Execution base, control method, program, and system

Abstract

This execution base, which has a RAN control function and an AI processing (RAN intelligent controller (RIC) or the like) function, comprises: a communication state acquisition unit that acquires a communication state between each of a plurality of nodes and the execution base; a centralized execution control unit that inputs, to a first learning model, data acquired from each of multiple nodes, of which the communication states satisfy a predetermined condition, among the plurality of nodes and transmits data output from the first learning model to the multiple nodes; and a distributed execution control unit that inputs, to a second learning model stored in advance, data from multiple nodes, of which the communication states do not satisfy the predetermined condition, among the plurality of nodes and causes the nodes to execute processing according to data that is output from the second learning model

Description

Execution platform, control method, program, and system

[0001] The present invention relates to an execution platform, a control method, a program, and a system.

[0002] Patent Document 1 describes a technology in the field of IoT (Internet of Things) in which an edge device switches between machine-learned models depending on the communication path. [Prior art documents] [Patent documents] [Patent document 1] International Publication No. 2019 / 193661

[0003] Reinforcement learning is another pillar of machine learning alongside unsupervised learning and supervised learning, and is a learning method in which an agent learns to maximize the reward it receives by taking action in an environment based on a policy. Multi-agent reinforcement learning targets problems in which there are multiple independent agents in a single environment, and they do not necessarily have to cooperate, but can also be adversarial or neutral.

[0004] There are three main problem configurations in multi-agent reinforcement learning. One is DTDE (Decentralized Training Decentralized Execution) (distributed learning-distributed execution), which is a problem involving multiple single agents sharing the same environment, and therefore has little value in solving it as multi-agent reinforcement learning. The other is CTCE (Centralized Training Centralized Execution) (centralized learning-centralized execution), which assumes infinite communication capacity and is therefore essentially equivalent to a single-agent problem. The other is CTDE (Centralized Training Decentralized Execution), which is a highly valuable setup for solving multi-agent reinforcement learning problems with limited information sharing.

[0005] CTDE is the most realistic setting, where agents can be trained while referencing information from other agents during learning, but each agent runs independently during runtime. CTCE, on the other hand, assumes infinite information sharing between agents, and is expected to have higher performance than CTDE. In a real-world environment where the communication environment (information sharing possibility) changes, learning and execution based on a policy that assumes CTDE from the beginning would result in inefficient use of information from other agents when there is communication availability, resulting in suboptimal performance. The system according to this embodiment provides a technique for switching between a policy learned with a CTCE setting and a policy learned with a CTDE setting, depending on the communication status of the multi-agent system. This makes it possible to use a policy learned with the optimal setting depending on the communication status.

[0006] As a specific example, the system according to this embodiment includes a distributed infrastructure having a RAN control function for controlling a RAN (Radio Access Network) and an AI processing function for performing AI processing, and a management infrastructure for managing multiple distributed infrastructures, and performs various processes using the distributed infrastructures and UEs that use the RAN as multi-agents.

[0007] Types of AI processing include AI processing related to RAN control (sometimes referred to as RAN-controlled AI processing) and AI processing not related to RAN control (sometimes referred to as non-RAN-controlled AI processing).

[0008] An example of RAN control AI processing is RIC (RAN Intelligent Controller). RIC is a technology that uses AI to optimize RAN radio resources and automate RAN operations. RIC includes Non-RT RIC (Non-Real Time RIC) and Near-RT RIC (Near-Real Time RIC). Non-RT RIC is sometimes called Centralized RIC. Non-RT RIC is located inside SMO (Service Management and Orchestration), which manages and orchestrates the RAN. Non-RT RIC generates and notifies policies related to RAN control and sends information to Near-RT RIC. For example, the Non-RT RIC performs machine learning using data collected from the RAN to generate a trained model for RAN control and transmits it to the Near-RT RIC. The Near-RT RIC is sometimes called a Distributed RIC. Compared to the Non-RT RIC, the Near-RT RIC is located closer to the RAN nodes (RU (Radio Unit), DU (Distributed Unit), CU (Central Unit)) and controls the RAN nodes, resources, etc. The Near-RT RIC performs processing with higher real-time performance than the Non-RT RIC. The Near-RT RIC performs inference processing related to RAN control using, for example, a trained model acquired from the Non-RT RIC. RAN control AI processing is not limited to the RIC.

[0009] The non-RAN control AI processing may correspond to a so-called MEC (Multi-access Edge Computing) application. Examples of the non-RAN control AI processing include a monitoring AI execution process that determines the situation within the imaging range of an input captured image, and a response AI execution process that outputs a response to an input user inquiry. However, this is not limited to these.

[0010] According to one embodiment of the present invention, an execution platform is provided. The execution platform may include a communication status acquisition unit that acquires communication status between each of a plurality of nodes and the execution platform. The execution platform may include a centralized execution control unit that inputs data acquired from each of a plurality of nodes whose communication status satisfies a predetermined condition into a first learning model and transmits data output from the first learning model to the plurality of nodes. The execution platform may include a distributed execution control unit that causes a plurality of nodes whose communication status does not satisfy the predetermined condition to input data from the node into a pre-stored second learning model and execute processing according to the data output from the second learning model.

[0011] The execution platform may include a first learning model generation unit that receives data from multiple nodes as input and generates the first learning model as output data for cooperative operation of the multiple nodes. The execution platform may include a second learning model generation unit that receives data from one node as input and generates the second learning model as output data for operating the one node. The execution platform may include a learning model transmission unit that transmits the second learning model to the multiple nodes. The first learning model generation unit may generate the first learning model according to a policy based on CTCE. The second learning model generation unit may generate the second learning model according to a policy based on CTDE. The first learning model generation unit may generate the first learning model for each combination of the multiple nodes, and the centralized execution control unit may input data acquired from each of multiple nodes whose communication conditions satisfy predetermined conditions into the first learning model corresponding to the multiple nodes, and transmit data output from the first learning model to the multiple nodes.

[0012] In any of the execution platforms, the centralized execution control unit may input data acquired from each of a plurality of nodes among the plurality of nodes whose communication conditions with the execution platform are better than a predetermined degree into the first learning model and transmit data output from the first learning model to the plurality of nodes. The distributed execution control unit may input data of the node into the second learning model for a plurality of nodes among the plurality of nodes whose communication conditions with the execution platform are better than the predetermined degree and cause the plurality of nodes to execute processing in accordance with the data output from the second learning model.

[0013] The execution platform may further include a RAN control unit that executes a RAN control function to control a RAN configured by a plurality of radio base stations. The communication status acquisition unit may acquire a communication status of communication between each of the plurality of nodes and the execution platform via the RAN.

[0014] According to one embodiment of the present invention, a program is provided for causing a computer to execute the following steps: a communication status acquisition step for acquiring communication status between each of a plurality of nodes and the execution platform; a centralized execution control step for inputting data acquired from each of a plurality of nodes among the plurality of nodes whose communication status satisfies predetermined conditions into a first learning model and transmitting data output from the first learning model to the plurality of nodes; and a distributed execution control step for inputting data from the node into a pre-stored second learning model for a plurality of nodes among the plurality of nodes whose communication status does not satisfy the predetermined conditions and causing processing according to the data output from the second learning model.

[0015] According to one embodiment of the present invention, there is provided a control method executed by a computer. The control method may include a communication status acquisition step of acquiring communication status between each of a plurality of nodes and the execution platform. The control method may include a centralized execution control step of inputting data acquired from each of a plurality of nodes whose communication status satisfies a predetermined condition into a first learning model and transmitting data output from the first learning model to the plurality of nodes. The control method may also include a distributed execution control step of inputting data from the node into a pre-stored second learning model and causing a plurality of nodes whose communication status does not satisfy the predetermined condition to execute processing in accordance with the data output from the second learning model.

[0016] According to one embodiment of the present invention, a system is provided. The system may include a plurality of distributed infrastructures and a management infrastructure that manages the plurality of distributed infrastructures. The management infrastructure may include a communication status acquisition unit that acquires, for each of the plurality of distributed infrastructures, a communication status between the distributed infrastructure and each of a plurality of nodes and a communication status between the distributed infrastructure and the management infrastructure. The management infrastructure may include a centralized execution control unit that inputs data acquired from each of a plurality of nodes among the plurality of nodes, where the communication status between the node, the distributed infrastructure, and the management infrastructure satisfies a predetermined condition, into a first learning model and transmits data output from the first learning model to the plurality of nodes. The management infrastructure may include a distributed execution control unit that inputs data of the node into a pre-stored second learning model and causes a plurality of nodes among the plurality of nodes, where the communication status between the node, the distributed infrastructure, and the management infrastructure does not satisfy the predetermined condition, to execute processing in accordance with the data output from the second learning model.

[0017] The above summary of the invention does not list all of the necessary features of the present invention, and subcombinations of these features may also constitute inventions.

[0018] 1A and 1B schematically illustrate an example of a system 10. FIG. 1B is an explanatory diagram illustrating a method for managing a plurality of nodes 400 by a distributed infrastructure 200. FIG. 1C is an explanatory diagram illustrating a method for managing a plurality of nodes 400 by the distributed infrastructure 200 in accordance with policies 510 and 520. FIG. 1D is an explanatory diagram illustrating reallocation of policies by the distributed infrastructure 200. FIG. 1E is an explanatory diagram illustrating reallocation of policies by the distributed infrastructure 200. FIG. 1F is an explanatory diagram illustrating an example of the functional configuration of the distributed infrastructure 200. FIG. 1F is an explanatory diagram illustrating an example of the flow of processing by the distributed infrastructure 200. FIG. 1F is an explanatory diagram illustrating a method for managing a plurality of nodes 400 by a management infrastructure 100. FIG. 1F is an explanatory diagram illustrating a method for managing a plurality of nodes 400 by the management infrastructure 100 in accordance with policies 510 and 520. FIG. 1C is an explanatory diagram illustrating an example of the functional configuration of the management infrastructure 100. FIG. 1D is an explanatory diagram illustrating an example of the hardware configuration of a computer 1200 that functions as the management infrastructure 100 or the distributed infrastructure 200.

[0019] The present invention will be described below through embodiments of the invention, but the following embodiments do not limit the scope of the invention as claimed. Furthermore, not all of the combinations of features described in the embodiments are necessarily essential to the solution of the invention.

[0020] 1 schematically illustrates an example of a system 10. The system 10 includes a distributed infrastructure 200. The distributed infrastructure 200 may be an example of an execution infrastructure. The system 10 may include a plurality of distributed infrastructures 200. The system 10 may include a management infrastructure 100 that manages the plurality of distributed infrastructures 200. In the system 10 according to this embodiment, the management infrastructure 100 and the plurality of distributed infrastructures 200 may cooperate to control the RAN 310 and perform AI processing.

[0021] The RAN 310 may be a virtualized vRAN (Virtual RAN), and the system 10 may control the vRAN. The RAN 310 may be a physical RAN, and the system 10 may control the physical RAN. In this embodiment, a case where the RAN 310 is a vRAN will be mainly described as an example.

[0022] The AI ​​processing performed by the system 10 may include RAN-controlled AI processing (sometimes referred to as RAN_AI). The AI ​​processing performed by the system 10 may include non-RAN-controlled AI processing (sometimes referred to as non-RAN_AI).

[0023] The distributed infrastructure 200 may be a data center located in various locations. The distributed infrastructure 200 may be configured by multiple devices. The distributed infrastructure 200 may be realized on a virtualization platform made up of multiple devices. The distributed infrastructure 200 may also be realized by a single device. In other words, the distributed infrastructure 200 may be a distributed device.

[0024] The management infrastructure 100 may be a data center that manages multiple distributed infrastructures 200. The management infrastructure 100 may be configured with multiple devices. The management infrastructure 100 may be realized on a virtualization infrastructure made up of multiple devices. The management infrastructure 100 may also be realized by a single device. In other words, the management infrastructure 100 may be a management device.

[0025] The management infrastructure 100 may be referred to as a Core Brain, and the distributed infrastructure 200 may be referred to as a Regional Brain. While FIG. 1 illustrates an example in which a single-level management infrastructure 100 is arranged below the management infrastructure 100, this is not limiting. The distributed infrastructure 200 may have multiple levels. For example, when a two-level distributed infrastructure 200 is arranged below the management infrastructure 100, the management infrastructure 100 may be referred to as a Core Brain, the distributed infrastructure 200 at the level below that may be referred to as a Regional Brain, and the distributed infrastructure 200 at the level further below that may be referred to as a Sub-Regional Brain.

[0026] The distributed infrastructure 200 may be arranged with one or more central processing units (CPUs). The distributed infrastructure 200 may be arranged with one or more graphics processing units (GPUs). The distributed infrastructure 200 may be arranged with multiple super chips, each of which has a CPU and a GPU connected via an interconnect. The interconnect may have memory consistency and may be capable of achieving high bandwidth and low latency. In this way, the distributed infrastructure 200 may have CPU resources and GPU resources as computational resources.

[0027] 2 is an explanatory diagram illustrating a method for managing a plurality of nodes 400 by a distributed infrastructure 200. The nodes 400 may be UEs 30. The UEs 30 may be mobile devices such as smartphones, tablet terminals, PCs, automobiles, and unmanned aerial vehicles, IoT terminals, and IoE (Internet of Everything) terminals.

[0028] The distributed infrastructure 200 acquires the communication status between the multiple nodes 400 and the distributed infrastructure 200. The communication status may include throughput. The higher the throughput, the better the communication status. The communication status may include communication delay. The lower the communication delay, the better the communication status. The communication status may include communication quality. The higher the communication quality, the better the communication status. The communication status may include communication capacity. The higher the communication capacity, the better the communication status. The communication status may include communication bandwidth. The larger the communication bandwidth, the better the communication status. The communication status may include network cost. The lower the network cost, the better the communication status.

[0029] The distributed infrastructure 200 may assign the policy 510 to one or more nodes 400 whose communication conditions satisfy predetermined conditions among the multiple nodes 400. The distributed infrastructure 200 may assign the policy 520 to one or more nodes 400 whose communication conditions do not satisfy predetermined conditions among the multiple nodes 400.

[0030] The policy 510 may be a policy for executing processing using data acquired from each of the plurality of nodes 400. The policy 510 may be a policy based on the CTCE.

[0031] The policy 520 may be a policy in which each node 400 performs execution processing using only its own data, without using data from other nodes 400. The policy 520 may be a policy based on the CTDE.

[0032] The predetermined condition may be a condition that is satisfied when the degree of goodness of the communication conditions is greater than the predetermined degree. If the communication conditions are throughput, the predetermined degree may be a threshold value for throughput. If the communication conditions are communication delay, the predetermined degree may be a threshold value for communication delay. If the communication conditions are communication quality, the predetermined degree may be a threshold value for communication quality. If the communication conditions are communication capacity, the predetermined degree may be a threshold value for communication capacity. If the communication conditions are communication bandwidth, the predetermined degree may be a threshold value for communication bandwidth. If the communication conditions are network cost, the predetermined degree may be a threshold value for network cost. These threshold values ​​may be arbitrarily settable or may be changeable after being set.

[0033] The distributed infrastructure 200 may assign the policy 510 to one or more nodes 400 whose communication conditions are more favorable than a predetermined level among the multiple nodes 400. The distributed infrastructure 200 may assign the policy 520 to one or more nodes 400 whose communication conditions are less favorable than a predetermined level among the multiple nodes 400.

[0034] 2 , of the five nodes 400, three nodes 400 have good communication conditions and policy 510 is assigned to the three nodes 400, and two nodes 400 have poor communication conditions and policy 520 is assigned to the two nodes 400. The number of nodes 400 is not limited to five, and may be four or less, or six or more.

[0035] FIG. 3 is an explanatory diagram for explaining a method for managing a plurality of nodes 400 by the distributed infrastructure 200 in accordance with the policy 510 and the policy 520. As shown in FIG.

[0036] The distribution platform 200 generates and stores the learning model 610 in advance. The learning model 610 receives data from the multiple nodes 400 as input and outputs data for causing the multiple nodes 400 to operate cooperatively. For example, if the multiple nodes 400 correspond to multiple joints of a robot, the learning model 610 plays a role in causing the multiple joints to operate cooperatively according to the conditions of the multiple joints. For example, if the multiple nodes 400 correspond to multiple autonomous vehicles, the learning model 610 plays a role in causing each of the multiple autonomous vehicles to perform autonomous driving while causing the multiple autonomous vehicles to operate cooperatively to smooth road traffic according to the conditions of the multiple autonomous vehicles. The distribution platform 200 may generate the learning model 610 in accordance with the CTCE policy. The distribution platform 200 may generate the learning model 610 by executing multi-agent reinforcement learning in accordance with the CTCE policy.

[0037] The distribution platform 200 generates a learning model 620 in advance and provides it to the multiple nodes 400. The learning model 620 receives data from one node 400 as input and outputs data for operating the node 400. For example, if the multiple nodes 400 correspond to multiple joints of a robot, the learning model 620 serves to operate one joint normally without considering the status of the other joints. For example, if the multiple nodes 400 correspond to multiple autonomous vehicles, the learning model 620 serves to cause one autonomous vehicle to perform autonomous driving without considering the status of the other autonomous vehicles. The distribution platform 200 may generate the learning model 620 in accordance with a CTDE policy. The distribution platform 200 may generate the learning model 620 by performing multi-agent reinforcement learning in accordance with the CTDE policy.

[0038] 3 , for three nodes 400 to which policy 510 is assigned, the distributed infrastructure 200 inputs data received from each of the three nodes 400 into the learning model 610 and transmits data output from the learning model 610 to the three nodes 400. The three nodes 400 execute processing in accordance with the data received from the distributed infrastructure 200. For two nodes 400 to which policy 520 is assigned, the distributed infrastructure 200 inputs only its own data into the learning model 620 and executes processing in accordance with the data output from the learning model 620.

[0039] This allows multiple nodes 400 in which the communication conditions are good and data can be sufficiently shared between the distributed infrastructure 200 and the nodes 400 to perform cooperative operations, while nodes 400 in which the communication conditions are not good and data cannot be sufficiently shared between the distributed infrastructure 200 and the nodes 400 to perform individual operations without being involved in cooperative operations, thereby contributing to optimizing performance according to the communication conditions.

[0040] The distributed infrastructure 200 may periodically acquire the communication status of multiple nodes 400 and reallocate policies in response to changes in the communication status. For example, when the communication status of a node 400 that previously satisfied a condition no longer satisfies the condition, the distributed infrastructure 200 assigns policy 520 to that node 400. For example, when the communication status of a node 400 that previously did not satisfy a condition now satisfies the condition, the distributed infrastructure 200 assigns policy 510 to that node 400. These can contribute to optimizing performance in response to changes in the communication status.

[0041] 4 is an explanatory diagram for explaining the reassignment of a policy by the distributed infrastructure 200. Here, the situation shown in FIG. 3 is illustrated in which the communication status of one of the three nodes 400 that previously satisfied the conditions no longer satisfies the conditions. The distributed infrastructure 200 assigns a policy 520 to the node 400.

[0042] For two nodes 400 to which policy 510 has been assigned, the distributed infrastructure 200 inputs data received from each of the two nodes 400 into the learning model 610 and transmits data output from the learning model 610 to the two nodes 400. The two nodes 400 execute processing in accordance with the data received from the distributed infrastructure 200. For three nodes 400 to which policy 520 has been assigned, the distributed infrastructure 200 inputs only its own data into the learning model 620 and executes processing in accordance with the data output from the learning model 620.

[0043] The distributed infrastructure 200 may generate and store in advance a learning model 610 for each combination of multiple nodes 400. Then, the distributed infrastructure 200 may use, from among the multiple learning models 610, a learning model 610 that corresponds to the combination of nodes 400 to which the policy 510 is assigned.

[0044] Taking an example in which the multiple nodes 400 correspond to the multiple joints of a robot, the distribution platform 200 generates and stores a learning model 610 in advance for each combination of the multiple joints. As a specific example, a learning model 610 corresponding to the multiple joints of the right arm (sometimes referred to as a right arm learning model) and a learning model 610 corresponding to the multiple joints of both arms (sometimes referred to as a both arm learning model) are stored. When a policy 510 is assigned to the multiple nodes 400 corresponding to the multiple joints of the right arm, the right arm learning model is used. When the communication situation changes and the policy 510 is also assigned to the multiple nodes 400 corresponding to the multiple joints of the left arm, the both arm learning model is used. This makes it possible to realize centralized execution using a learning model suitable for a combination of the multiple nodes 400 whose communication situation satisfies certain conditions.

[0045] 5 is an explanatory diagram for explaining the reassignment of a policy by the distributed infrastructure 200. Here, the situation shown in FIG. 3 is illustrated in which the communication situation of one of the two nodes 400 whose communication situation did not satisfy the condition has now become one that satisfies the condition. The distributed infrastructure 200 assigns a policy 510 to the node 400.

[0046] For the four nodes 400 to which policy 510 has been assigned, the distributed infrastructure 200 inputs data received from each of the four nodes 400 into the learning model 610 and transmits data output from the learning model 610 to the four nodes 400. The four nodes 400 execute processing in accordance with the data received from the distributed infrastructure 200. For the one node 400 to which policy 520 has been assigned, the distributed infrastructure 200 inputs only its own data into the learning model 620 and executes processing in accordance with the data output from the learning model 620.

[0047] 6 shows an example of the functional configuration of the distributed infrastructure 200. The distributed infrastructure 200 includes a storage unit 222, a learning data acquisition unit 224, a learning model generation unit 226, a learning model generation unit 228, a learning model transmission unit 230, a communication status acquisition unit 234, a centralized execution control unit 236, a distributed execution control unit 238, and a RAN control unit 240. Note that it is not essential for the distributed infrastructure 200 to include all of these units.

[0048] The learning data acquisition unit 224 acquires learning data and stores the acquired learning data in the storage unit 222.

[0049] The learning data acquisition unit 224 may acquire learning data for generating the learning model 610. The learning data acquisition unit 224 may acquire learning data for performing intensive learning of multi-agent reinforcement learning for a plurality of nodes 400.

[0050] The learning data acquiring unit 224 may acquire learning data for generating the learning model 620. The learning data acquiring unit 224 may acquire learning data for each of the multiple nodes 400 for individual learning.

[0051] The learning model generation unit 226 generates a learning model 610 using the learning data stored in the storage unit 222. The learning model generation unit 226 stores the generated learning model 610 in the storage unit 222. The learning model generation unit 226 may be an example of a first learning model generation unit. The learning model generation unit 226 may generate the learning model 610 in accordance with the policy of the CTCE.

[0052] The learning model generation unit 226 may generate the learning model 610 by treating the multiple nodes 400 as multi-agents, collectively inputting the observations of each agent into a single learning model 610 that outputs a policy of each agent, and performing reinforcement learning that gives a reward for a desired action. Generating the learning model 610 may mean generating a new learning model 610 by updating the learning model 610. Generating the learning model 610 may mean generating the learning model 610 from scratch. The learning model generation unit 226 may generate a learning model 610 for each combination of multiple nodes 400.

[0053] The learning model generation unit 228 generates a learning model 620 using the learning data stored in the storage unit 222. The learning model generation unit 228 stores the generated learning model 620 in the storage unit 222. The learning model generation unit 228 may be an example of a second learning model generation unit. The learning model generation unit 228 may generate the learning model 620 in accordance with the policy of the CTDE.

[0054] The learning model generation unit 228 may generate the learning model 620 by treating the multiple nodes 400 as multi-agents, inputting observations of each agent individually into the learning model 620, and performing reinforcement learning that provides a reward for a desired behavior. Generating the learning model 620 may mean generating a new learning model 620 by updating the learning model 620. Generating the learning model 620 may mean generating the learning model 620 from scratch. The learning model generation unit 226 may generate a learning model 620 for each of the multiple nodes 400.

[0055] The learning model transmission unit 230 transmits the learning model 620 generated by the learning model generation unit 226 to the plurality of nodes 400. The learning model transmission unit 230 may transmit the learning model 620 generated for each of the plurality of nodes 400 to each of the plurality of nodes 400.

[0056] The communication status acquisition unit 234 acquires the communication status between each of the plurality of nodes 400 and the distributed infrastructure 200. The communication status acquisition unit 234 may acquire the communication status of communication between each of the plurality of nodes 400 and the distributed infrastructure 200 via the RAN 310.

[0057] The communication status acquisition unit 234 may acquire, according to a predetermined schedule, the communication status between each of the plurality of nodes 400 and the distributed infrastructure 200. For example, the communication status acquisition unit 234 periodically acquires the communication status between each of the plurality of nodes 400 and the distributed infrastructure 200 according to a predetermined cycle.

[0058] The centralized execution control unit 236 inputs data acquired from each of the plurality of nodes 400 for which the communication status acquired by the communication status acquisition unit 234 satisfies a predetermined condition into the learning model 610, and transmits data output from the learning model 610 to the plurality of nodes 400. The centralized execution control unit 236 may input data acquired from each of the plurality of nodes 400 for which the degree of goodness of the communication status with the distributed infrastructure 200 is greater than a predetermined degree into the learning model 610, and transmit the data output from the learning model 610 to the plurality of nodes 400.

[0059] The distributed execution control unit 238 inputs data of the nodes 400 into a pre-stored learning model 620 for multiple nodes 400 for which the communication status acquired by the communication status acquisition unit 234 does not satisfy a predetermined condition, and causes the multiple nodes 400 to execute processing in accordance with the data output from the learning model 620. The distributed execution control unit 238 may input data of the nodes 400 into the learning model 620 for multiple nodes 400 for which the degree of goodness of the communication status with the distributed infrastructure 200 is lower than a predetermined degree, and cause the multiple nodes 400 to execute processing in accordance with the data output from the learning model 620.

[0060] The RAN control unit 240 executes a RAN control function that controls the RAN 310 configured by a plurality of radio base stations 300. The RAN control unit 240 may execute a function of a so-called vRAN (Virtual RAN). The RAN control unit 240 controls communications in the RAN 310 by controlling the plurality of radio base stations 300 that configure the RAN 310 and coordinating with other distributed infrastructures 200.

[0061] 7 shows an example of the flow of processing by the distributed infrastructure 200. Here, the description will be given assuming that the distributed infrastructure 200 stores a learning model 610 and each of the multiple nodes 400 stores a learning model 620 as a starting state.

[0062] In step (sometimes abbreviated as S) 102, the communication status acquisition unit 234 acquires the communication status between each of the plurality of nodes 400 and the distributed infrastructure 200. In S104, the communication status acquisition unit 234 assigns a policy to each of the plurality of nodes 400. The communication status acquisition unit 234 assigns policy 510 to the plurality of nodes 400 whose communication status satisfies predetermined conditions, and assigns policy 520 to the plurality of nodes 400 whose communication status does not satisfy the predetermined conditions.

[0063] In S106, the centralized execution control unit 236 and the distributed execution control unit 238 start execution control. The centralized execution control unit 236 starts a process of acquiring data from each of the multiple nodes 400 to which the policy 510 is assigned, inputting the acquired data into the learning model 610, and transmitting data output from the learning model 610 to the multiple nodes 400. If the centralized execution control unit 236 stores a learning model 610 for each combination of nodes 400, it may use the learning model 610 corresponding to the multiple nodes 400 to which the policy 510 is assigned. The distributed execution control unit 238 causes one or more nodes 400 to which the policy 520 is assigned to input their own data into the pre-stored learning model 620 and start processing in accordance with the data output from the learning model 620.

[0064] In S108, the communication status acquisition unit 234 acquires the communication status between each of the multiple nodes 400 and the distributed infrastructure 200. In S110, the communication status acquisition unit 234 determines whether or not there is a node 400 for which a policy should be changed. The communication status acquisition unit 234 determines that there is a node 400 for which a policy should be changed if the communication status of the node 400 to which policy 510 was assigned no longer satisfies the conditions, or if the communication status of the node 400 to which policy 520 was assigned now satisfies the conditions. If it is determined that there is a node 400 for which a policy should be changed, the communication status acquisition unit 234 reallocates the policy. Then, the process proceeds to S112. If it is determined that there is no node 400 for which a policy should be changed, the process proceeds to S114.

[0065] In S112, the centralized execution control unit 236 and the distributed execution control unit 238 perform change control. After the policy change, the centralized execution control unit 236 acquires data from each of the multiple nodes 400 to which the policy 510 is assigned, inputs the acquired data into the learning model 610, and starts processing to transmit data output from the learning model 610 to the multiple nodes 400. If the centralized execution control unit 236 stores a learning model 610 for each combination of nodes 400, it may use the learning model 610 corresponding to the multiple nodes 400 to which the policy 510 is assigned after the policy change. After the policy change, the distributed execution control unit 238 causes one or more nodes 400 to which the policy 520 is assigned to input their own data into the pre-stored learning model 620 and start processing in accordance with the data output from the learning model 620.

[0066] If the process is not to be ended (NO in S114), the process returns to S108, and if the process is to be ended (YES in S114), the process ends.

[0067] 8 is an explanatory diagram for explaining a method for managing a plurality of nodes 400 by the management board 100. Here, differences from FIG. 2 will be mainly explained.

[0068] The management infrastructure 100 acquires, for each of the plurality of distributed infrastructures 200 , the communication status between the distributed infrastructure 200 and each of the plurality of nodes 400 , and the communication status between the distributed infrastructure 200 and the management infrastructure 100 .

[0069] The management infrastructure 100 may assign a policy 510 to one or more nodes 400 among the multiple nodes 400, for which the communication status between the node 400, the distributed infrastructure 200, and the management infrastructure 100 satisfies a predetermined condition. The management infrastructure 100 may determine that the communication status between the node 400, the distributed infrastructure 200, and the management infrastructure 100 satisfies the predetermined condition when the communication status between the node 400 and the distributed infrastructure 200 and the management infrastructure 100 satisfies the predetermined condition, and may otherwise determine that the communication status between the node 400, the distributed infrastructure 200, and the management infrastructure 100 does not satisfy the predetermined condition. The management infrastructure 100 may assign a policy 520 to one or more nodes 400 among the multiple nodes 400, for which the communication status between the node 400, the distributed infrastructure 200, and the management infrastructure 100 does not satisfy the predetermined condition.

[0070] In the example shown in Figure 8, of the five nodes 400, the communication status of two nodes 400 meets the conditions, and policy 510 is assigned to those two nodes 400, while the communication status of three nodes 400 does not meet the conditions, and policy 520 is assigned to those three nodes 400.

[0071] 9 is an explanatory diagram for explaining a method for managing a plurality of nodes 400 by the management board 100 in accordance with the policies 510 and 520. Here, differences from FIG. 3 will be mainly explained.

[0072] 9 , for two nodes 400 to which policy 510 is assigned, the management infrastructure 100 inputs data received from each of the two nodes 400 into the learning model 610 and transmits data output from the learning model 610 to the two nodes 400. The two nodes 400 execute processing in accordance with the data received from the management infrastructure 100. For three nodes 400 to which policy 520 is assigned, the management infrastructure 100 inputs only its own data into the learning model 620 and executes processing in accordance with the data output from the learning model 620. The management infrastructure 100 may periodically acquire communication statuses of multiple nodes 400 and reassign policies in accordance with changes in the communication status.

[0073] 10 shows an example of the functional configuration of the management infrastructure 100. The management infrastructure 100 includes a memory unit 122, a learning data acquisition unit 124, a learning model generation unit 126, a learning model generation unit 128, a learning model transmission unit 130, a communication status acquisition unit 134, a centralized execution control unit 136, a distributed execution control unit 138, and a RAN control unit 140. Note that it is not essential for the management infrastructure 100 to include all of these units.

[0074] The learning data acquisition unit 124 acquires learning data and stores the acquired learning data in the storage unit 122.

[0075] The learning data acquiring unit 124 may acquire learning data for generating the learning model 610. The learning data acquiring unit 124 may acquire learning data for performing intensive learning of multi-agent reinforcement learning for a plurality of nodes 400.

[0076] The learning data acquiring unit 124 may acquire learning data for generating the learning model 620. The learning data acquiring unit 124 may acquire learning data for each of the multiple nodes 400 for individual learning.

[0077] The learning model generation unit 126 generates a learning model 610 using the learning data stored in the storage unit 122. The learning model generation unit 126 stores the generated learning model 610 in the storage unit 122. The learning model generation unit 126 may generate the learning model 610 in accordance with the policy of the CTCE.

[0078] The learning model generation unit 126 may generate the learning model 610 by treating the multiple nodes 400 as multi-agents, collectively inputting the observations of each agent into a single learning model 610 that outputs a policy of each agent, and performing reinforcement learning that gives a reward for a desired action. Generating the learning model 610 may mean generating a new learning model 610 by updating the learning model 610. Generating the learning model 610 may mean generating the learning model 610 from scratch. The learning model generation unit 126 may generate a learning model 610 for each combination of multiple nodes 400.

[0079] The learning model generation unit 128 generates a learning model 620 using the learning data stored in the storage unit 122. The learning model generation unit 128 stores the generated learning model 620 in the storage unit 122. The learning model generation unit 128 may generate the learning model 620 in accordance with the policy of the CTDE.

[0080] The learning model generation unit 128 may generate the learning model 620 by treating the multiple nodes 400 as multi-agents, inputting observations of each agent individually into the learning model 620, and performing reinforcement learning that provides a reward for a desired behavior. Generating the learning model 620 may mean generating a new learning model 620 by updating the learning model 620. Generating the learning model 620 may mean generating the learning model 620 from scratch. The learning model generation unit 126 may generate a learning model 620 for each of the multiple nodes 400.

[0081] The learning model transmission unit 130 transmits the learning model 620 generated by the learning model generation unit 126 to the plurality of nodes 400. The learning model transmission unit 130 may transmit the learning model 620 generated for each of the plurality of nodes 400 to each of the plurality of nodes 400.

[0082] The communication status acquisition unit 134 acquires, for each of the multiple distributed infrastructures 200, the communication status between the distributed infrastructure 200 and each of the multiple nodes 400, and the communication status between the distributed infrastructure 200 and the management infrastructure 100. The communication status acquisition unit 134 may acquire the communication status of communication between the distributed infrastructure 200 and each of the multiple nodes 400 via the RAN 310.

[0083] The communication status acquisition unit 134 may acquire, for each of the multiple distributed infrastructures 200, the communication status between the distributed infrastructure 200 and each of the multiple nodes 400, and the communication status between the distributed infrastructure 200 and the management infrastructure 100, according to a predetermined schedule. For example, the communication status acquisition unit 134 periodically acquires these communication statuses according to a predetermined cycle.

[0084] The centralized execution control unit 136 inputs data acquired from each of the plurality of nodes 400 for which the communication status between the node 400, the distributed infrastructure 200, and the management infrastructure 100 satisfies a predetermined condition into the learning model 610, and transmits data output from the learning model 610 to the plurality of nodes 400. The centralized execution control unit 136 may input data acquired from each of the plurality of nodes 400 for which the degree of goodness of the communication status between the node 400, the distributed infrastructure 200, and the management infrastructure 100 is greater than a predetermined degree into the learning model 610, and transmits data output from the learning model 610 to the plurality of nodes 400.

[0085] The distributed execution control unit 138 causes, among the plurality of nodes 400, a plurality of nodes 400 for which the communication conditions between the node 400, the distributed infrastructure 200, and the management infrastructure 100 do not satisfy predetermined conditions to input data of the node 400 into a pre-stored learning model 620 and execute processing in accordance with the data output from the learning model 620. The distributed execution control unit 138 may cause, among the plurality of nodes 400, a plurality of nodes 400 for which the degree of goodness of the communication conditions between the node 400, the distributed infrastructure 200, and the management infrastructure 100 is less than a predetermined degree to input data of the node 400 into the learning model 620 and execute processing in accordance with the data output from the learning model 620.

[0086] The RAN control unit 140 executes a RAN control function for controlling the RAN 310. The RAN control unit 140 may control the RAN 310 in cooperation with the RAN control units 240 of the multiple distribution infrastructures 200.

[0087] 11 schematically illustrates an example of the hardware configuration of a computer 1200 that functions as the management infrastructure 100 or the distribution infrastructure 200. A program installed on the computer 1200 can cause the computer 1200 to function as one or more "parts" of an apparatus according to the present embodiment, or can cause the computer 1200 to execute operations associated with the apparatus according to the present embodiment or one or more "parts," and / or can cause the computer 1200 to execute a process according to the present embodiment or steps of the process. Such a program can be executed by the CPU 1212 to cause the computer 1200 to perform specific operations associated with some or all of the blocks in the flowcharts and block diagrams described herein.

[0088] The computer 1200 according to this embodiment includes a CPU 1212, a GPU 1213, a RAM 1214, and a graphics controller 1216, which are interconnected by a host controller 1210. The computer 1200 also includes input / output units such as a communications interface 1222, a storage device 1224, a DVD drive 1226, and an IC card drive, which are connected to the host controller 1210 via an input / output controller 1220. The DVD drive 1226 may be a DVD-ROM drive, a DVD-RAM drive, or the like. The storage device 1224 may be a hard disk drive, a solid-state drive, or the like. The computer 1200 also includes a ROM 1230 and legacy input / output units such as a keyboard, which are connected to the input / output controller 1220 via an input / output chip 1240.

[0089] The CPU 1212 operates according to programs stored in the ROM 1230 and the RAM 1214, thereby controlling each unit. The graphics controller 1216 acquires image data generated by the CPU 1212 into a frame buffer or the like provided in the RAM 1214 or into the graphics controller 1216 itself, and causes the image data to be displayed on the display device 1218.

[0090] The communication interface 1222 communicates with other electronic devices via a network. The storage device 1224 stores programs and data used by the CPU 1212 in the computer 1200. The DVD drive 1226 reads programs or data from a DVD-ROM 1227 or the like and provides them to the storage device 1224. The IC card drive reads programs and data from an IC card and / or writes programs and data to an IC card.

[0091] The ROM 1230 stores therein a boot program or the like that is executed by the computer 1200 upon activation, and / or programs that depend on the hardware of the computer 1200. The input / output chip 1240 may also connect various input / output units to the input / output controller 1220 via a USB port, a parallel port, a serial port, a keyboard port, a mouse port, etc.

[0092] The programs are provided by a computer-readable storage medium such as a DVD-ROM 1227 or an IC card. The programs are read from the computer-readable storage medium, installed in the storage device 1224, RAM 1214, or ROM 1230, which are also examples of computer-readable storage media, and executed by the CPU 1212. Information processing described in these programs is read by the computer 1200, and brings about cooperation between the programs and the various types of hardware resources described above. An apparatus or a method may be configured by implementing operations or processing of information in accordance with the use of the computer 1200.

[0093] For example, when communication is performed between computer 1200 and an external device, CPU 1212 may execute a communication program loaded into RAM 1214 and instruct communication interface 1222 to perform communication processing based on the processing described in the communication program. Under the control of CPU 1212, communication interface 1222 reads transmission data stored in a transmission buffer area provided in RAM 1214, storage device 1224, DVD-ROM 1227, or a recording medium such as an IC card, and transmits the read transmission data to a network, or writes received data received from the network to a reception buffer area or the like provided on the recording medium.

[0094] Furthermore, the CPU 1212 may cause all or a necessary portion of a file or database stored in an external recording medium such as the storage device 1224, the DVD drive 1226 (DVD-ROM 1227), an IC card, etc. to be read into the RAM 1214, and may perform various types of processing on the data on the RAM 1214. The CPU 1212 may then write back the processed data to the external recording medium.

[0095] Various types of information, such as various types of programs, data, tables, and databases, may be stored on the recording medium and may undergo information processing. The CPU 1212 may perform various types of processing on data read from the RAM 1214, including various types of operations, information processing, conditional judgment, conditional branching, unconditional branching, information search / replacement, etc., as described throughout this disclosure and specified by the instruction sequences of the programs, and write the results back to the RAM 1214. The CPU 1212 may also search for information in a file, database, etc. on the recording medium. For example, if multiple entries, each having an attribute value of a first attribute associated with an attribute value of a second attribute, are stored on the recording medium, the CPU 1212 may search for an entry whose attribute value of the first attribute matches a specified condition from among the multiple entries, read the attribute value of the second attribute stored in the entry, and thereby obtain the attribute value of the second attribute associated with the first attribute that satisfies a predetermined condition.

[0096] The above-described programs or software modules may be stored in a computer-readable storage medium on or near the computer 1200. A recording medium such as a hard disk or RAM provided in a server system connected to a dedicated communication network or the Internet can also be used as a computer-readable storage medium, thereby providing the programs to the computer 1200 via the network.

[0097] The blocks in the flowcharts and block diagrams in the present embodiments may represent stages of a process in which an operation is performed or "parts" of a device responsible for performing the operation. Particular stages and "parts" may be implemented by dedicated circuitry, programmable circuitry provided with computer-readable instructions stored on a computer-readable storage medium, and / or a processor provided with computer-readable instructions stored on a computer-readable storage medium. The dedicated circuitry may include digital and / or analog hardware circuits, and may include integrated circuits (ICs) and / or discrete circuits. The programmable circuitry may include reconfigurable hardware circuits, such as field programmable gate arrays (FPGAs) and programmable logic arrays (PLAs), including AND, OR, XOR, NAND, NOR, and other logical operations, flip-flops, registers, and memory elements.

[0098] A computer-readable storage medium may include any tangible device capable of storing instructions that are executed by an appropriate device, such that a computer-readable storage medium having instructions stored thereon comprises an article of manufacture, including instructions that can be executed to create means for performing the operations specified in the flowcharts or block diagrams. Examples of computer-readable storage media may include electronic, magnetic, optical, electromagnetic, and semiconductor storage media. More specific examples of computer-readable storage media may include floppy disks, diskettes, hard disks, random access memories (RAMs), read-only memories (ROMs), erasable programmable read-only memories (EPROMs or flash memories), electrically erasable programmable read-only memories (EEPROMs), static random access memories (SRAMs), compact disc read-only memories (CD-ROMs), digital versatile discs (DVDs), Blu-ray discs, memory sticks, integrated circuit cards, and the like.

[0099] The computer readable instructions may include either assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state-setting data, or source or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk®, JAVA®, C++, etc., and conventional procedural programming languages ​​such as the “C” programming language or similar programming languages.

[0100] The computer-readable instructions may be provided to a general-purpose computer, a special-purpose computer, or another programmable data processing device processor or programmable circuit, either locally or via a local area network (LAN), a wide area network (WAN) such as the Internet, so that the processor or programmable circuit of the programmable data processing device, such as a computer, executes the computer-readable instructions to generate means for performing the operations specified in the flowcharts or block diagrams. Here, the computer may be a personal computer (PC), a tablet computer, a smartphone, a workstation, a server computer, a general-purpose computer, a special-purpose computer, or the like, or may be a computer system in which multiple computers are connected. Such a computer system in which multiple computers are connected is also called a distributed computing system, and is a broad definition of computer. In a distributed computing system, multiple computers collectively execute a program by each executing a portion of the program and passing data between the computers as needed during program execution.

[0101] Examples of processors include computer processors, central processing units (CPUs), processing units, microprocessors, digital signal processors, controllers, microcontrollers, etc. A computer may have one processor or multiple processors. In a multiprocessor system with multiple processors, each processor executes a portion of a program and passes data between processors as needed during program execution, allowing the multiple processors to collectively execute the program. For example, in multitasking, each of the multiple processors may execute a portion of each task in small chunks by switching tasks at each time slice. In this case, which portion of a program each processor executes changes dynamically. Which portion of a program each of the multiple processors executes may also be statically determined by multiprocessor-aware programming.

[0102] Although the present invention has been described above using the embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various modifications and improvements can be made to the above embodiments. It is clear from the claims that such modifications and improvements can also be included within the technical scope of the present invention.

[0103] It should be noted that the execution order of each process, such as operations, procedures, steps, and stages, in the devices, systems, programs, and methods shown in the claims, specifications, and drawings is not specifically stated as "before," "prior to," etc., and that the processes can be performed in any order unless the output of a previous process is used in a subsequent process. Even if the operational flow in the claims, specifications, and drawings is described using "first," "next," etc. for convenience, this does not mean that the processes must be performed in this order.

[0104] 10 System, 30 UE, 100 Management base station, 122 Memory unit, 124 Learning data acquisition unit, 126 Learning model generation unit, 128 Learning model generation unit, 130 Learning model transmission unit, 134 Communication status acquisition unit, 136 Centralized execution control unit, 138 Distributed execution control unit, 140 RAN control unit, 200 Distribution base station, 222 Memory unit, 224 Learning data acquisition unit, 226 Learning model generation unit, 228 Learning model generation unit, 230 Learning model transmission unit, 234 Communication status acquisition unit, 236 Centralized execution control unit, 238 Distributed execution control unit, 240 RAN control unit, 300 Radio base station, 310 RAN, 400 Node, 510 Policy, 520 Policy, 610 Learning model, 620 Learning model, 1200 Computer, 1210 Host controller, 1212 CPU, 1213 GPU, 1214 RAM, 1216 graphics controller, 1218 display device, 1220 input / output controller, 1222 communication interface, 1224 storage device, 1226 DVD drive, 1227 DVD-ROM, 1230 ROM, 1240 input / output chip

Claims

1. An execution platform comprising: a communication status acquisition unit that acquires communication status between each of a plurality of nodes and the execution platform; a centralized execution control unit that inputs data acquired from each of a plurality of nodes among the plurality of nodes whose communication status satisfies predetermined conditions into a first learning model and transmits data output from the first learning model to the plurality of nodes; and a distributed execution control unit that inputs data from a plurality of nodes among the plurality of nodes whose communication status does not satisfy the predetermined conditions into a second learning model stored in advance and causes the plurality of nodes to execute processing in accordance with the data output from the second learning model.

2. The execution platform described in claim 1, comprising: a first learning model generation unit that receives data from multiple nodes as input and generates the first learning model that outputs data for causing the multiple nodes to operate cooperatively; a second learning model generation unit that receives data from one node as input and generates the second learning model that outputs data for causing the one node to operate; and a learning model transmission unit that transmits the second learning model to the multiple nodes.

3. The execution platform described in claim 2, wherein the first learning model generation unit generates the first learning model in accordance with a policy based on CTCE (Centralized Training Centralized Execution), and the second learning model generation unit generates the second learning model in accordance with a policy based on CTDE (Centralized Training Decentralized Execution).

4. The execution platform described in claim 2 or 3, wherein the first learning model generation unit generates the first learning model for each combination of the plurality of nodes, and the centralized execution control unit inputs data obtained from each of the plurality of nodes whose communication conditions satisfy predetermined conditions into the first learning model corresponding to the plurality of nodes, and transmits the data output from the first learning model to the plurality of nodes.

5. An execution platform described in any one of claims 1 to 4, wherein the centralized execution control unit inputs data obtained from each of a plurality of nodes among the plurality of nodes whose communication conditions with the execution platform have a degree of goodness greater than a predetermined degree into the first learning model and transmits data output from the first learning model to the plurality of nodes, and the distributed execution control unit inputs data from the node into the second learning model for a plurality of nodes among the plurality of nodes whose communication conditions with the execution platform have a degree of goodness less than the predetermined degree and causes the nodes to execute processing in accordance with the data output from the second learning model.

6. An execution platform as described in any one of claims 1 to 5, further comprising: a RAN control unit that executes a RAN control function that controls a RAN (Radio Access Network) consisting of a plurality of radio base stations; and the communication status acquisition unit acquires the communication status of communication via the RAN between each of the plurality of nodes and the execution platform.

7. A program that, when executed by a computer, causes the computer to execute the following steps: a communication status acquisition step of acquiring communication status between each of a plurality of nodes and the computer; a centralized execution control step of inputting data acquired from each of a plurality of nodes whose communication status satisfies predetermined conditions into a first learning model and transmitting data output from the first learning model to the plurality of nodes; and a distributed execution control step of inputting data from a plurality of nodes whose communication status does not satisfy the predetermined conditions into a second learning model stored in advance and causing the plurality of nodes whose communication status does not satisfy the predetermined conditions to execute processing in accordance with the data output from the second learning model.

8. A control method executed by a computer, comprising: a communication status acquisition step of acquiring communication status between each of a plurality of nodes and the computer; a centralized execution control step of inputting data acquired from each of a plurality of nodes whose communication status satisfies predetermined conditions into a first learning model and transmitting data output from the first learning model to the plurality of nodes; and a distributed execution control step of inputting data from the node into a second learning model stored in advance for a plurality of nodes whose communication status does not satisfy the predetermined conditions and causing the nodes to execute processing in accordance with the data output from the second learning model.

9. A system comprising: a plurality of distributed infrastructures; and a management infrastructure that manages the plurality of distributed infrastructures, wherein the management infrastructure has: a communication status acquisition unit that acquires, for each of the plurality of distributed infrastructures, the communication status between the distributed infrastructure and each of a plurality of nodes, and the communication status between the distributed infrastructure and the management infrastructure; a centralized execution control unit that inputs data acquired from each of a plurality of nodes among the plurality of nodes, for which the communication status between the node, the distributed infrastructure, and the management infrastructure satisfies predetermined conditions, into a first learning model and transmits data output from the first learning model to the plurality of nodes; and a distributed execution control unit that, for a plurality of nodes among the plurality of nodes, for which the communication status between the node, the distributed infrastructure, and the management infrastructure does not satisfy the predetermined conditions, inputs the data of the node into a second learning model that is pre-stored, and causes processing in accordance with the data output from the second learning model.

Images