claim, when said computer program is executed on a computer system.