Method and node for clustering nodes of multi-agent intelligent network

Abstract

A method for clustering nodes of a multi-agent intelligent network comprises each node sending a task information message to one or more nodes connected to that node. The task information message includes a task id of an origin node providing an identifier for an original task, an input token and a counter for the original local memory state that includes an original local memory state resulting from the original task, and each node receiving a response message to that node. The response message comprises an acknowledgement or a non-acknowledgement indicating whether the original local memory state was useful or not to the node for obtaining the output and for each the response message received and including an acknowledgement, adding a pair comprising an identifier of the origin node and an identifier of the node from which the response message was received to a node cluster table and clustering nodes.

Description

[0001] METHOD AND NODE FOR CLUSTERING NODES OF MULTI-AGENT INTELLIGENT NETWORK

[0002] TECHNICAL FIELD

[0003] The present disclosure relates generally to the field of wireless communication network management and, more specifically, to a method for clustering nodes of a multi-agent intelligent network. Furthermore, the present disclosure relates more specifically to a node to be used in the multi-agent intelligent network, such as by providing a protocol for devices’ clustering based on linear attention.

[0004] BACKGROUND

[0005] In rapidly advancing domain of communication systems, upcoming future communication systems are anticipated to heavily rely on cooperation of distributed agents that are used to combine processing compatibility of each of the intelligent agents and produce results. Moreover, each of the distributed agents are used to jointly process multimodal data (e.g., images, videos or sensing data collected by each of the distributed agent) to adapt changing environments, solve various complex tasks, and share obtained data with other distributed agents, such as by distributing all relevant data (i.e., tokens) across the distributed agents. However, such operation poses significant technical challenges that include inefficient performance of the distributed agents when sharing relevant input information with each other, as none of the agents are informed in advance if the data received is useful or not. Additionally, the standard approach of exchanging all input data among all the distributed agents for joint evaluation is often infeasible due to communication bandwidth limitations and privacy constraints.

[0006] Conventional communication systems, such as conventional self-attention mechanism cannot be computed in a distributed manner and require all the relevant data (i.e., the tokens) to be transmitted to a central server for joint computation. On the other hand, the transmission of the relevant data from each of the distributed agents to the central server is not feasible due to limitations of communication bandwidth and privacy constraints. Moreover, certain attempts have been made to overcome the communication bandwidth and privacy constraints, such as by using self-attention of a transformer, ring attention, retentive networks, message passing neural networks, and the like. However, such attempts fail due to many reasons, such as unavailability of the relevant data at a single place, complex computational cost, restrictive data flow, and the like. Therefore, there exists a technical problem of how to allow each node to discover a group of other nodes to collaborate with and share local knowledge.

[0007] Therefore, in light of the foregoing discussion, there exists a need to overcome the aforementioned drawbacks associated with the conventional methods and conventional nodes for clustering nodes of the multi-agent intelligent network.

[0008] SUMMARY

[0009] The present disclosure provides a method for clustering nodes of a multi-agent intelligent network and a node to be used in the multi-agent intelligent network. The present disclosure provides a solution to the existing problem of how to allow each node to discover a group of other nodes to collaborate with and share local knowledge. 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 the method for clustering the nodes of the multi-agent intelligent network and the node to be used in the multi-agent intelligent network, such as for clustering devices based on linear attention.

[0010] 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. In one aspect, the present disclosure provides a method for clustering nodes of a multi-agent intelligent network. Moreover, each node is arranged to receive input tokens and to process the input tokens to obtain an output of the node related to a task and the method includes sending a task information message to one or more nodes connected to that node and the task information message comprises a task id of an origin node providing an identifier for an original task, an input token and a counter for the original local memory state. In addition, the input token includes an original local memory state resulting from the original task, and each node receiving a response message from the one or more nodes connected to that node. The response message comprises an acknowledgement or a non-acknowledgement indicating whether the original local memory state was useful or not to the node for obtaining the output, and the method comprises that node for each the response message received and including an acknowledgement, adding a pair comprising an identifier of the origin node and an identifier of the node from which the response message was received to a node cluster table, and wherein the method comprises clustering nodes based on the cluster table.

[0011] Advantageously, the method is used for clustering nodes in the multi-agent intelligent network in order to enhance the overall multi-agent intelligent network efficiently with an improved communication performance. By allowing the nodes to share taskspecific data and memory states without the need for centralized control or raw data sharing, the method is used to allow distributed processing and collaboration between relevant nodes. The use of task identifiers, memory state counters, and clustering tables ensures that nodes communicate only with the most relevant peers, reducing unnecessary data transmissions and optimizing resource utilization. Additionally, the memory state that is useful indicates that the input data of the origin node is also useful, which is beneficial for future collaboration and knowledge sharing between the origin node and the receiving node. Furthermore, the application of the decay factor to outdated data and using positional embeddings further refine the ability of the multi-agent intelligent network to prioritize recent and meaningful information. As a result, the clustering of the nodes along with linear attention and learned matrices, enhances the capability of the multi-agent network to handle large- scale, complex environments while maintaining data privacy, reducing computational overhead, and improving decisionmaking accuracy.

[0012] In another aspect, the present disclosure provides a node configured to be used in a multi-agent intelligent network, wherein the node is arranged to receive input tokens and to process the input tokens to obtain an output of the node related to a task, wherein the node is configured to send a task information message to one or more nodes connected to the node. The task information message comprises a task id of an origin node providing an identifier for an original task, an input token and a counter for the original local memory state. Moreover, the input token includes an original local memory state resulting from the original task, and to receive a response message from the one or more nodes connected to that node. In addition, the response message comprises an acknowledgement or a non-acknowledgement indicating whether the original local memory state was useful or not to the node for obtaining the output, and the node is further configured to, for each the response message received and including an acknowledgement, add a pair comprising an identifier of the origin node and an identifier of the node from which the response message was received to a node cluster table. Furthermore, the node is configured to cluster nodes based on the cluster table.

[0013] The node achieves all the advantages and technical effects of the method of the present disclosure.

[0014] It is to be appreciated that all the aforementioned implementation forms can be combined.

[0015] 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 that 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.

[0016] 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.

[0017] BRIEF DESCRIPTION OF THE DRAWINGS

[0018] 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.

[0019] Embodiments of the present disclosure will now be described, by way of example only, with reference to the following diagrams wherein:

[0020] FIG. 1 is a block diagram that illustrates a multi-agent intelligent network providing distributed attention, in accordance with an embodiment of the present disclosure;

[0021] FIG. 2 is a flowchart of a method for clustering nodes of a multi-agent intelligent network, in accordance with an embodiment of the present disclosure;

[0022] FIG. 3 is a diagram that depicts a node configured to be used in a multi-agent intelligent network, in accordance with an embodiment of the present disclosure;

[0023] FIG. 4 is a diagram that illustrates an exemplary scenario depicting each node in a mesh topology of distributed nodes to discover groups of other nodes to collaborate with, based on the mutual usefulness of local data, in accordance with an embodiment of the present disclosure;

[0024] FIG. 5 is a diagram that illustrate an exemplary scenario for multi-round message exchanges between the nodes, in accordance with an embodiment of the present disclosure; and

[0025] FIG. 6 is a diagram that illustrates an exemplary scenario for evaluating usefulness of the received state, in accordance with an embodiment of the present disclosure.

[0026] 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.

[0027] DETAILED DESCRIPTION OF EMBODIMENTS

[0028] 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. FIG. 1 is a block diagram that illustrates a multi-agent intelligent network providing distributed attention, in accordance with an embodiment of the present disclosure. With reference to FIG. 1 , there is shown a decentralized mesh network of independent agents, such as a first node 102A, a second node 102B, a third node 102C, a fourth node 102D, and a fifth node 102E.

[0029] There is provided the multi-agent intelligent network 100 provides distributed attention by allowing multiple agents to independently process and focus on different parts of their local data while communicating and collaborating with each other in order to enhance the efficiency, robustness, and data privacy while handling complex and large-scale operations in a decentralized and efficient manner with reduced communication overhead. Each agent (e.g., the first node 102A, the second node 102B, the third node 102C, the fourth node 102D, and the fifth node 102E) in the wireless network includes multi-head network model configured to receive local input data (i.e., a sequence of input tokens) and further provide output data, which is a sequence of output tokens translated into a response that is further utilized for computing the distributed attention. Moreover, each of the multi-head network model includes at least one linear attention layer, which may be multi-layer and multi-head.

[0030] In operation, each of the node in the multi-agent intelligent network 100 receives its own set of local input embeddings that represent the information about the local data input of the agent. The nodes, such as the first node 102A, the second node 102B, the third node 102C, the fourth node 102D, and the fifth node 102E applies positional embedding to its local input embeddings to provide information about the position of each input element, allowing for accurate and context-aware predictions. Furthermore, the nodes are configured to process their local data independently, determining sets of keys, values, and queries using learned matrices in order to compute local memory states based on these processed inputs. Thereafter, the nodes are configured to share these processed memory states with other nodes in the multi-head intelligent network 100. Upon receiving the memory states from other nodes, each of the node evaluates the usefulness of the received information using its linear attention mechanism. If the received memory state is deemed useful, the corresponding node is configured to incorporate the received memory state into its combined memory state that allows for collaborative problem-solving and knowledge sharing without the need to exchange raw data, thereby maintaining privacy and reducing communication overhead. Moreover, the mesh structure of the network, as depicted in FIG. 1 , allows for flexible and dynamic communication patterns between the nodes. Each of the node is configured to connect and share information with any other agent in the multi-head intelligent network 100, enabling efficient information flow and adaptive collaboration based on the relevance and usefulness of data. Therefore, such decentralized, mesh-based architecture supports scalable and flexible accommodation of new agents without requiring changes to the existing network structure, enabling seamless expansion and integration of additional nodes as per the requirement.

[0031] FIG. 2 is a flowchart of a method for clustering nodes of a multi-agent intelligent network, in accordance with an embodiment of the present disclosure. With reference to FIG. 2, there is shown a flowchart of a method 200 for clustering nodes of the multi-agent intelligent network 100.

[0032] There is provided the method 200 for clustering nodes of the multi-agent intelligent network 100 and a node of the multi-agent intelligent network 100 is arranged to receive input tokens and to process the input tokens to obtain an output of the node related to a task. The method 200 is used to enhance the efficiency and performance of the multi-agent intelligent network 100 by enabling nodes to identify and collaborate with other relevant nodes without the need for centralized control or raw data sharing.

[0033] At step 202, the method 200 includes sending a task information message to one or more nodes connected to that node. In an implementation, the node is configured to generate the task information message that includes information, such as ID, specific task ID, and any relevant metadata or requirements associated with the task. Thereafter, the node is configured to send the generated task information to the one or more nodes that are connected with the node. Moreover, the one or more connected nodes include direct nodes that have been identified as potentially relevant based on previous data exchanges. Moreover, the task information message includes a task id of an origin node providing an identifier for an original task, an input token and a counter for the original local memory state. Additionally, the input token includes an original local memory state resulting from the original task. In an implementation, the task ID of the origin node provides an identifier for the task being executed by the origin node, allowing other nodes to understand the specific task context while the input token includes the original local memory state that is generated by processing the task at the origin node. Moreover, the memory state encapsulates the processed data that may be useful to the one or more nodes. In another implementation, the counter for the original local memory state is configured to track the relevance and freshness of the original memory state, helping receiving nodes assess the timeliness of the data. As a result, the method 200 is used to allow the node to communicate specific task requirement and share relevant information with the one or more connected nodes for facilitating effective collaboration and clustering within the multi-agent intelligent network 100.

[0034] Furthermore, at step 204, the method 200 includes receiving a response message from the one or more nodes connected to that node. By receiving the response messages from the one or more nodes that are connected to the node allows the node to ensure an accurate, efficient, and targeted future communications within the multi-agent intelligent network 100. As a result, the method 200 is used to enhance the overall interactions between the one or more nodes based on real-time feedback, thereby reducing unnecessary data exchanges.

[0035] In accordance with an embodiment, the information message further comprises an identifier of that node. The identifier in the information message is that it enhances the overall efficiency and coordination within the multi-agent intelligent network. By incorporating an identifier, each node can easily recognize the source of the data, preventing redundant evaluation of the same information that reduces unnecessary data processing and transmission, optimizing resource usage across the network. The node identifier also allows for more accurate and targeted communication, as nodes can update their forwarding tables based on the sender's ID, ensuring that future interactions are more relevant and efficient.

[0036] In accordance with an embodiment, the information message includes an identifier of the node. The identifier is used to prevent unnecessary re-evaluation of the same data, improving overall efficiency of the multi-agent intelligent network 100. In an implementation, the node generates an information message that includes the identifier along with data, such as the memory states, which is then transmitted to neighbouring nodes. Upon receiving the message, the neighbouring nodes use the identifier to recognize the source and, if the data is deemed useful, update their forwarding tables with the sender’s ID. As a result, the inclusion of the node improves the coordination among the nodes with optimized resource utilization and enhanced scalability for the multi-agent intelligent network 100 that leads to an efficient communication, distributed processing, and privacy in the multi-agent intelligent network 100.

[0037] In accordance with an embodiment, the response message further includes the identifier of the origin node, the identifier of the node from which the response message was received, and the identifier of a task of the origin node. The inclusion of the identifier of the origin node allows the receiving node to associate the response message with the specific source of the memory state in order to ensure that the response is correctly matched to the data. Moreover, the identifier of the node from which the response message is received provides the information about the node that has transmitted the response message, helping the origin node. Additionally, the identifier of the task of the origin node provides a context about the specific task related to the memory state that is evaluated, which helps the receiving node to understand the relevance of the data in relation to the task thereby improving the accuracy of the evaluation process. As a result, by incorporating the identifiers into the response message, the multi-agent intelligent network 100 is configured to enhance the efficiency and effectiveness of the multi-agent intelligent network 100, leading to an informed decision-making and better alignment of tasks and data exchanges. The response message includes an acknowledgement or a non-acknowledgement indicating whether the original local memory state was useful or not to the node for obtaining the output and for the response message received, including an acknowledgement and adding a pair comprising an identifier of the origin node and an identifier of the node from which the response message was received to a node cluster table. When the node receives the response message that includes an acknowledgment, then, in that case, the node adds a pair to the node cluster table, which includes the identifier of the origin node and the identifier of the node that sent the acknowledgment in order to enhance the clustering process by building a record of effective collaborations thereby allowing the node to recognize the one or more nodes that are aligned with the data and tasks. Moreover, such node cluster table is further utilized for optimizing future communications and data exchanges thereby ensuring that the node interacts primarily with the one or more nodes that are useful in order to reduce any unnecessary data transmissions and fostering more effective task execution across the multi-agent intelligent network 100.

[0038] Furthermore, the method 200 includes clustering the nodes based on the cluster table. By clustering nodes based on the cluster table, the method 200 is used to optimize the communication within the multi-agent intelligent network 100 thereby, ensuring that only the most relevant nodes exchange data, thereby reducing unnecessary transmissions. This clustering is dynamic and can change over time as the network evolves and new data becomes available, making the system adaptive to changing conditions.

[0039] In accordance with an embodiment, the method 200 further includes receiving task information message(s) from at least one of the one or more nodes connected to that node, and evaluating the usefulness of received memory states using linear attention, and if the evaluation shows that the memory state was useful, then send a response message including an acknowledgement, and if not useful, then send a response message including a non-acknowledgement. In an implementation, if the evaluation shows that the memory state is useful, then, in that case, the node is configured to generate the response message and send the response message back to the node that includes the acknowledgment (ACK). In such an implementation, the acknowledgment indicates that the shared memory state has provided valuable information that can be further utilized for future communications. In another implementation, if the evaluation shows that the memory state is not useful, then, in that case, the corresponding node is configured to send the response message including the non-acknowledgment (NACK), signalling that the data was not relevant or useful. As a result, evaluating and responding to received memory states is used for maintaining efficient communication within the multi-agent intelligent network 100 and by providing immediate feedback on the usefulness of shared data, the one or more nodes are configured to refine their interactions and focus on exchanging only the most relevant information with the node. Hence, the unnecessary data traffic is reduced while the collaboration between the nodes of the multi-agent intelligent network 100 is enhanced leading to an effective and adaptive clustering of the nodes.

[0040] In accordance with an embodiment, the method 200 further includes receiving a set of local input embeddings (Xn,Xl-X3) as the input information. The node is configured to receive the set of local input embeddings as the input information to initialize the data processing at the local node level. Moreover, the set of local input embeddings, for example, a first local input embedding, a second local input embedding, and a third local input embedding, refers to a set of tokens that represents the input information about the local data input of the one or more nodes within the multi-agent intelligent network 100. As a result, the node is configured to receive the set of local input embeddings as the input information to process and interpret the local data efficiently for further computations and decision-making processes within the multi-agent intelligent network 100.

[0041] Furthermore, the method 200 includes applying positional embedding to the set of local input embeddings (Xn, X1-X3). In other words, the application of the positional embedding to the set of local input embeddings is used to encode the position of each input element in the sequence, allowing the node to distinguish between different positions. In an implementation, the node is configured to generate the positional embeddings, such as by using predefined or learned functions. Thereafter, the generated positional embeddings are added to the local input embeddings. As a result, by incorporating positional embedding, the node is configured to distinguish between different positions of the one or more nodes, thereby allowing the node to handle various types of sequential data effectively, making the multi-agent intelligent network 100 adaptable to different applications, such as text, audio, and video processing.

[0042] Furthermore, the method 200 includes determining a set of keys (Kn) by multiplying the set of local input embeddings (Xn,Xl- X3) with a learned key matrix (WK). In an implementation, the set of local input embeddings, such as the first local input embedding, the second local input embedding, and the third local input embedding, are multiplied with learned key matrix in order to obtain the set of keys, such as a first key, a second key, and a third key. Moreover, the determination of the set of keys involves transforming the set of local input embeddings by using the learned key matrix to provide the set of keys that are further utilized to compute the memory states. Moreover, by using the learned key matrix, the node is configured to provide an efficient and effective representation of the data that can be used for computing the linear attention.

[0043] Furthermore, the method 200 includes determining a set of values (Vn) by multiplying the set of local input embeddings (Xn,Xl- X3) with a learned value matrix (Wv) and a set of queries (Qn) by multiplying the set of local input embeddings (Xn,Xl-X3) with a learned query matrix (WQ). In an example, the set of local input embeddings, such as the first local input embedding, the second local input embedding, and the third local input embedding are multiplied with learned value matrix in order to obtain the set of values, such as a first value, a second value, and a third value. Similarly, the set of local input embeddings, such as the first local input embedding, the second local input embedding, and the third local input embedding, are multiplied with learned query matrix to determine the set of queries that includes a first query, a second query, and a third query. The determination of the set of queries and the set of values are used to improve the efficiency, scalability, and overall performance of the node while minimizing the communication overhead within the multi-agent intelligent network 100. Moreover, the node is characterized in that the node is further configured to determine a set of local memory states based on the set of local input embeddings, a learned key matrix, and a learned value matrix. In an implementation, the set of local memory states refers to input partial tokens that are associated with each node that is derived through a series of matrix multiplications involving the input token and a set of learned matrices, such as the set of values, set of keys, and the set of queries. Moreover, the local memory states, for example, a first local memory state, a second local memory state, and a third local memory state, are configured to encapsulate the information contained in the input tokens in a manner that is suitable for further processing within the node and for sharing the same with other nodes of the multi-agent intelligent network 100. Firstly, the set of input embeddings (or tokens) are received by the node locally. Thereafter, the set of keys, the set of values, and the set of queries are determined in order to further determine the set of local memory states for each node of the multi-agent intelligent network 100, such as based on the set of local embeddings, the set of learned key matrix, the learned value matrix. The set of local memory states allows the node to perform attention distribution in order to allow information sharing within the multi-agent intelligent network 100 locally, thereby maintaining the data privacy within the multi-agent intelligent network 100. As a result, the set of local memory states is determined to provide an efficient local processing, distributed attention, data privacy, and reduced communication overhead in the multi-agent intelligent network 100.

[0044] Moreover, the method 200 further includes determining a set of local memory states (S’n) based on the set of local input embeddings (Xn, X1-X3), a learned key matrix (WK) and a learned value matrix (Wv). By using learned matrices, the node is configured to transform the raw input embeddings into a format that highlights the key features and values in order to integrate and share such information with other nodes in the multi-agent intelligent network 100. By determining the local memory states by the multiplication of the set of input embeddings with the learned key and the learned value matrices, the multi-agent intelligent network 100 is configured to ensure that each node accurately encodes and processes the local data for efficient transformation of the raw data into meaningful representations, facilitating effective distributed processing and information sharing.

[0045] Furthermore, the method 200 includes receiving one or more node memory states (S1) from at least one of the one or more other nodes, the received one or more node memory states forming a set of received node memory states (Sn). By receiving the set of node memory states from the at least one or more other nodes, the node is configured to integrate external information with the local memory state of the node in order to facilitate a distributed linear attention mechanism without sharing raw data.

[0046] Furthermore, the method 200 includes determining a set of combined node memory states (Sn) based on the received set of node memory states (Sn) and the set of local memory states (S’n). In other words, each node of the multi-agent intelligent network 100 determines the local memory states from the set of input embeddings by using learned matrices and transmits the memory states to one or more other nodes. Finally, the received memory states are integrated with the local memory states of the node to further determine the combined memory states that are used to compute the linear attention. The combined memory state is scaled and rotated, which is added to the first memory state, for example, at operation in order to obtain a combined memory state. Furthermore, the combined memory state is scaled and rotated and is further add to the second memory state to obtain a combined memory state, for example, at operation. Thereafter, the combined memory state is scaled and rotated to further add with the third memory state, for example, to obtain a combined memory state. Hence, the combined memory state includes information from the combined memory state. As a result, the set of combined node memory states is used to facilitate distributed attention in order to ensure that each node has access to a comprehensive set of node combined memory states with an efficient and effective data integration and processing with reduced communication overhead, such as by transmitting aggregated states rather than raw data. Additionally, the determination of the set of node combined memory states is used to maintain data privacy, such as by sharing the combined node memory state with an improved overall performance of the multiagent intelligent network 100.

[0047] Furthermore, the method 200 includes determining the usefulness of a received memory state based on a product of the received memory state (S1) and the query (Qk) and a product of the query and a determined combined memory states (Sn). The product of the received memory state and the query allows the node to compare both the received memory state and the combined memory states from prior exchanges, which is used to ensure that only the most relevant data is incorporated into the decisionmaking process of the node. As a result, the determination of the usefulness of the received memory state based on the product of the received memory state and the query is used to reduce the computational load by focusing only on relevant data, allowing nodes to process large amounts of information more efficiently. In addition, such determination also minimizes the communication overhead by limiting unnecessary exchanges of unhelpful data, conserving network bandwidth and improving the execution of the data transmission within the multi-agent intelligent network 100.

[0048] In accordance with an embodiment, the method 200 further includes determining that the received memory state is useful if the product of the received memory state (S1) and the query (Qk) minus the product of the query and the determined combined memory states (Sn), divided by the product of the query and the determined combined memory states (Sn) exceeds a threshold value (C). By using the threshold value, the node is configured to filter out the memory states that do not contribute to the improvement of the task performance. Moreover, such calculation is used to ensure that the node incorporates the data that adds meaningful value with reduced redundancy. As a result, the likelihood of unnecessary data processing and data transmission is reduced thereby improving the overall efficiency of the multi-agent intelligent network 100. Additionally, the performance of the multi-agent intelligent network 100 is enhanced with reduced computational and communication overhead while maintaining high task accuracy.

[0049] In accordance with an embodiment, the method 200 further includes providing the set of combined node memory states (Sn) to at least one next node and determine an output (On) based on the determined set of queries (Qn) and the set of combined node memory states (Sn). After combining the memory states from the local data and the data received from the one or more nodes, the corresponding node shares the set of combined memory states with other one or more connected nodes. Thereafter, the at least one next node is configured to further utilize the received combined memory states to determine the output. Therefore, by transmitting the set of combined memory states to the at least one next node, each node contributes to a comprehensive understanding of the task at hand. The next node can then determine the output based on the relevant queries and the combined memory states, thereby improving the overall decision-making capability of the multi-agent intelligent network 100 by pooling the knowledge and resources of multiple nodes.

[0050] In an implementation, the method 200 further comprises that node applying a rotational factor ((<|>). For example, when the task information message is received and the memory state counter is determined to be below the threshold, the node applies a rotational factor to the memory state. This process modifies the memory state by rotating it in a defined way, which could involve adjusting the parameters of the data representation to capture different aspects of the information. After applying the rotational factor, the node updates the task information message with the modified memory state and forwards the modified memory state to the next set of connected nodes. As a result, the memory states are shared across the multi-agent intelligent network 100 in order to enhance decision-making outcomes without increasing the computational load.

[0051] In another implementation, the method 200 includes applying a decay factor to the received memory state. By applying the decay factor, each node can prioritize more relevant and recent information, which is crucial in dynamic environments where the state of the multi-agent intelligent network 100 is constantly changing. The decay factor ensures that outdated or less significant data does not unduly influence the system’s current decisions, thereby enhancing the efficiency and accuracy of data processing within the multi-agent intelligent network 100.

[0052] In accordance with an embodiment, the multi-agent intelligent network 100 has a fixed mesh topology. In an implementation, the fixed mesh topology allows the one or more nodes to be connected to the node, forming a reliable, decentralized system for data sharing . Moreover, the fixed nature of the fixed mesh topology ensures consistent and predictable communication patterns, for managing the exchange of task information messages and coordinating tasks in distributed multi-agent environments.

[0053] Advantageously, the method 200 is used for clustering nodes in the multi-agent intelligent network 100 in order to enhance the multi-agent intelligent network 100 efficiently with an improved communication performance. By allowing the nodes to share task-specific data and memory states without the need for centralized control or raw data sharing, the method 200 is used to allow distributed processing and collaboration between relevant nodes. The use of task identifiers, memory state counters, and clustering tables ensures that nodes communicate only with the most relevant peers, reducing unnecessary data transmissions and optimizing resource utilization. Additionally, the application of the decay factor to outdated data and using positional embeddings further refine the ability of the multi-agent intelligent network 100 to prioritize recent and meaningful information. As a result, the clustering of the nodes along with linear attention and learned matrices, enhances the capability of the multiagent intelligent network 100 to handle large-scale, complex environments while maintaining data privacy, reducing computational overhead, and improving decision-making accuracy.

[0054] The steps 202 to 204 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.

[0055] There is further provided a computer program product comprising program instructions for performing the method 200 when executed by one or more processors in the multi-agent intelligent network 100. The computer program product is implemented as an algorithm, embedded in a software stored in a non-transitory computer-readable storage medium. The non-transitory computer-readable storage means may include, but are not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. Examples of implementation of computer-readable storage medium, 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. FIG. 3 is a diagram that depicts a node configured to be used in a multi-agent intelligent network, in accordance with an embodiment of the present disclosure. With reference to FIG. 3, there is shown a node 302 that includes a controller 304, a memory 306, and a network interface 308.

[0056] The node 302 refers to an individual device within the multi-agent intelligent network 100 that processes local data and communicates with the one or more nodes, such as user equipment (UEs) and the like of the multi-agent intelligent network 100.

[0057] The controller 304 may include suitable logic, circuitry, interfaces, hardware, software, or code that is configured to provide a signaling protocol for providing distributed attention in the multi-agent intelligent network 100. Examples of implementation of the controller 304 may include but are not limited to a central data processing device, a microprocessor, a microcontroller, a complex instruction set computing (CISC) processor, an application-specific integrated circuit (ASIC) processor, a reduced instruction set (RISC) processor, a very long instruction word (VLIW) processor, a state machine, and other processors or control circuitry.

[0058] The memory 306 may include suitable logic, circuitry, interfaces, or code that is configured to store data for generating a relevant result. In an implementation, the memory 306 corresponds to a local memory, such as an Electrically Erasable Programmable Read-Only Memory (EEPROM), Random Access Memory (RAM), Read-Only Memory (ROM), a central processing unit (CPU) cache memory, and the like. In another implementation, the memory 306 corresponds to disc storage memory, such as a Hard Disk Drive (HDD), Flash memory, Solid-State Drive (SSD), and the like.

[0059] The network interface 308 includes hardware or software that is configured to establish communication among the controller 304 and the memory 306. Examples of the network interface 308 may include but are not limited to, a computer port, a network socket, a network interface controller (NIC), and any other network interface device.

[0060] There is provided the node 302 configured to be used in the multi-agent intelligent network 100. Moreover, the node 302 is arranged to receive input tokens and to process the input tokens to obtain an output of the node related to a task. The node 302 configured to enhance the efficiency and performance of the multi-agent intelligent network 100 by enabling nodes to identify and collaborate with other relevant nodes without the need for centralized control or raw data sharing.

[0061] In operation, the node 302 is configured to send a task information message to one or more nodes connected to the node 302. In an implementation, the node 302 is configured to generate the task information message that includes information, such as ID, specific task ID, and any relevant metadata or requirements associated with the task. Thereafter, the node 302 is configured to send the generated task information to the one or more nodes that are connected with the node. Moreover, the one or more connected nodes include direct nodes that have been identified as potentially relevant based on previous data exchanges. Moreover, the task information message includes a task id of an origin node providing an identifier for an original task, an input token and a counter for the original local memory state. Additionally, the input token includes an original local memory state resulting from the original task. In an implementation, the task ID of the origin node provides an identifier for the task being executed by the origin node, allowing other nodes to understand the specific task context while the input token includes the original local memory state that is generated by processing the task at the origin node. Moreover, the memory state encapsulates the processed data that may be useful to the one or more nodes. In another implementation, the counter for the original local memory state is configured to track the relevance and freshness of the original memory state, helping receiving nodes assess the timeliness of the data. As a result, the node 302 is configured to allow the node to communicate specific task requirement and share relevant information with the one or more connected nodes for facilitating effective collaboration and clustering within the multi-agent intelligent network 100. Furthermore, the node 302 is configured to receive a response message from the one or more nodes connected to that node 302. By receiving the response messages from the one or more nodes that are connected to the node allows the node to ensure an accurate, efficient, and targeted future communications within the multi-agent intelligent network 100. As a result, the method 200 is used to enhance the overall interactions between the one or more nodes based on real-time feedback, thereby reducing unnecessary data exchanges.

[0062] The response message includes an acknowledgement or a non-acknowledgement indicating whether the original local memory state was useful or not to the node for obtaining the output and for the response message received, including an acknowledgement and adding a pair comprising an identifier of the origin node and an identifier of the node from which the response message was received to a node cluster table. When the node receives the response message that includes an acknowledgment, then, in that case, the node adds a pair to the node cluster table, which includes the identifier of the origin node and the identifier of the node that sent the acknowledgment in order to enhance the clustering process by building a record of effective collaborations thereby allowing the node to recognize the one or more nodes that are aligned with the data and tasks. Moreover, such node cluster table is further utilized for optimizing future communications and data exchanges thereby ensuring that the node interacts primarily with the one or more nodes that are useful in order to reduce any unnecessary data transmissions and fostering more effective task execution across the multi-agent intelligent network 100.

[0063] Furthermore, the node 302 is configured to cluster nodes based on the cluster table. By clustering nodes based on the cluster table, the method 200 is used to optimize the communication within the multi-agent intelligent network 100 thereby, ensuring that only the most relevant nodes exchange data, thereby reducing unnecessary transmissions. This clustering is dynamic and can change over time as the network evolves and new data becomes available, making the system adaptive to changing conditions.

[0064] In accordance with an embodiment, the node 302 is further configured to receive task information message(s) from at least one of the one or more nodes connected to that node and evaluate the usefulness of received memory states using linear attention, and if the evaluation shows that the memory state was useful, then send a response message including an acknowledgement, and if not useful, then send a response message including a non-acknowledgement. In an implementation, if the evaluation shows that the memory state is useful, then, in that case, the node is configured to generate the response message and send the response message back to the node that includes the acknowledgment (ACK). In such an implementation, the acknowledgment indicates that the shared memory state has provided valuable information that can be further utilized for future communications. In another implementation, if the evaluation shows that the memory state is not useful, then, in that case, the corresponding node is configured to send the response message including the non-acknowledgment (NACK), signaling that the data was not relevant or useful. As a result, evaluating and responding to received memory states is used for maintaining efficient communication within the multi-agent intelligent network 100 and by providing immediate feedback on the usefulness of shared data, the one or more nodes are configured to refine their interactions and focus on exchanging only the most relevant information with the node. Hence, the unnecessary data traffic is reduced while the collaboration between the nodes of the multi-agent intelligent network 100 is enhanced leading to an effective and adaptive clustering of the nodes.

[0065] In accordance with an embodiment, the node 302 is further configured to receive a set of local input embeddings (Xn,Xl-X3) as the input information. The node 302 is configured to receive the set of local input embeddings as the input information to initialize the data processing at the local node level. Moreover, the set of local input embeddings, for example, a first local input embedding, a second local input embedding, and a third local input embedding, refers to a set of tokens that represents the input information about the local data input of the one or more nodes within the multi-agent intelligent network 100. As a result, the node is configured to receive the set of local input embeddings as the input information to process and interpret the local data efficiently for further computations and decision-making processes within the multi-agent intelligent network 100. Furthermore, the node 302 is further configured to apply positional embedding to the set of local input embeddings (Xn, XI- X3). In other words, the application of the positional embedding to the set of local input embeddings is used to encode the position of each input element in the sequence, allowing the node to distinguish between different positions. In an implementation, the node is configured to generate the positional embeddings, such as by using predefined or learned functions. Thereafter, the generated positional embeddings are added to the local input embeddings. As a result, by incorporating positional embedding, the node is configured to distinguish between different positions of the one or more nodes, thereby allowing the node to handle various types of sequential data effectively, making the multi-agent intelligent network 100 adaptable to different applications, such as text, audio, and video processing.

[0066] Furthermore, the node 302 is further configured to determine a set of keys (Kn) by multiplying the set of local input embeddings (Xn,Xl-X3) with a learned key matrix (WK). In an implementation, the set of local input embeddings, such as the first local input embedding, the second local input embedding, and the third local input embedding, are multiplied with learned key matrix in order to obtain the set of keys, such as a first key, a second key, and a third key. Moreover, the determination of the set of keys involves transforming the set of local input embeddings by using the learned key matrix to provide the set of keys that are further utilized to compute the memory states. Moreover, by using the learned key matrix, the node is configured to provide an efficient and effective representation of the data that can be used for computing the linear attention.

[0067] Furthermore, the node 302 is further configured to determine a set of values (Vn) by multiplying the set of local input embeddings (Xn,Xl-X3) with a learned value matrix (Wv) and determine a set of queries (Qn) by multiplying the set of local input embeddings (Xn,Xl-X3) with a learned query matrix (WQ). In an example, the set of local input embeddings, such as the first local input embedding, the second local input embedding, and the third local input embedding are multiplied with learned value matrix in order to obtain the set of values, such as a first value, a second value, and a third value. Similarly, the set of local input embeddings, such as the first local input embedding, the second local input embedding, and the third local input embedding, are multiplied with learned query matrix to determine the set of queries that includes a first query, a second query, and a third query. The determination of the set of queries and the set of values are used to improve the efficiency, scalability, and overall performance of the node while minimizing the communication overhead within the multi-agent intelligent network 100. Moreover, the node is characterized in that the node is further configured to determine a set of local memory states based on the set of local input embeddings, a learned key matrix, and a learned value matrix. In an implementation, the set of local memory states refers to input partial tokens that are associated with each node that is derived through a series of matrix multiplications involving the input token and a set of learned matrices, such as the set of values, set of keys, and the set of queries. Moreover, the local memory states, for example, a first local memory state, a second local memory state, and a third local memory state, are configured to encapsulate the information contained in the input tokens in a manner that is suitable for further processing within the node and for sharing the same with other nodes of the multi-agent intelligent network 100. Firstly, the set of input embeddings (or tokens) are received by the node locally. Thereafter, the set of keys, the set of values, and the set of queries are determined in order to further determine the set of local memory states for each node of the multi-agent intelligent network 100, such as based on the set of local embeddings, the set of learned key matrix, the learned value matrix. The set of local memory states allows the node to perform attention distribution in order to allow information sharing within the multi-agent intelligent network 100 locally, thereby maintaining the data privacy within the multi-agent intelligent network 100. As a result, the set of local memory states is determined to provide an efficient local processing, distributed attention, data privacy, and reduced communication overhead in the multi-agent intelligent network 100.

[0068] Furthermore, the node 302 is further configured to determine a set of local memory states (S’n) based on the set of local input embeddings (Xn, X1-X3), a learned key matrix (WK) and a learned value matrix (Wv). By using learned matrices, the node is configured to transform the raw input embeddings into a format that highlights the key features and values in order to integrate and share such information with other nodes in the multi-agent intelligent network 100. By determining the local memory states by the multiplication of the set of input embeddings with the learned key and the learned value matrices, the multi-agent intelligent network 100 is configured to ensure that each node accurately encodes and processes the local data for efficient transformation of the raw data into meaningful representations, facilitating effective distributed processing and information sharing.

[0069] Furthermore, the node 302 is further configured to receive one or more node memory states (S1) from at least one of the one or more other nodes the received one or more node memory states forming a set of received node memory states (Sn). By receiving the set of node memory states from the at least one or more other nodes, the node is configured to integrate external information with the local memory state of the node in order to facilitate a distributed linear attention mechanism without sharing raw data.

[0070] Furthermore, the node 302 is further configured to determine a set of combined node memory states (Sn) based on the received set of node memory states (Sn) and the set of local memory states (S’n). In other words, each node of the multi-agent intelligent network 100 determines the local memory states from the set of input embeddings by using learned matrices and transmits the memory states to one or more other nodes. Finally, the received memory states are integrated with the local memory states of the node to further determine the combined memory states that are used to compute the linear attention. The combined memory state is scaled and rotated, which is added to the first memory state, for example, at operation in order to obtain a combined memory state. Furthermore, the combined memory state is scaled and rotated and is further add to the second memory state to obtain a combined memory state, for example, at operation. Thereafter, the combined memory state is scaled and rotated to further add with the third memory state, for example, to obtain a combined memory state. Hence, the combined memory state includes information from the combined memory state. As a result, the set of combined node memory states is used to facilitate distributed attention in order to ensure that each node has access to a comprehensive set of node combined memory states with an efficient and effective data integration and processing with reduced communication overhead, such as by transmitting aggregated states rather than raw data. Additionally, the determination of the set of node combined memory states is used to maintain data privacy, such as by sharing the combined node memory state with an improved overall performance of the multiagent intelligent network 100.

[0071] Furthermore, the node 302 is further configured to determining the usefulness of a received memory state based on a product of the received memory state (S1) and the query (Qk) and a product of the query and a determined combined memory states (Sn). The product of the received memory state and the query allows the node to compare both the received memory state and the combined memory states from prior exchanges, which is used to ensure that only the most relevant data is incorporated into the decision-making process of the node. As a result, the determination of the usefulness of the received memory state based on the product of the received memory state and the query is used to reduce the computational load by focusing only on relevant data, allowing nodes to process large amounts of information more efficiently. In addition, such determination also minimizes the communication overhead by limiting unnecessary exchanges of unhelpful data, conserving network bandwidth and improving the execution of the data transmission within the multi-agent intelligent network 100.

[0072] In accordance with an embodiment, the node is further configured to determine that the received memory state is useful if the product of the received memory state (S1) and the query (Qk) minus the product of the query and the determined combined memory states (Sn), divided by the product of the query and the determined combined memory states (Sn) exceeds a threshold value (C). By using the threshold value, the node is configured to filter out the memory states that do not contribute to the improvement of the task performance. Moreover, such calculation is used to ensure that the node incorporates the data that adds meaningful value with reduced redundancy. As a result, the likelihood of unnecessary data processing and data transmission is reduced thereby improving the overall efficiency of the multi-agent intelligent network 100. Additionally, the performance of the multi-agent intelligent network 100 is enhanced with reduced computational and communication overhead while maintaining high task accuracy. In accordance with an embodiment, the node is further configured to provide the set of combined node memory states (Sn) to at least one next node and determine an output (On) based on the determined set of queries (Qn) and the set of combined node memory states (Sn). After combining the memory states from the local data and the data received from the one or more nodes, the corresponding node shares the set of combined memory states with other one or more connected nodes. Thereafter, the at least one next node is configured to further utilize the received combined memory states to determine the output. Therefore, by transmitting the set of combined memory states to the at least one next node, each node contributes to a comprehensive understanding of the task at hand. The next node can then determine the output based on the relevant queries and the combined memory states, thereby improving the overall decision-making capability of the multi-agent intelligent network 100 by pooling the knowledge and resources of multiple nodes.

[0073] In accordance with an embodiment, the node is further configured to apply a decay factor to the received memory state. For example, when the task information message is received and the memory state counter is determined to be below the threshold, the node applies a rotational factor to the memory state. This process modifies the memory state by rotating it in a defined way, which could involve adjusting the parameters of the data representation to capture different aspects of the information. After applying the rotational factor, the node updates the task information message with the modified memory state and forwards the modified memory state to the next set of connected nodes. As a result, the memory states are shared across the multi-agent intelligent network 100 in order to enhance decision-making outcomes without increasing the computational load.

[0074] In accordance with an embodiment, the node is further configured to apply a rotational factor. By applying the decay factor, each node can prioritize more relevant and recent information, which is crucial in dynamic environments where the state of the multi-agent intelligent network 100 is constantly changing. The decay factor ensures that outdated or less significant data does not unduly influence the system’s current decisions, thereby enhancing the efficiency and accuracy of data processing within the multi-agent intelligent network 100.

[0075] Advantageously, the node 302 is configured to cluster nodes in the multi-agent intelligent network 100 in order to enhance the multi-agent intelligent network 100 efficiently with an improved communication performance. By allowing the nodes to share task-specific data and memory states without the need for centralized control or raw data sharing, the method 200 is used to allow distributed processing and collaboration between relevant nodes. The use of task identifiers, memory state counters, and clustering tables ensures that nodes communicate only with the most relevant peers, reducing unnecessary data transmissions and optimizing resource utilization. Additionally, the application of the decay factor to outdated data and using positional embeddings further refine the ability of the multi-agent intelligent network 100 to prioritize recent and meaningful information. As a result, the clustering of the nodes along with linear attention and learned matrices, enhances the capability of the multiagent intelligent network 100 to handle large-scale, complex environments while maintaining data privacy, reducing computational overhead, and improving decision-making accuracy.

[0076] FIG. 4 is a diagram that illustrates an exemplary scenario depicting each node in a mesh topology of distributed nodes to discover groups of other nodes to collaborate with, based on the mutual usefulness of local data, in accordance with an embodiment of the present disclosure. FIG. 4 is described in conjunction with elements from FIG. 1 to 3. With reference to FIG. 4, there is shown a diagram 400 that illustrates an exemplary scenario for clustering of the nodes in the multi-agent intelligent network 100.

[0077] In an exemplary scenario, all the nodes, such as a first node 402A, a second node 402 B, a third node 402C, a fourth node 402D, and a fifth node 402E are equipped with linear attention modules with parameters WQ, WK, Wv (learned matrices of dimension d x d), and a decay coefficient y (a fixed vector of dimension d), and a phase shift <p (a fixed vector of dimension d). The parameters are the same for all nodes. Furthermore, each node has access to their own input data. Moreover, all the nodes use local data to produce memory states Snand progressively exchange the states with the neighbour nodes for data usefulness evaluation. After each exchange, each node updates a forwarding table (or a cluster table) which lists all nodes having useful data for the node (e.g., the third node 406), and all the nodes who need data from other nodes connected to node. In an implementation, each node uses its linear attention module to process its input data and obtain memory states to initialize an empty forwarding table, and set the maximum number of rounds N_max. Thereafter, the node is configured to share its local information with neighbour nodes and each node sends the sender’s ID (here, ID ( sender )=ID(k) ), the ID of the node producing the original memory states (here ID(origin)=ID(k) ), local memory states of ID(origin) (here, a memory state of the corresponding node). Furthermore, from each of the neighbour node, the node is configured to receive message with the sender’s ID, the ID of the node producing the original memory states (here ID (origin)=ID(k)), ACK (the transmitted memory states were useful) or NACK (the states were not useful). Whenever ACK is received from some node, the corresponding node adds a pair (ID (origin), ID(n)) to its cluster table and the node receives neighbours’ local information for usefulness evaluation. From each neighbour, the node receives a message of type A with ID (sender), ID (origin) and state and further evaluates the usefulness of received memory states using linear attention. Thereafter, the node responds to each neighbour with a message of type B with ID(k), ID (origin), ACK / NACK. If states received from some node n were useful, node k adds a pair (ID (origin), ID(k)) to the cluster table. The node is configured to share the memory states received in a previous round with selected neighbour nodes and the node sends a message of type A with ID(k), ID(origin), and the states to selected neighbour nodes. Furthermore, each node receives a message of type B with ACK / NACK if the states were useful or not. If ACK is received from some node n, node k adds a pair (ID (origin), ID(n)) to its forwarding table (i.e., cluster table). Moreover, if the cluster table of the node has no pair (ID (origin), ID(k)), i.e., the data of node origin are useful for node n and not for the node, then the node will act just as a relay from node origin to node n. In that case, the node is configured to forward recursively the message with ACK until ID (origin) is reached. Thereafter, the node is configured to receive the forwarded states from neighbour nodes for joint evaluation from selected neighbours, node k receives a message of type A with ID (sender), ID (origin) and states, evaluate the usefulness of received memory states using linear attention, and responds to the senders with a message of type B and ACK / NACK. Moreover, if a state received from some node n was useful, node k adds a pair (ID (origin), ID) to the cluster table. As a result, by using linear attention modules with shared parameters across nodes is that it enables efficient and scalable memory state sharing and evaluation, minimizing computational load while enhancing coordination and data usefulness across the multi-agent intelligent network 100 that optimizes the resource utilization by reducing redundancy in communication and updating forwarding tables based on relevant, useful data exchanges.

[0078] FIG. 5 is a diagram that illustrate an exemplary scenario for multi-round message exchanges between the nodes, in accordance with an embodiment of the present disclosure. FIG. 5 is described in conjunction with elements from FIG. 1 to 4. With reference to FIG. 5, there is shown a diagram 500 that illustrates an exemplary scenario for multi-round message exchanges between the nodes.

[0079] In an implementation scenario, the FIG. 5 represents a first node 502 and a second node 504 configured to exchange the local memory state with other nodes, which is a first exchange 516 of the local memory. At operation 506, the second node 504 is configured to receive a message with ID(sender), ID(origin) and memory states. Moreover, the second node 504 is configured to evaluate the usefulness of received memory states using linear attention and further send the ACK / NACK if the memory states were useful or not to the first node 502, such as at operation 508 (i.e., at round 1). Similarly, the second node 504 is configured to evaluate the usefulness of received memory states using linear attention and further send the ACK / NACK if the memory states were useful or not to the first node 502, such as at operation 512 (i.e., at round k). As a result, the exchange of local memory state is used to perform cluetering of the nodes more efficiently and accurately.

[0080] FIG. 6 is a diagram that illustrates an exemplary scenario for evaluating usefulness of the received state, in accordance with an embodiment of the present disclosure. FIG. 6 is described in conjunction with elements from FIG. 1 to 5. With reference to FIG. 6, there is shown a diagram 600 that depicts the evaluation of the usefulness of the received state by the first node 502. In an implementation scenario, the first node 502 is configured to receive the memory state (i.e., S(rec)) for usefulness evaluation. Moreover, the first node 502 includes a linear attention module with parameters, such as learned query matrix, learned key matrix, learned value matrix, decay factor, rotating factor, and the like. Firstly, the first node 502 is configured to computes queries, keys, values, and a memory' state from the local data as given in the below-mentioned equation (4), such as at operation 602, through input tokens 604: -

[0081] QJ= WQXJ KJ = WKXJ VJ = WVXJ S(J) = / C / VJ (4)

[0082] Thereafter, the first node 502 is configured to combine the local memory state with states received in previous rounds linearly as given in the below-mentioned equation (5): -

[0083] S= S(K) + S7(ei<p)CTR® O S(j) (5)

[0084] Moreover, the combined memory state represents all the information that is acquired by the first node 502. In the above- mentioned equation (5), the CTR(j) is a counter of the received memory state, which considers the decaying relevance of the individual memory states from the one or more nodes. Finally, the corresponding node is configured to compute the relative information contribution of the new received state by the below mentioned equation (6): -

[0085] C=||QkS(rec)-QkS||\||QkS|| (6)

[0086] Moreover, if the contribution is larger than the threshold value, then, the new memory state (i.e., S(rec)) is added to the relevant information. Alternately, the memory state (i.e., S(rec)) is considered as not relevant. Hence, the first node 502 is configured to cluster the one or more nodes of the multi-agent intelligent network 100.

[0087] 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

CLAIMS1. A method (200) for clustering nodes of a multi-agent intelligent network (100), wherein each node is arranged to receive input tokens and to process the input tokens to obtain an output of the node related to a task, wherein the method (200) comprises: each node sending a task information message to one or more nodes connected to that node, wherein the task information message comprises: a task id of an origin node providing an identifier for an original task, an input token and a counter for the original local memory state, wherein the input token includes an original local memory state resulting from the original task, and each node receiving a response message from the one or more nodes connected to that node, wherein the response message comprises: an acknowledgement or a non-acknowledgement indicating whether the original local memory state was useful or not to the node for obtaining the output, and wherein the method (200) comprises that node for each the response message received and including an acknowledgement, adding a pair comprising an identifier of the origin node and an identifier of the node from which the response message was received to a node cluster table, and wherein the method (200) comprises: clustering nodes based on the cluster table.

2. The method (200) according to claim 1 , wherein the method (200) farther comprises: each node receiving task information message(s) from at least one of the one or more nodes connected to that node, and evaluating the usefulness of received memory states using linear attention, and if the evaluation shows that the memory state was useful, then send a response message including an acknowledgement, and if not useful, then send a response message including a non-acknowledgement.

3. The method (200) according to claim 2, wherein the method (200) farther comprises each node: receiving a set of local input embeddings (Xn,Xl-X3) as the input information, applying positional embedding to the set of local input embeddings (Xn, X1-X3), determining a set of keys (Kn) by multiplying the set of local input embeddings (Xn,Xl-X3) with a learned key matrix (WK), determining a set of values (Vn) by multiplying the set of local input embeddings (Xn,Xl-X3) with a learned value matrix (WV), determining a set of queries (Qn) by multiplying the set of local input embeddings (Xn,Xl-X3) with a learned query matrix (WQ), wherein the method (200) further comprises the node determining a set of local memory states (S’n) based on the set of local input embeddings (Xn, X1-X3), a learned key matrix (WK) and a learned value matrix (WV), receiving one or more node memory states (Si) from at least one of the one or more other nodes (L, M), the received one or more node memory states forming a set of received node memory states (Sn), determining a set of combined node memory states (Sn) based on the received set of node memory states (Sn) and the set of local memory states (S’n), and determining the usefulness of a received memory state based on a product of the received memory state (Si) and the query (Qk) and a product of the query and a determined combined memory states (Sn).

4. The method (200) according to claim 3, wherein the method (200) further comprises that node determining that the received memory state is useful if the product of the received memory state (Si) and the query (Qk) minus the product of thequery and the determined combined memory states (Sn), divided by the product of the query and the determined combined memory states (Sn) exceeds a threshold value (C).

5. The method (200) according to claim 4, wherein the method (200) further comprises that node providing the set of combined node memory states (Sn) to at least one next node and determine an output (On) based on the determined set of queries (Qn) and the set of combined node memory states (Sn).

6. The method (200) according to any preceding claim, wherein the method (200) further comprises that node applying a decay factor to the received memory state.

7. The method (200) according to claim 6, wherein the method (200) further comprises that node applying a rotational factor.

8. The method (200) according to any preceding claim, wherein the information message further comprises an identifier of that node.

9. The method (200) according to any preceding claim, wherein the response message further includes: the identifier of the origin node, the identifier of the node from which the response message was received, and the identifier of a task of the origin node.

10. The method (200) according to any preceding claim, wherein the multi-agent intelligent network (100) has a fixed mesh topology.

11. A computer program product comprising program instructions for performing the method (200) according to any preceding claim, when executed by one or more processors in a multi-agent intelligent network (100).

12. A node (302) configured to be used in a multi-agent intelligent network (100), wherein the node (302) is arranged to receive input tokens and to process the input tokens to obtain an output of the node (302) related to a task, wherein the node (302) is configured to: send a task information message to one or more nodes connected to the node, wherein the task information message comprises: a task id of an origin node providing an identifier for an original task, an input token and a counter for the original local memory state, wherein the input token includes an original local memory state resulting from the original task, and to receive a response message from the one or more nodes connected to that node, wherein the response message comprises: an acknowledgement or a non-acknowledgement indicating whether the original local memory state was useful or not to the node (302) for obtaining the output, and wherein the node (302) is further configured to, for each the response message received and including an acknowledgement, add a pair comprising an identifier of the origin node and an identifier of the node (302) from which the response message was received to a node cluster table, and wherein the node (302) is configured to: cluster nodes based on the cluster table.

13. The node (302) according to claim 12, wherein the node (302) is further configured to receive task information message(s) from at least one of the one or more nodes connected to that node (302) and evaluate the usefulness of receivedmemory states using linear attention, and if the evaluation shows that the memory state was useful, then send a response message including an acknowledgement, and if not useful, then send a response message including a non-acknowledgement.

14. The node (302) according to claim 13, wherein the node (302) is further configured to: receive a set of local input embeddings (Xn,Xl-X3) as the input information, apply positional embedding to the set of local input embeddings (Xn, X1-X3), determine a set of keys (Kn) by multiplying the set of local input embeddings (Xn,Xl-X3) with a learned key matrix (WK), determine a set of values (Vn) by multiplying the set of local input embeddings (Xn,Xl-X3) with a learned value matrix (WV), determine a set of queries (Qn) by multiplying the set of local input embeddings (Xn,Xl-X3) with a learned query matrix (WQ), determine a set of local memory states (S’n) based on the set of local input embeddings (Xn, X1-X3), a learned key matrix (WK) and a learned value matrix (WV), receive one or more node memory states (Si) from at least one of the one or more other nodes the received one or more node memory states forming a set of received node memory states (Sn), determining a set of combined node memory states (Sn) based on the received set of node memory states (Sn) and the set of local memory states (S’n), and determining the usefulness of a received memory state based on a product of the received memory state (Si) and the query (Qk) and a product of the query and a determined combined memory states (Sn).

15. The node (302) according to claim 14, wherein the node (302) is further configured to determine that the received memory state is useful if the product of the received memory state (Si) and the query (Qk) minus the product of the query and the determined combined memory states (Sn), divided by the product of the query and the determined combined memory states (Sn) exceeds a threshold value (C).

16. The node (302) according to claim 15, wherein the node (302) is further configured to provide the set of combined node memory states (Sn) to at least one next node and determine an output (On) based on the determined set of queries (Qn) and the set of combined node memory states (Sn).

17. The node (302) according to claim 16, wherein the node (302) is further configured to apply a decay factor to the received memory state.

18. The node (302) according to claim 17, wherein the node (302) is further configured to apply a rotational factor.19

Images