Node and method for processing data in node of multi-agent intelligent network

Abstract

A method for processing data in node of multi-agent intelligent network. Furthermore, node is arranged to receive input token and to execute task on input token to provide output. Moreover, method comprises node determine local state of node based on input token determine task identifier and reset counter, send state message to direct neighbours according to forwarding table. Furthermore, forwarding table indicates which nodes is to transmit messages to, and state message includes identifier, task and local state. The method further comprises node receive state message from another node and determine if counter in received state message is less than maximum counter value, and then decay state in the received state message, increase counter in received state message, and forward decay of state in received state message to target node based on identifier of received state message, and provide output as product of query and local state.

Description

[0001] NODE AND METHOD FOR PROCESSING DATA IN NODE 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 node and a method for processing data in a node of a multi-agent intelligent network, such as by signalling for inference over a mesh network using linear attention.

[0004] BACKGROUND

[0005] In rapidly advancing domain of communication systems, upcoming future communication systems are anticipated to rely heavily on the cooperation of distributed agents that are used to combine the processing compatibility of each of the intelligent agents and produce results. Moreover, each of the distributed agents is used to jointly process multimodal data (e.g., images, videos or sensing data collected by each of the distributed agents) to adapt to 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. In addition, each of the distributed agents also fails to provide an efficient knowledge sharing among agents, and less around distributed attention. Furthermore, each of the distributed agent is not capable enough to determine the data useful for other nodes and therefore, transmit all input data to all the nodes for joint evaluation, which is often infeasible due to communication and privacy constraints.

[0006] Conventional communication systems, such as conventional self-attention mechanisms, 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 the 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 the unavailability of the relevant data in a single place, complex computational costs, restrictive data flow, and the like. Therefore, there exists a technical problem of how to share data among the distributed agents across a distributed agent network.

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

[0008] SUMMARY

[0009] The present disclosure provides a node of a multi-agent intelligent network and a method for processing data in the node of the multi-agent intelligent network. The present disclosure provides a solution to the existing problem of how to share data among the distributed agents across a distributed agent network. 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 node of the multi-agent intelligent network and the method for processing data in the node of the multi-agent intelligent network.

[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.

[0011] In one aspect, the present disclosure provides a method for processing data in a node of the multi-agent intelligent network. The node is arranged to receive an input token and to execute a task on the input token to provide an output. The method comprises the node to determine a local state of the node based on the input token, determine a task identifier and reset a counter, and send a state message to direct neighbours according to a forwarding table. Moreover, the forwarding table indicates which nodes in the multi-agent intelligent network the node is to transmit messages to, and the state message includes the identifier, the task and the local state. The method further comprises the node to receive a state message from another node and determine if a counter in the received state message is less than a maximum counter value, and then decay the state in the received state message, increase the counter in the received state message, and forward the decay the state in the received state message to a target node according to the forwarding table based on the identifier of the received state message, and then provide the output as the product of the query and the local state.

[0012] Advantageously, the method for processing data in the node of the multi-agent intelligent network allows each node within the multi-agent intelligent network to function independently and efficiently within the network. Each node is designed to handle specific tasks without relying on constant instructions from a centralized system or a server. Furthermore, each node in the multi-agent intelligent network is configured to receive the input tokens and process the input tokens locally. Moreover, each node only processes data that is relevant to the tasks assigned to that node, which conserves bandwidth and reduces unnecessary communication within the multi-agent intelligent network. By virtue of processing the input token or input data locally, the node is configured to minimize the need to send raw data across the multi-agent intelligent network, which improves the speed and responsiveness of the data processing while conserving resources like bandwidth. The method enables each node to identify the specific task the node needs to perform by determining a task identifier (ID). Moreover, each node uses counters to track how many times a task or message has been processed or forwarded. Additionally, each node is configured to send messages only to relevant neighbours according to the forwarding table, which lists the specific nodes for further communication, thereby reducing unnecessary network traffic and improving the efficiency of data transmission. Further, the node receiving a state does not only apply a decay, but can also enrich the state with own local states (if this is helpful for the next agent) that is used to combine the knowledge from several nodes in one forwarded state in order to further save bandwidth. Moreover, the updated state forwarded by the node could be either a received state decayed by a constant factor, or a combination of the received state or the local state of the node , which depends on the local forwarding table of the node. Therefore, the method further allows each node to decay the state in the received message, thereby gradually reducing the importance of older data while prioritizing newer, more relevant information, leading to better decision-making.

[0013] In another aspect, the present disclosure provides a node configured to process data in a node of the multi-agent intelligent network. The node is configured to receive input token and to execute a task on the input token to provide an output, wherein the node is configured to determine a local state of the node based on the input token, determine a task identifier and reset a counter, send a state message to direct neighbours according to a forwarding table, wherein the forwarding table indicates which nodes in the multi-agent intelligent network the node is to transmit messages to, and wherein the state message includes the identifier, the task and the local state. The node is further configured to receive a state message from another node and determine if a counter in the received state message is less than a maximum counter value, and then decay the state in the received state message, increase the counter in the received state message, and forward the decay the state in the received state message to a target node according to the forwarding table based on the identifier of the received state message, and then provide the output as the product of the query and the local state.

[0014] The node achieves all the advantages and technical effects of the method of the present disclosure.

[0015] It is to be appreciated that all the aforementioned implementation forms can be combined.

[0016] 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.

[0017] 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.

[0018] BRIEF DESCRIPTION OF THE DRAWINGS

[0019] 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.

[0020] Embodiments of the present disclosure will now be described, by way of example only, with reference to the following diagrams wherein:

[0021] 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;

[0022] FIG. 2 is a flowchart of a method for processing data in a node of the multi-agent intelligent network, in accordance with an embodiment of the present disclosure;

[0023] FIG. 3 is a diagram that depicts a node configured to process data in a node of a multi-agent intelligent network, in accordance with an embodiment of the present disclosure;

[0024] FIG. 4 is a diagram that illustrates an exemplary scenario of a distributed mesh topology, in accordance with an embodiment of the present disclosure; and

[0025] FIG. 5 is a diagram that illustrates an exemplary scenario of knowledge flow between agents in a fixed mesh network, 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.

[0029] 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. 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 processing data in a node of the multi-agent intelligent network, in accordance with an embodiment of the present disclosure. FIG. 2 is described in conjunction with elements from FIG. 1. With reference to FIG. 2, there is shown a flowchart of method 200 that includes steps 202 to 224.

[0032] There is provided the method 200 for processing data in a node of the multi-agent intelligent network 100. Moreover, each node within the multi-agent intelligent network 100 is configured to carry out one or more task assigned to the node without the requirement of any constant direction from a centralized system or a server, as each node has its own computing power and intelligence, allowing each node to perform complex data processing tasks on its own.

[0033] In accordance with an embodiment, the multi-agent intelligent network 100 has a fixed mesh topology. The fixed mesh topology includes multiple nodes, where each node is connected to a fixed number of neighbouring nodes and are configured to use the forwarding table to establish communication with a fixed set of neighbours, ensuring that messages are transmitted efficiently without the need for dynamic adjustments to the connections. Advantageously, the fixed mesh topology enhances the overall stability of the multi-agent intelligent network 100 and reduces the chances of communication errors.

[0034] At step 202, the method 200 includes receiving the input token. Furthermore, the node is arranged to receive an input token (i.e., X). In an implementation, the input token may represent a packet of data or an instruction that needs to be processed by the node, which is configured to serve as the starting point for the data processing operation within the node. Each node in the multi-agent intelligent network 100 is configured to handle certain tasks based on the received input token. Moreover, once the node receives the input token, thereafter, the node is configured to interpret the information, which is included in the input token and further performs the required data processing operations. Therefore, by allowing each node to receive input tokens independently, the method 200 is used to reduce the unnecessary data transfer across the multi-agent intelligent network 100.

[0035] At step 204, the method 200 includes executing the task on the input token to provide output. In an implementation, the executed task may include computation, transformation, or analysis that is required to produce useful results from the received input tokens. Once the node receives the input token, thereafter, the node is configured to apply the pre-defined algorithms, functions, or processing rules to the input token, which may include performing calculations, analyzing patterns, filtering data, or transforming the input in other ways. Moreover, the one or more nodes can process data parallelly threreby enhancing the overall data processing speed and responsiveness of the multi-agent intelligent network 100 while reducing the unnecessary data transmission, and improving communication efficiency of the multi-agent intelligent network 100.

[0036] At step 206, the method 200 includes determining the local state of the node based on the input token. In an implementation, the node is configured to determine its local state (i.e., S) based on the input token, such as by analyzing the data included in the input token, which may involve extracting relevant features, comparing the input with previous states, or applying predefined rules and algorithms to the received input token. Furthermore, based on the analysis, the node is configured to update the internal variables, setting the local state to reflect the current conditions. In an implementation, the local state may include information such as the current task priority, resource availability, or the specific processing mode used by the node. Therefore, the determination of the local state enables the node to allow an improved resource allocation thereby, ensuring that the data processing tasks can be handled in the most efficient manner possible.

[0037] In accordance with an embodiment, the method 200 further includes determining the local state (i.e., S) of the node based on the input tokens (i.e., X) by determining a key (i.e., K) being the product of a key weight (i.e., WK) and the input tokens (X) (K=WKX) and a value (i.e., V) being the product of a value weight (i.e., Wv) and the input token (X) (V=WvX), wherein the local state is determined as the product of a transpose of the key and the value (V) (S = KTV). By applying the key weight, the node is configured to capture the critical features from the input tokens, enhancing the ability of the node to process the received information efficiently. Moreover, the value weight allows the node to interpret the input tokens in a way that aligns with the overall objectives of the task.

[0038] At step 208, the method 200 includes determining a task identifier. Furthermore, the node is configured to determine a task identifier (i.e., ID). By determining the task identifier, the node is configured to identify the specific task that needs to be performed by the node in order to process the input data accurately and efficiently. Moreover, the task identifier allows the node to understand which operation or computation must be applied to the data after being received by the node. In addition, the node is configured to use the task identifier to select the appropriate processing routines or algorithms that match the task in order to differentiate the processing requirementsbetween various tasks, especially when multiple tasks are assigned to different nodes or when each node is configured to handle a variety of tasks. Therefore, by incorporating the task identifiers, the method 200 is used to streamline the operations of the node and ensure that the tasks are distributed and processed accurately across the multi-agent intelligent network 100.

[0039] At step 210, the method 200 includes resetting the counter. Furthermore, the node is configured to reset the counter (i.e., CTR). The counter refers to a numerical value that tracks the number of tasks and the data operations that has been processed or forwarded. Mroeover, by resetting the counter, the node is configured to accurately track the current processing cycle or transmission of data without any confusion from previous tasks. In addition, the node resets the counter by reinitializing the counter variable to a starting value, usually zero, as soon as the counter begins the data processing a new task or data input, which may be automatically triggered by the arrival of a new input token or at the end of a previous task. Advantageously, the counter allows the node to monitor and control how the data moves through the multi-agent intelligent network 100 in a structured manner.

[0040] At step 212, the method 200 includes sending a state message to direct neighbours according to the forwarding table. Furthermore, the node is configured to send the state message to direct neighbours according to the forwarding table. The forwarding table indicates which node in the multi-agent intelligent network 100, the node is configured to transmit messages to. The state message includes the identifier, the task and the local state. Initially, the node is configured to check the forwarding table for the neighbours for sharing data or working together. After checking the forwarding table, the node is configured to create the message, which includes certain details like the task that the node is working on, the task ID, the current state, and the like. The node is further configured to send the message only to the neighbours, that are listed in the forwarding table, making sure that the node shares the data only with the relevant nodes. Thus, by using the forwarding table, the node is configured to avoid broadcasting messages to every other node in the multi-agent intelligent network 100, which conserves bandwidth and reduces unnecessary network traffic.

[0041] At step 214, the method 200 includes receiving the state message from another node. Furthermore, the node is further configured to receive the state message from another node. In an implementation, the node is configured to receive the state message via a communication link that connects the node to other nodes. Moreover, once the state message is received, the node is configured to process the information inside the state message, which includes the task identifier and local state of the sender. In addition, the node may use the processed information and combine the processed information with the local data, and forward the combination of processed information and local data to other nodes if needed.

[0042] At step 216, the method 200 includes determining if the counter in the received state message is less than the maximum counter value. Furthermore, the node is configured to determine if the counter in the received state message is less than the maximum counter value. In an implementation, the node is configured the state message, which includes the counter value that has been incremented each time another node forwards the state message. The node compares the counter value to a preset maximum. Moreover, if the counter is still below the preset maximum, the node can either process the state message or forward the state message to other nodes. If the counter has reached or exceeded the maximum, the node stops further processing or forwarding of the message. Thus, by setting a limit on how many times a message can be forwarded, the method 200 conserves bandwidth and avoids unnecessary data traffic within the multi-agent intelligent network 100.

[0043] At step 218, the method 200 includes decaying the state in the received state message. Furthermore, the node is configured to decay the state in the received state message. In an implementation, the node is configured to apply a mathematical function, such as multiplying the state value by a decay factor, which gradually reduces the value of the state. The decay factor can be a fixed number or may depend on specific conditions such as the number of hops the state message made. After decaying the state, the node can either use the degraded state message for further processing or pass the degraded state message along to neighbouring nodes, where the decay may continue. Thus, by decaying the state in the received state message, the method 200 gradually reduces the influence of older data and prioritizes newer, more accurate information, thereby ensuring better decisionmaking at each node.

[0044] In accordance with an embodiment, the method 200 further includes decaying the state in the received state message by applying a decay factor (i.e., y) to the received state. By applying the decay factor, the method 200 is configured to reduce the influence of older or less relevant data as the state message travels through the multi-agent intelligent network 100. In an implementation, the application of the decay factor to the received state may involve multiplying the state data by the decay factor (i.e., y), which reduces its value each time it is passed through any node within the multi-agent intelligent network 100. By decaying the state, the multi-agent intelligent network 100 focuses on the most up-to-date and relevant information, which leads to better decision-making and more efficient processing. In accordance with an embodiment, the method 200 further comprises decaying the state in the received state message by applying a rotating factor (i.e., 4>) to the received state. The rotating factor refers to a mathematical value used to adjust or modify the state of data, often by altering the direction of the shared data, phase, or value over time. Furthermore, the rotating factor is used to reduce the influence of older data as the data moves through the network. Thus, by applying the rotating factor, the data is gradually rotated or adjusted so that the data becomes less significant, making sure that newer data has a stronger impact. The node is configured to apply the rotating factor to the received state, which could involve adjusting the value or direction of the state based on a fixed or dynamic mathematical rule. By virtue of applying the rotating factor, the method 200 is used to reduce the risk of outdated or overly repetitive data dominating the network, ensuring that more relevant information is prioritized.

[0045] In accordance with an embodiment, the rotating factor (i.e., 4>) is proportional to the counter (CTR* <[>). In an implementation, when the node receives the state message, the node is configured to check the counter value (i.e., CTR), which shows how many times the message has been forwarded. Furthermore, the node is configured to multiply the counter by the rotating factor (i.e., 4>) to determine how much rotation should be applied to the state. As the counter increases, the effect of the rotation becomes stronger. Moreover, after applying the rotation, the node can either process the adjusted state or forward it to the next node with the updated counter and rotating factor. By resetting the counter, the method 200 ensures an efficient use of resources, as no unnecessary data from previous operations interferes with the current task.In accordance with an embodiment, the method 200 further comprises resetting the counter (i.e., CTR) by setting the counter (CTR) to 0. The node is configured to reset the counter by setting the value of the counter to 0 at the beginning of a new task or when a specific condition triggers the reset. Once the counter is reset, the counter begins incrementing again as the task is processed or forwarded across the network.

[0046] In accordance with an embodiment, the method further comprises combining the local state to the received state by applying a decay factor (i.e., y) to the received state and applying the rotating factor (i.e., 4>) to the local state and adding them together to a combined state. In an implementation, when the node receives the state message, the node is configured to apply the decay factor (y) to the received state message, gradually reducing the value of the received state message as the received state message travels through the multi-agent intelligent network 100. Simultaneously, the node is configured to apply the rotating factor (<[>) to the local state, adjusting the local state based on the current conditions. Moreover, the node is configured to add the decayed received state message and the rotated local state together, producing a combined state that integrates both sets of information. Advantageously, the method 200 is used to improve the accuracy of the multi-agent intelligent network 100 by ensuring that the final combined state reflects both local and external knowledge, but with an emphasis on fresh, relevant data.

[0047] At step 220, the method 200 includes increasing the counter in the received state message. Moreover, the node is configured to increase the counter in the received state message. The counter indicates how many times the state message has already been forwarded. After receiving the state message, the node is configured to add the counter before either processing the received state message further or sending the received state message to the next node. Each time the message is passed along, the counter continues to increase. In an implementation, the state message can’t be forwarded by the node if the value of the counter exceeds a predefined threshold, which ensures that the outdated state messages are not endlessly forwarded to the nodes, thereby improving the quality of data and decision-making within the multi-agent intelligent network 100.

[0048] At step 222, the method 200 includes forwarding the decay of the state in the received state message to a target node according to the forwarding table based on the identifier of the received state message. Furthermore, the node is configured to forward the decay of the state in the received state message to a target node according to the forwarding table based on the identifier of the received state message. In an implementation, after applying the decay to the state in the received message, the node is configured to follow the forwarding table to identify which target node should receive the data next. The node is further configured to forward the degraded state to the target node, using the identifier in the message to ensure that the node handles the right task. At step 224, the method 200 includes providing the output as the product of the query and the local state. Furthermore, the node is configured to provide the output as the product of the query and the local state. By virtue of providing the output as the product of the query and the local state, the method 200 that the output becomes highly targeted and efficient processing.

[0049] In accordance with an embodiment, the method 200 further includes determining that the node has the same destination as the identifier of the received state message, combining the local state to the state of the received state message and appending an identifier of the node to the identifier of the received state message and provide the output as the product of the query and the combined state. Initially, the node is configured to compare the destination in the received message with its destination. Moreover, when both destinations match, the node is configured to combine its local state with the state from the message, which reflects both the local and external data. After combining both states, the node is configured to add the identifier to the received message to show that the received message has contributed to the processing. Finally, the node computes the output by multiplying the query with the combined state, generating a result that is based on both the local and external data.

[0050] In accordance with an embodiment, the method 200 further includes determining the process identifier (i.e., ID) as a time stamp. In an implementation, the node is configured to generate the process identifier by capturing the exact time when the process begins via some time-fracking mechanism. The timestamp is further used as the unique identifier, which may include elements such as the date, hours, minutes, seconds, and even milliseconds to ensure that no two processes share the same task identifier.

[0051] Advantageously, the method 200 for processing data in a node of the multi-agent intelligent network 100 allows each node within the multi-agent intelligent network 100 to function independently and efficiently within the network. Each node is designed to handle specific tasks without relying on constant instructions from a centralized system or a server. Furthermore, each node in the multi-agent intelligent network 100 is configured to receive the input tokens and process the input tokens locally. Moreover, each node only processes data that is relevant to the tasks assigned to that node, which conserves bandwidth and reduces unnecessary communication within the multi-agent intelligent network 100. By virtue of processing the input token or input data locally, the node is configured to minimize the need to send raw data across the multi-agent intelligent network 100, which improves the speed and responsiveness of the data processing while conserving resources like bandwidth. The method 200 is used to allow each node to identify the specific task the node needs to perform by determining a task identifier (ID). Moreover, each node uses counters to frack how many times a task or message has been processed or forwarded. Additionally, each node is configured to send messages only to relevant neighbours according to the forwarding table, which lists the specific nodes for further communication, thereby reducing unnecessary network traffic and improving the efficiency of data transmission. The method 200 enables each node to reduce the significance of older data in the received message, giving priority to newer information for improved decision-making.

[0052] The steps 202 to 224 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.

[0053] 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-fransitory computer-readable storage medium. The non-fransitory 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.

[0054] FIG. 3 is a diagram that depicts a node configured to process data in a node of a multi-agent intelligent network, in accordance with an embodiment of the present disclosure. FIG. 3 is described in conjunction with elements from FIG. 1 to 2. With reference to FIG. 3, there is shown a node 302 that includes a controller 304, a memory 306, and a network interface 308.

[0055] 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.

[0056] The controller 304 may include suitable logic, circuitry, interfaces, hardware, software, or code that is configured to provide a signalling 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.

[0057] 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.

[0058] The network interface 308 includes hardware or software that is configured to establish communication between 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.

[0059] There is provided the node 302 configured to process data in the node of the multi-agent intelligent network 100. The node 302 is configured to process the data by receiving input from various sources, including the local data storage and information transmitted from other nodes in the multi-agent intelligent network 100. Moreover, the node is configured to utilise specific algorithms and predefined tasks to analyze the received input, generate insights, and make decisions. Furthermore, the node 302 is configured to share the relevant processed information with other nodes within the multi-agent intelligent network 100 according to the forwarding table, contributing to the collective intelligence of the multi-agent intelligent network 100.

[0060] Furthermore, the node 302 is configured to receive the input token (i.e., X) and execute a task on the input token to provide an output (i.e., O). In an implementation, the node 302 is configured to receive the input token through the communication interfaces. Moreover, the node is configured to apply a predefined task or algorithm to the received input token, which could involve mathematical operations, logical decisions, or data transformations. The execution of the predefined task or algorithm results in the output, which can be stored locally, used for further processing, or transmitted to other nodes in the multi-agent intelligent network 100. Advantageously, input-task-output configuration of the node 302 allows each node to be assigned different tasks, improving the ability of the multi-agent intelligent network 100 to handle a wide range of data processing needs. Furthermore, the node 302 is configured to determine the local state (i.e., S) of the node 302 based on the input token (i.e., X). The node 302 is configured to process the input token through a predefined algorithm or function, which may involve calculations, data transformations, or pattern recognition. Moreover, the node 302 is configured to extract relevant information or features from the input token to update the internal state of the node 302. The local state serves as a compact representation of processed information, reducing memory requirements and facilitating efficient information sharing between nodes. Additionally, maintaining a local state allows for temporal consistency in decision-making, as the node 302 can consider both current inputs and past processed information, leading to the execution more robust and context-aware tasks within the multiagent intelligent network 100.

[0061] In accordance with an embodiment, the node 302 is further configured to determine the local state (i.e., S) of the node based on the input tokens (i.e., X) by determining the key (i.e., K) being the product of a key weight (i.e., WK) and the input tokens (X) (K=WKX) and a value (V) being the product of a value weight (Wv) and the input token (X) (V=WvX), wherein the local state is determined as the product of a transpose of the key (K) and the value (V) (S = KTV)). By applying the key weight, the node 302 is configured to capture the critical features from the input tokens, enhancing the ability of the node 302 to process the received information efficiently. Moreover, the value weight allows the node 302 to interpret the input tokens in a way that aligns with the overall objectives of the task. Additionally, the determination of the local state as the product of the transpose of the key and the value provides a highly efficient way to combine the extracted features and insights from the input tokens.

[0062] Furthermore, the node is configured to determine the task identifier (i.e., ID) and reset the counter (i.e., CTR). The node is configured to assign a unique task identifier to each new processing operation, typically using a combination of timestamps, node identifiers, and task-specific information. The counter (CTR) is further initialized to a starting value, usually zero. As the task progresses through the multi-agent intelligent network 100, the counter is incremented at each step or node interaction, providing a measure of the task's progress or the number of hops in the multi-agent intelligent network 100. As a result, the counter prevents infinite loops in data propagation by setting a maximum number of allowed hops, enhancing the overall stability of the multi-agent intelligent network 100.

[0063] Furthermore, the node 302 is configured to send a state message to direct neighbours according to a forwarding table. The node 302 is configured to check the forwarding table to determine which neighbouring nodes should receive the state message. Furthermore, the forwarding table indicates which nodes in the multi-agent intelligent network the node 302 are to transmit messages to. The forwarding table contains rules or mappings that specify the recipients for different types of information or tasks. Once the recipient nodes are identified, the node 302 is configured to package the state information into a message format and transmit the state message to the selected neighbouring nodes through the communication channels of the multi-agent intelligent network 100. By sending the targeted state messages based on the forwarding table, the node 302 enhances the overall scalability of the multi-agent intelligent network 100.

[0064] Furthermore, the state message includes the identifier, the task and the local state. In an implementation, the node 302 is configured to construct the state message by combining the unique identifier assigned to the current operation, the details of the task being executed, and the current local state of the node 302. Moreover, the state message is formatted into a structured message format before transmission. While sending the message, the node 302 includes all three components, ensuring that receiving nodes can accurately interpret and utilize the information. As a result, the identifier is used to allow the nodes to recognize and handle duplicate messages, enhancing the reliability of the multi-agent intelligent network 100.

[0065] Furthermore, the node 302 is further configured to receive the state message from another node and determine if a counter in the received state message is less than a maximum counter value. In an implementation, when the node 302 receives the state message, the node is configured to extract the counter value from the state message. Moreover, the node 302 is configured to compare the counter value to a preset maximum counter value. Furthermore, if the extracted counter value is less than the maximum value, the node 302 is configured to proceed with processing or forwarding the message. Furthermore, if the counter value equals or exceeds the maximum value, the node 302 is then configured to further processing of the message. In an implementation scenario, whenever the node 302 obtains some new data, the node 302 is configured to send a message that may comprise various entities, such as ID( sender), {ID(origin)}, {task(origin)}, state(origin), a counter set to 0, and a unique process identifier to neighboring nodes listed in the forwarding table. In another implementation, whenever the node 302 receives a message with a state from {ID(origin)} and a counter smaller than CTR_MAX, it increases the counter, updates the state, and forwards the updated state to next nodes selected from the forwarding table. Furthermore, the updated state forwarded by the node 302 could be either a received state decayed by a constant factor, or a combination of the received state or the local state of the node 302, which depends on the local forwarding table of the node 302. In an implementation scenario, if the updated state is a combination of the received and the local states, the node 302 is configured to appends the ID to {ID(origin)} and its task to {task(origin)} . Additionally, the node 302 is configured to perform inference based on local and received states. Therefore, the inclusion of the counter allows for controlled information sharing, ensuring that data reaches an optimal number of nodes without overwhelming the network. Furthermore, the node 302 is configured to decay the state in the received state message. In an implementation, when the node 302 receives the state message, the node 302 is configured to apply a predefined decay function to the state information. The predefined decay function may involve mathematical operations such as multiplication by a factor less than one, subtraction of a fixed value, or more complex transformations. The degraded state is then used for local processing or included in outgoing messages, representing a diminished version of the original state. Advantageously, the degradation of the state message contributes to the ability of the multi-agent intelligent network 100 to adapt to changing conditions.

[0066] In accordance with an embodiment, the node 302 is further configured to decay the state in the received state message by applying a decay factor (i.e., y) to the received state. By applying the decay factor, the node 302 is configured to reduce the influence of older or less relevant data as the state message travels through the multi-agent intelligent network 100. In an implementation, the application of the decay factor to the received state may involve multiplying the state data by the decay factor (i.e., y), which reduces the value of the received state message each time it is passed through any node within the multiagent intelligent network 100. By applying the decay factor, the node 302 focuses on the most up-to-date and relevant information, which leads to better decision-making and more efficient processing.

[0067] In accordance with an embodiment, the node 302 is further configured to decay the state in the received state message by applying a rotating factor (i.e., 4>) to the received state. The rotating factor is used to reduce the influence of older data as the data moves through the network. By applying the rotating factor, the data is gradually rotated or adjusted so that the data becomes less significant, making sure that newer data has a stronger impact. The node 302 is configured to apply the rotating factor to the received state, which could involve adjusting the value or direction of the state based on a fixed or dynamic mathematical rule. By virtue of gradually rotating the state, the node 302 is configured to reduce the risk of outdated or overly repetitive data dominating the network, ensuring that more relevant information is prioritized.

[0068] In accordance with an embodiment, the rotating factor is proportional to the counter (i.e., CTR*<[>). In an implementation, when the node 302 receives the state message, the node 302 is configured to check the counter value (i.e., CTR), which shows how many times the message has been forwarded. Furthermore, the node 302 is configured to multiply the counter by the rotating factor (i.e., 4>) to determine how much rotation should be applied to the state. As the counter increases, the effect of the rotation becomes stronger. Moreover, after applying the rotation, the node 302 can either process the adjusted state or forward it to the next node with the updated counter and rotating factor. By resetting the counter, the node 302 ensures an efficient use of resources, as no unnecessary data from previous operations interferes with the current task. information. Advantageously, the node 302 is configured to improve the accuracy of the multi-agent intelligent network 100 by ensuring that the final combined state reflects both local and external knowledge, but with an emphasis on fresh, relevant data.

[0069] Furthermore, the node 302 is configured to increase the counter in the received state message. In an implementation, when the node 302 receives the state message, the node 302 is configured to extract the current counter value from the state message. The node 302 is further configured to increments the current counter value by a predetermined amount, typically by one. The increased counter value is then incorporated into the message before any further processing or forwarding occurs. The incrementing counter facilitates the creation of self-terminating message chains, enhancing the stability of the multi-agent intelligent network 100 and preventing congestion from endlessly circulating messages.

[0070] Furthermore, the node 302 is configured to forward the decay of the state in the received state message to a target node according to the forwarding table based on the identifier of the received state message. In an implementation, while processing the received state message, the node 302 is configured to check the forwarding table, using the identifier message as a key. The forwarding table determines the appropriate target nodes for the degraded state information. The node 302 is further configured to package the degraded state into a new message and transmit the delayed state to the identified target nodes, propagating the information while preserving the diminished influence of the delayed state.

[0071] Furthermore, the node 302 is configured to provide the output as the product of the query and the local state. In an implementation, when the node 302 receives a query, the node 302 is configured to retrieve the current local state. Moreover, the node 302 is configured to perform a multiplication operation between the query and the local state, which becomes the output of the node 302. The combination of query and local state in the output calculation allows for more informative responses.

[0072] In accordance with an embodiment, the node 302 is further configured to determine that the node 302 has the same destination as the identifier of the received state message, combines the local state to the state of the received state message and appends an identifier of the node to the identifier of the received state message and provide the output as the product of the query and the combined state. Initially, node 302 is configured to compare the destination in the received message with its own destination. Moreover, when both destinations match, the node 302 is configured to combine its local state with the state from the message, which reflects both the local and external data. After combining both states, the node 302 is configured to add the identifier to the received message to show that the received message has contributed to the processing. Finally, the node 302 computes the output by multiplying the query with the combined state, generating a result that is based on both the local and external data.

[0073] Advantageously, the node 302 enables efficient, adaptive, and distributed information processing within the multi-agent intelligent network 100. The configuration of the node 302 allows for scalable and robust network operations by implementing local data processing, selective information sharing, and controlled message propagation. Further, the capacity of the node 302 to decay and forward state information, manage message counters, and combine queries with local states supports dynamic decision-making and adaptability to changing network conditions. By utilizing forwarding tables and unique identifiers, the node 302 ensures targeted communication, reducing network congestion while maintaining relevant information flow.

[0074] FIG. 4 is a diagram that illustrates an exemplary scenario of a distributed mesh topology, 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 of a distributed mesh topology that includes a first node 402A, a second node 402B, and a third node 402C.

[0075] In an implementation scenario, the first node 402A, the second node 402B, and the third node 402C utilize local data to produce memory, where each node shares knowledge with neigh boring nodes using a local attention-based forwarding table. In an example, the second node 402B and the third node 402C exchanges the memory states with the first node 402A, and the first node 402A and the third node 402C exchanges the memory states with the second node 402B. Moreover, after such data exchange, each node is configured to update the cluster table which lists all the nodes having useful data for each node. Furthermore, the first node is configured to preform a first task 404A. Similarly, the second node 402B is configured to perform a second task 404B, and the third node is configured to perform a third task 404C. Moreover, each node in the distributed mesh topology is equipped with linear attention modules with parameters WQ, WK, Wv (learned matrices of dimension d x d), a decay fcator 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 the nodes present in the distributed mesh topology. Furthermore, each node has access to its own input data. Additionally, each node in the distributed mesh topology is configured to use local data to produce memory states and send states to neighbour devices listed in the local forwarding tables. In an example, if the first node 402A receives a state that is also useful for another node, that is, the second node 402B and the third node 402C, the first node 402A combines linearly the received and local states and sends a combined state to the second node 402B and the third node 402C.

[0076] In the first scenario, the first node 402A is configured to compute a local memory state based on the first input data 406 using the linear attention module. Furthermore, the first node 402A is configured to compute a local memory state based on the input data using the linear attention module. Moreover, the first node 402A is configured to assign a unique inference (ID) based on the combination of the timestamp and the geo location. The first node 402A is further configured to initialize the counter (CTR) to keep track of how many times the message is forwarded. The first node 402A is further configured to send the first state message 408 to the neighbouring nodes (i.e., the second node 402B, and the third node 402C) listed in the forwarding table of the first node. In an implementation, the first state message may include essential information, such as the ID of the sender node, the task ID, and the memory state. Furthermore, the first node 402A is configured to send the first state message 408 to the nodes listed in the forwarding table, which includes the ID of the sender (i.e., ID (sender )=ID(k)), the ID of the node producing the original state (here, ID(sender)=ID(k) ), the task id of the node producing the original state (i.e., the task id of the first node 402A), the unique inference ID, the memory state (i.e., the memory state of the first node 402A), the counter of the memory state (i.e., CTR = CTR_init). The initialization and sharing of the first state to the neighbour nodes ensures that any node with new knowledge can independently initiate the sharing process, which makes the distributed mesh topology flexible and responsive to changing conditions in the multi-agent intelligent network 100. Moreover, the initialization process also provides transparency by clearly identifying the origin of the data and the task that it is related to.

[0077] In a second scenario, when any node (i.e., the first node 402A, the second node 402B, and the third node 402C) is configured to receive the second state message 410 from another node, where the state message may include details such as the origin of the data, a combined memory state that may contain contributions from multiple agents, and a counter that tracks how many times the message has been forwarded. Further, the second state message 410 includes the ID of the sender, the ID of all nodes contributing to the received state (i.e., {ID (origin)} _all (e.g., {ID (1), ID(2), .., ID(n)}) ), the task id of all nodes contributing to the received state (e.g., {task(l), task(2), .., task(n)} ), the unique inference ID, the combined memory state combining knowledge from one or more nodes, the counter of the memory state (i.e., CTR). In an example, the first node 402A receives the state message, and then the first node 402A is configured to check if the counter has reached a maximum value (i.e., CTR_max) or if the state message has been forwarded to all relevant destinations (or nodes) listed in the forwarding table. Moreover, if the counter has reached to a maximum point or if there are no more new destinations, the first node 402A is configured to stop forwarding the state message. Furthermore, when new destinations are still present in the forwarding table, then the first node 402A is configured to combine the received memory state with its own local state, append the ID and task to the state message, and increase the counter to the next value. The updated state message is then forwarded to the next destination. In some implementations, if the destination (or the node) is not relevant for the first node 402A, then the first node 402A is configured to decay the received state message using a decay coefficient (i.e., y), increases the counter, and forwards the degraded state to other nodes, thereby ensuring that only relevant information is shared across the multi-agent intelligent network 100, and outdated or less important data is decayed to maintain efficiency.

[0078] In a third scenario, each node (i.e., the first node 402A, the second node 402B, and the third node) is configured to use both the local memory states and any received memory states to process an input data (i.e., the first input data 406), which involves attending to relevant data from other nodes using the linear attention module. Thus, by using both the local memory states and any received memory states to process an input data, the nodes can make better-informed decisions, as the nodes now have access to a broader range of knowledge from other nodes in the multi-agent intelligent network 100.

[0079] Advantageously, each node in the multi-agent intelligent network 100 is configured to independently process data and share knowledge with the neighbour nodes in a highly organized and efficient manner, prioritizing the new information, minimizing unnecessary data transmissions, and ensuring that relevant data is shared with the right node at the right time. The inclusion of decay mechanisms, counters, and forwarding tables ensures that the multi-agent intelligent network 100 remains scalable and avoids overload, even as more nodes are added. The use of attention-based forwarding ensures that only relevant information is shared with the necessary nodes, making the multi-agent intelligent network 100 more reliable and efficient. Moreover, the decay mechanism prevents old or irrelevant data from overwhelming the network, and the counters ensure that messages do not circulate endlessly.

[0080] FIG. 5 is a diagram that illustrates an exemplary scenario of knowledge flow between agents in a fixed mesh network, 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. 4, there is shown a diagram 500 that illustrates an exemplary scenario of the knowledge flow between agents in the fixed mesh network, which includes the first node 402A, the second node 402B, the third node 402C, a fourth node 502A, a fifth node 502B.

[0081] In an exemplary scenario, the fixed mesh network includes multiple interconnected nodes, labelled with identifiers such as the first node 402A, the second node 402B, the third node 402C, the fourth node 502A, and the fifth node 502B andeach node in the fixed mesh network includes specific data, such as the first node 402A includes a first local data 504A, the first input data 406, and is configured to perform the first task 404A. Similarly, the second node 402B includes a second local data 504B, a second input data 508A, and is configured to perform the second task 404B. Similarly, the third node 402C includes a third local data 504C, a third input data 508B, and is configured to perform the third task 404C. Similarly, the fourth node 502A includes a fourth local data 504D, a fourth input data 508C, and is configured to perform the fourth task 506A. Similarly, the fifth node 502B includes the fourth local data 504D, a fifth input data 508D, and is configured to perform the fifth task 506B.

[0082] In an implementation, each node (i.e., the first node 402A, the second node 402B, the third node 402C, the fourth node 502A, and the fifth node 504B) receives input data from other nodes, processes the data alongside the local data, and performs the assigned task. Moreover, each node (i.e., the first node 402A, the second node 402B, the third node 402C, the fourth node 502A, and the fifth node 504B) is configured to use forwarding tables to determine which information to share and with whom. As nodes process the tasks using combined local and received data, they generate new insights or results, which are then shared with relevant neighbouring nodes. Moreover, the continuous cycle of receiving, processing, and transmitting data creates a fluid knowledge flow throughout the multi-agent intelligent network 100. Advantageously, the fixed mesh network ensures that each node knows exactly to which other nodes they are connected, using predetermined communication pathways. Furthermore, the fixed mesh network supports scalability, efficient data flow, and improved collaboration between nodes, which enhances the performance and decision-making ability of the multi-agent intelligent network 100.

[0083] 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 processing data in a node (302) of a multi-agent intelligent network (100), wherein the node (302) is arranged to receive input token (X) and to execute a task on the input token to provide an output (O), wherein the method (200) comprises the node (302) determine a local state (S) of the node (302) based on the input token (X) determine a task identifier (ID) and reset a counter (CTR), send a state message to direct neighbours according to a forwarding table, wherein the forwarding table indicates which nodes in the multi-agent intelligent network (100) the node (302) is to transmit messages to, and wherein the state message includes the identifier, the task and the local state, wherein the method (200) further comprises the node (302) receive a state message from another node (302) and determine if a counter in the received state message is less than a maximum counter value, and then decay the state in the received state message, increase the counter in the received state message, and forward the decay of the state in the received state message to a target node according to the forwarding table based on the identifier of the received state message, and then provide the output as the product of the query and the local state.

2. The method (200) according to claim 1 , wherein the method (200) further comprises determine that the node (302) has a same destination as the identifier of the received state message, and then also combine the local state to the state of the received state message and appending an identifier of the node (302) to the identifier of the received state message, and provide the output as the product of the query and the combined state.

3. The method (200) according to claim 1 or 2, wherein the method (200) further comprises: decay the state in the received state message by applying a decay factor (y) to the received state.

4. The method (200) according to any preceding claim, wherein the method (200) further comprises decay the state in the received state message by applying a rotating factor (<[>) to the received state.

5. The method (200) according to any preceding claim, wherein the rotating factor is proportional to the counter (CTR*<[>).

7. The method (200) according to any preceding claim, wherein the method (200) further comprises determine the local state (S) of the node (302) based on the input tokens (X) by determining a Key (K) being the product of a key weight (WK) and the input tokens (X) (K=WKX) and a Value (V) being the product of a value weight (WV) and the input token (X) (V=WVX), wherein the local state is determined as the product of a transpose of the key (K) and the value (V) (S = KTV)).

8. The method (200) according to any preceding claim, wherein the method (200) further comprises determine the process identifier (ID) as a time stamp.

9. The method (200) according to any preceding claim, wherein the method (200) further comprises reset the counter (CTR) by setting the counter (CTR) to 0.

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 process data in a node (302) of a multi-agent intelligent network (100), wherein the node (302) is configured to receive input token (X) and to execute a task on the input token to provide an output (O), wherein the node (302) is configured to: determine a local state (S) of the node (302) based on the input token (X) determine a task identifier (ID) and reset a counter (CTR), send a state message to direct neighbours according to a forwarding table, wherein the forwarding table indicates which node (302) s in the multi-agent intelligent network (100) the node (302) is to transmit messages to, and wherein the state message includes the identifier, the task and the local state, wherein the node (302) is further configured to: receive a state message from another node (302) and determine if a counter in the received state message is less than a maximum counter value, and then decay the state in the received state message, increase the counter in the received state message, and forward the decay the state in the received state message to a target node according to the forwarding table based on the identifier of the received state message, and then provide the output as the product of the query and the local state.

13. The node (302) according to claim 12, wherein the node (302) is further configured to: determine that the node (302) has a same destination as the identifier of the received state message, and then also combine the local state to the state of the received state message and appending an identifier of the node (302) to the identifier of the received state message, and provide the output as the product of the query and the combined state.

14. The node (302) according to claim 12 or 13, wherein the node (302) is further configured to decay the state in the received state message by applying a decay factor (y) to the received state.

15. The node (302) according to any of claims 12 to 14, wherein the node (302) is further configured to decay the state in the received state message by applying a rotating factor (<[>) to the received state.

16. The node (302) according to any of claims 12 to 15, wherein the rotating factor is proportional to the counter (CTR*<[>).

18. The node (302) according to any of claims 12 to 17, wherein the node (302) is further configured to determine the local state (S) of the node (302) based on the input tokens (X) by determining a Key (K) being the product of a key weight (WK) and the input tokens (X) (K=WKX) and a Value (V) being the product of a value weight (WV) and the input token (X) (V=WVX), wherein the local state is determined as the product of a transpose of the key (K) and the value (V) (S = KTV).

Images