Building a Federated Learning Framework

Abstract

Embodiments relate to an intelligent computer platform for building a federated learning framework, which includes creating a hierarchy of machine learning models (MLMs). The hierarchy of MLMs has a primary MLM in a primary layer. Training the primary MLM includes capturing contribution model updates through at least one communication channel. A secondary MLM is created and logically placed in a secondary layer of the hierarchy. The secondary MLM is operably coupled to the primary MLM through at least one communication channel. Initializing the created secondary MLM includes cloning the weights and framework of the primary MLM into the secondary MLM, and populating the created secondary MLM with secondary data. The populated data has model updates local to the created secondary MLM. The secondary MLM is logically stored locally in the secondary layer, and access to the secondary MLM is restricted to the secondary layer.

Description

Background Art

[0001] The present embodiment relates to an artificial intelligence platform and optimization method for creating and supporting a federated learning framework. The framework represents a hierarchical layering of learning models, and the embodiment relates to training and maintaining neural models within the hierarchy, and utilizing the models in response to the hierarchy. Summary of the invention

[0002] Embodiments include computer systems, computer program products, and methods for building and maintaining a federated learning framework.

[0003] In one aspect, a computer system having a processor and a memory is provided for use with an artificial intelligence (AI) platform processing tool. The processor is operably coupled to the memory and communicates with the AI ​​platform. The AI ​​platform tool includes a hierarchy manager, a training manager, and an MLM manager. The hierarchy manager is used to build a joint learning framework including a hierarchy, more specifically, to create a hierarchy of machine learning models (MLMs). The hierarchy includes a primary MLM in a primary layer of the hierarchy. The hierarchy manager is also used to create a secondary MLM logically located in a secondary layer of the hierarchy. The secondary MLM is operably coupled to the primary MLM through at least one communication channel. The training manager is used to train the primary MLM, which includes capturing contribution model updates through at least one communication channel. The training manager is used to initialize the created secondary MLM, including cloning the weights and framework of the primary MLM into the secondary MLM, and filling the created secondary MLM with secondary data. The populated data includes model updates to the local (local) of the created secondary MLM. The MLM manager is used to logically place the secondary MLM locally in the secondary layer, restrict access to the secondary MLM in the secondary layer, and logically place the primary MLM data globally. The secondary MLM can access the primary MLM.

[0004] On the other hand, a computer program device is provided to construct a joint learning framework. The computer program product includes a computer-readable storage medium having a program code contained thereon. The program code can be executed by a processor to construct a joint learning framework, including program code for creating a hierarchy of machine learning models (MLMs). The hierarchy of MLMs has a primary MLM logically located in a primary layer of the hierarchy. The program code trains the primary MLM, including capturing contribution model updates through at least one communication channel. Similarly, the program code creates a secondary MLM logically located in a secondary layer of the hierarchy. The secondary MLM is operably coupled to the primary MLM through at least one communication channel. In addition, the program code initializes the created secondary MLM, which includes cloning the weights and framework of the primary MLM into the secondary MLM and filling the created secondary MLM with secondary data. The populated data includes model updates to the created secondary MLM local. In addition, the program code logically stores the secondary MLM locally in the secondary layer and restricts access to the secondary layer MLM to members of the secondary group local to the secondary layer. The program code logically and globally stores the primary MLM and makes the primary MLM accessible to the secondary MLM.

[0005] On the other hand, a method for building a federated learning framework is provided, comprising creating a hierarchy of machine learning models (MLMs). The hierarchy of MLMs has a primary MLM logically located in a primary layer. Training the primary MLM includes capturing contribution model updates through at least one communication channel. A secondary MLM is created and logically placed in a secondary layer of the hierarchy. The secondary MLM is operably coupled to the primary MLM through at least one communication channel. Initializing the created secondary MLM includes cloning the weights and framework of the primary MLM into the secondary MLM and populating the created secondary MLM with secondary data. The populated data exists locally in the created secondary MLM in the form of model updates. The secondary MLM is logically stored locally in the secondary layer, and access to the secondary MLM is limited to the secondary layer.

[0006] These and other features and advantages will become apparent from the following detailed description of the presently preferred embodiments, taken in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] The accompanying drawings referred to herein form a part of the specification. Unless expressly stated otherwise, the features shown in the accompanying drawings are intended to illustrate some embodiments only, not all embodiments.

[0008] Figure 1A system diagram illustrating a computer system is shown;

[0009] Figure 2 Shows the description as Figure 1 A block diagram and associated node hierarchy of an exemplary federated learning framework for a computer system as shown and described;

[0010] Figure 3 A block diagram illustrating a tool from a computing system and its associated application program interface is shown;

[0011] Figure 4 A flowchart illustrating construction of a federated learning framework and creation of a secondary MLM is shown;

[0012] Figure 5 A flow chart illustrating training a local model by first determining the sufficiency of training data is shown;

[0013] Figure 6 An example of a computer system / server illustrating a cloud-based support system is shown to implement the above-mentioned Figures 1 to 5 A block diagram of the systems and processes described in;

[0014] Figure 7 A block diagram illustrating a cloud computer environment is shown;

[0015] Figure 8 A block diagram illustrating a set of functional abstraction model layers provided by a cloud computing environment is shown. DETAILED DESCRIPTION

[0016] It will be readily understood that, as generally described and illustrated in the drawings herein, the components of the embodiments of the present invention may be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the apparatus, system, method, and computer program product of the present embodiments, as illustrated in the drawings, is not intended to limit the scope of the claimed embodiments, but is merely representative, selected embodiments.

[0017] Throughout the specification, references to "a select embodiment," "one embodiment," or "an embodiment" mean that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Thus, the appearances of the phrases "a select embodiment," "in one embodiment," or "in an embodiment" in various places throughout this specification are not necessarily referring to the same embodiment.

[0018] The illustrated embodiments will be best understood by referring to the drawings, wherein like parts are represented by like numerals throughout.The following description is intended to be exemplary only and simply illustrates certain selected embodiments of apparatuses, systems, and processes consistent with the embodiments claimed herein.

[0019] Artificial Intelligence (AI) is a field of computer science that deals with computers and computer behavior in relation to humans. AI refers to the intelligence of machines when they are able to make decisions based on information, which can maximize the chances of success on a given subject. More specifically, AI is able to learn from data sets to solve problems and provide relevant recommendations. For example, in the field of artificial intelligence computer systems, natural language systems (such as Artificial intelligence computer systems (or other natural language query answering systems) process natural language based on the knowledge acquired by the system. To process natural language, the system can be trained using data obtained from a database or knowledge base, but the results produced may be incorrect or inaccurate for a variety of reasons.

[0020] Machine Learning (ML) is a subset of Artificial Intelligence (AI) that utilizes algorithms to learn from data and create predictions based on that data. More specifically, ML is the application of AI by creating neural networks that can demonstrate learning behavior by performing tasks that are not explicitly programmed. Deep learning is a type of ML in which the system can complete complex tasks. Deep learning is a method of ML that incorporates neural networks in successive layers to learn from data in an iterative manner. Neural networks are models of how nervous systems work. The basic units are called neurons and are usually organized into layers. Neural networks work by simulating a large number of interconnected processing units that resemble abstract versions of neurons. Neural networks typically have three parts, including: an input layer with units that represent the input domain; one or more hidden layers; and an output layer with one or more units that represent the target domain. These units are connected with different connection strengths or weights. Input data is presented to the first layer, and values ​​are propagated from each neuron to each neuron in the next layer. Ultimately, the results are delivered from the output layer. Deep Learning Complex neural networks are designed to simulate the way the human brain works, so computers can be trained to support poorly defined abstractions and problems. Neural networks and deep learning are often used in image recognition, speech, and computer vision applications.

[0021] Regarding personalized local models, the heterogeneity of data from different groups and the immediate response required in online learning settings can affect the effectiveness of AI training. For example, data heterogeneity may be caused by using different languages ​​or different session types in the group. A large number of participants (clients / accounts) can be segmented into different groups, for example, different projects and languages. Therefore, the need for specific online learning and data privacy may increase.

[0022] Privacy is the ability of an individual to determine what data can be shared and to adopt access controls. If data is in the public domain, it poses a threat to individual privacy because the data is kept by the data holder. Examples of data holders can include public or semi-public places such as, but not limited to, social networking applications, websites, mobile applications, e-commerce sites, banks, and hospitals. In general, it is the responsibility of the data holder to ensure the privacy of user data. In addition to data kept in the public domain, users themselves can also cause data leakage, knowingly or unknowingly. For example, many mobile applications seek access to our contacts, files, and cameras. If terms and conditions are not chosen, opportunities for data leakage will abound. There are a variety of techniques that can be used to protect data privacy. These techniques include, but are not limited to, k-anonymity, l-diversity, t-closeness, randomization, data distribution, cryptographic methods, and anonymization based on multidimensional sensitivity.

[0023] Systems, methods, and computer program products are provided and described and directed to a federated learning framework (FLF) with localization adapted for local group patterns. A FLF is shown herein where multiple MLMs are organized into a hierarchy or hierarchical representation with layers and dependencies. A hierarchy is understood in the art as an inverted tree structure where multiple items are grouped in the layers. Examples of hierarchical structures are described in Figure 2 As shown and described in detail below, the FLF includes a hierarchy of nodes and one or more corresponding MLMs. A node may be a processor, a communication controller, a cluster controller, or one or more client-accessible terminals. Clustering techniques may be used to form groups of multiple nodes and one or more corresponding MLMs, referred to herein as local MLMs. Local MLMs provide personalized local models designed to capture local group pattern data and make decisions and predictions on a local basis, while global models in the hierarchy capture general patterns in groups formed in the hierarchy. More specifically, local models are trained for local group levels in the hierarchy and react on a local basis, and do not require the same amount of data as a federated dataset or global model.

[0024] Embodiments of this article can provide multiple advantages. By maintaining local models and capturing local patterns that can emphasize local behavior, the performance of the group can be enhanced. Typically, local models are maintained at the group or space level to alleviate the client's need for large training data sets and reduce training time. The global model can be used to initialize the local model of the newly formed group to ensure performance without a large amount of local training data. Typically, when the current local training data set is insufficient, the local model will be synchronized with the global model. Local models can be used to identify local patterns, make decisions, make predictions, reduce the client's computing and data requirements, and identify the naturally formed groups of the client to help local pattern learning. In this way, the computer system operates efficiently and effectively by providing customized learning to the client. In addition, privacy protection techniques can be added to protect the client's data, such as different privacy standards tailored for specific customers. Therefore, embodiments of this article can provide privacy protection, personalized learning, and online learning in a collaborative learning framework.

[0025] Typically, local models are based on local data or groupings of local data, hereinafter referred to as groups. Typically, due to limited training time budgets, models are usually trained in an online manner, and clients usually wait for classification results. More training data can ensure performance, especially when a large number of participants improves accuracy (for example, under different privacy schemes). Defining local models associated with groups allows obtaining trained models in limited training time.

[0026] Reference Figure 1, a schematic diagram of a computing system (100) is shown. As shown, a server (110) is provided to communicate with multiple computing devices (180), (182), (184), (186), (188) and (190) across a computer network (105) via links (102) and (104). The server (110) is configured with a processor (112) that communicates with a memory (116) via a bus (114). The server (110) is shown with an artificial intelligence (AI) platform (150) for performing cognitive computing, including natural language processing and machine learning, from one or more computing devices (180), (182), (184), (186), (188) and (190) via a computer network (105). More specifically, computing devices (180), (182), (184), (186), (188) and (190) communicate with each other and with other devices or components via one or more wired and / or wireless data communication links. Each communication link may include one or more of a line, a router, a switch, a transmitter, a receiver, etc. In this network arrangement, the server (110) and the computer network (105) implement communication detection, identification and resolution. Other embodiments of the server (110) may be used with components, systems, subsystems and / or devices other than those described herein.

[0027] The AI ​​platform (150) shown here is configured with tools to support the configuration and management of a federated learning framework (FLF) with corresponding and, in one embodiment, logically located local neural models and global neural models. These tools include, but are not limited to, a hierarchy manager (152), a machine learning model (MLM) manager (154), and a training manager (156). Figure 2 As shown and described, the AI ​​platform (150) creates and manages (FLF) (130) to logically organize and position neural models in a hierarchy including at least one global or primary model and at least one local or secondary model. Tools (152)-(156) are shown operably coupled to the FLF (130) via at least one communication link (158).

[0028] The AI ​​platform (150) can receive input from the computer network (105) and utilize a data source (160) (also referred to herein as a corpus or knowledge base) to selectively access data. As shown, the data source (160) is configured with a library (162) having multiple models, which are created and managed by the hierarchy manager (152). The details of how the models are created will be shown and detailed below. It should be understood that different domains, such as different groups or clients, can each be classified as a domain. In the examples shown herein, the domains include, but are not limited to, the domain A (162 A ),domain B (162 B ) and domain C (162 C ). Although only three domains are shown and represented here, the number should not be considered limiting. In one embodiment, there may be a different number of domains. Similarly, domains may be added to the library (162). The MLM manager (154) stores or categorizes corresponding activity data for each domain. As shown in the figure, the domain A (162 A ) including activity data A (164 A ),domain B (162 B ) including activity data B (164 B ),domain C (162 C ) including activity data C (164 C ).

[0029] It should be understood that supervised learning utilizes data from a data source. As shown in this article, the data source is referred to as a knowledge base (160) and is configured to have domains in the form of models and logically grouped activity data (also referred to as query response data in this article). The MLM manager (154) is used to collect or extract data in the form of model updates from various computing devices (180), (182), (184), (186), (188) and (190) that communicate with the computer network (105). Once collected, the training manager (156) organizes or arranges the data collected from one or more computing devices into one or more corresponding models. More specifically, the training manager (156) trains the model by issuing queries to clients in the domain and collecting the client's responses to the query. Models can be created based on intra-domain activity or inter-domain activity. Two models are shown here, although the number of models and the relationship of the models to the domains should not be considered limiting. Model A (166 A ) is shown as being operatively coupled to the activity data (164 A), and is an in-domain active model. Model B (166 B ) is shown as being operatively coupled to the active data B (164 B ) and activity data C (164 C ), and is an inter-domain activity model, also referred to herein as a multi-class classification model. When communication is detected, the activity data is updated, and each model configured and operatively coupled to the activity data is dynamically updated by a training manager (156).

[0030] The hierarchy manager (152) builds the FLF (130) and logically organizes the hierarchy (132) of machine learning models (MLMs). In one embodiment, the MLMs are maintained in a library (162) and organized into FLFs (130). Typically, the hierarchy (132) contains multiple layers of MLMs, with the number of layers of MLMs increasing from level one to level two to level three.

[0031] As described in further detail below, the MLM manager (154) operatively and logically locates one or more secondary MLMs within the FLF (130) locally at the secondary layer and restricts access to the secondary MLMs. In addition, the MLM manager (154) may operatively locate and configure the primary MLMs globally, where the primary MLMs are accessible to the secondary MLMs. Additionally, the MLM manager (154) may synchronize the secondary MLMs with the primary MLMs, including aggregating weight parameters under the secondary MLM settings, and updating the primary MLMs using aggregated local neural model weights included in the captured contribution data. Furthermore, as described below, the MLM manager (154) may logically locate or form new secondary MLMs by using clustering techniques. Figure 4 The details of the logical positioning and clustering of MLMs within a hierarchy are shown and described in .

[0032] The training manager (156) trains the primary MLM, which includes issuing queries to one or more secondary MLMs or associated data sources and collecting responses or model updates from one or more secondary MLMs or associated data sources. In addition, the training manager (156) initializes the created secondary MLM, as shown and described in detail below, which includes cloning the weights and framework of the primary MLM into the secondary MLM and populating the created secondary MLM with secondary data, the populated data including the client's responses to the queries that are locally or operatively coupled to the created secondary MLM.

[0033] A response output (172) in the form of one or more derived actions, such as an action sequence or a modified action sequence, is transmitted or otherwise communicated to the processor (112) for execution. In one embodiment, the response output (172) is transmitted to a corresponding network device, shown here as a visual display (170), which is operably coupled to the server (110) or, in one embodiment, is operably coupled to one or more computing devices (180)-(190) on the computer network (105).

[0034] Reference Figure 2 , a block diagram (200) is provided showing a schematic diagram of a federated learning framework (FLF). As shown, the FLF shows a hierarchical structure (210) having multiple layers, referred to herein as layers, and is shown here to have three layers, referred to as layers 0 (220), Layer 1 (240) and layer 2 (270). The hierarchy (210) represents a structure for organizing and associating multiple neural networks, including creating dependencies and gradients between layers. As shown and described in the example, the hierarchy (210) is an inverted tree structure in which multiple items are associated in layers. The hierarchy (210) is shown as having multiple layers, such as three layers, but the number of layers should not be considered limiting. Although the hierarchy (210) can be considered a grouping, it should be understood in the context of the embodiments of this document that grouping also occurs within each layer. Each layer can have one or more items, referred to herein as nodes. As shown in this example, the layers 0 (220) includes nodes operably coupled to a model (224) (hereinafter referred to as a global model) 0,0 (222). In one embodiment, the node 0,0 (222) is called a primary node or an independent node. 1 (240) has three nodes, including operatively coupled to a local or secondary model A (242 A ) 1,0 (242), which can be operatively coupled to local or secondary models B (244 A ) and the nodes 1,1 (244), which can be operatively coupled to a local or secondary model C (246 A ) 1,2 (246). As described below, as shown by the dashed line, the node 1,0 (242) may be a newly formed node. 1,0 (242), node 1,1 (244) and node 1,2(246) and node 0,0 (222) relationships, they are called subordinate nodes. 2 (270) has eight nodes, including node 2,0 (272), node 2,1 (274), node 2,2 (276), node 2,3 (278), node 2,4 (280), node 2,5 (282), node 2,6 (284) and nodes 2,7 (286), called directly dependent on the node 1,0 (242), node 1,1 (244) and node 1,2 (246) is a subordinate node and indirectly depends on an independent node 0,0 (222). Node 2,0 (272) and data 2,0 (272 A ) can communicate operatively, nodes 2,1 (274) and data 2,1 (274 A ) can communicate operatively, nodes 2,2 (276) and data 2,2 (276 A ) can communicate operatively, nodes 2,3 (278) and data 2,3 (278 A ) can communicate operatively, nodes 2,4 (280) and data 2,4 (280 A ) can communicate operatively, nodes 2,5 (282) and data 2,5 (282 A ) can communicate operatively, nodes 2,6 (284) and data 2,6 (284 A ) are operable to communicate, and the nodes 2,7 (286) and data 2,7 (286 A ) are operable to communicate.

[0035] Provide communication channels across the various layers of the hierarchy. As shown in the figure, at least one communication channel (230) enables the first-level node 0 , 0(222) is operably coupled to a slave node 1,0 (242), slave node 1,1 (244) and slave nodes1,2 (246). Provides a local communication channel to connect layers 0,1 Nodes and layers in (240) 0,2 The nodes in (270) are operably coupled. The local communication channel (250) enables the nodes to communicate via the communication channel (260). 1,0 (242) and node 2,0 (272) and nodes 2,1 (274) is operably coupled, the local communication channel (252) enables the node to communicate via the communication channel (262) 1,1 (244) and node 2,2 (276), Node 2,3 (278) and nodes 2,4 (280) is operably coupled, and the local communication channel (254) enables the node to communicate via the communication channel (264) 1,2 (246) and node 2,5 (282), Node 2,6 (284) and nodes 2,7 (286) are operably coupled.

[0036] Level 1 Node 0,0 (222) From the base layer (e.g. layer 1 (240)) captures the contribution data. As shown in the figure, the first-level node 0,0 (222) is locally attached to a global model (224), which can be any suitable MLM, such as a long short-term memory (LSTM). 1,0 (242), slave node 1,1 (244) and slave nodes 1,2 (246) may independently include a group or secondary server operably coupled to a local or secondary MLM, which may be any suitable MLM, such as a long short-term memory (LSTM). As described herein, the slave nodes 1,0 (242) and local model A (242A) performs local communication, slave node 1,1 (244) and local model B (244 B ) for local communication, while slave nodes 1,2 (246) and local model C (246 C ) for local communication. 1 Each local model shown in (240) is based on the layer 2 (270). In the example shown in this article, the local model A (242A ) Based on the data 2,0 (272 A ) and data 2.1 (274 A ) Received model updates, local models B (244 A ) Based on the data 2,2 (276 A ),data 2,3 (278 A ) and data 2,4 (280 A ) received model updates, and local models C (244 A ) Based on the data 2,5 (282 A ),data 2,6 (284 A ) and data 2,7 (286 A ) received. Across the various layers of the hierarchy (for example, at the layer 1 (240) and layer 2 The data communicated between (270) can be gradients or model parameter information, so that 1 (240) Training a local model. The gradient may be based on the local data of at least one client, and the gradient may occur when training the corresponding MLM because the gradient from the layer 1 (240) A response to the query is issued. In one embodiment, a local group model can be applied by layer 1 The nodes in (240) average all the gradients. Similarly, the slave nodes 1,0 (242), node 1,1 (244) and node 1,2 The isolation of (246) creates privacy differences, which provides more impetus for localized training. Accordingly, these features allow for localization and separation learning of nodes in the hierarchy (210).

[0037] Level 1 Node 0,0 (222) Monitor the joint learning process. First-level node 0,0 The functions of (222) include issuing queries, collecting query responses, and aggregating the collected responses to update the global model (224). In one embodiment, the first-level node 0,0 (222) is referred to as an aggregator. The query response is provided by a client, which in one embodiment is also referred to herein as a participant or party. The query response is based on local data. For example, a node represented as 2,1 (274) The query response provided by the client is based on local data 2,1(274A). The reply to the query is from the client (e.g., node 2,1 (274)) can be communicated via the local model A (242 A ) transmits the model updates to the global model (224).

[0038] The MLM of the hierarchy (210) may be stored in a knowledge base (160) and operably coupled to the server (110). In an exemplary embodiment, the hierarchy (210) may be hosted on a single central server that is accessed only by clients that are members of the FLF. Alternatively, the central server may maintain the global model (224), while a set of separate servers may be responsible for one or more sets of local models, such as for local models. A (242A) local server and for local models B (244 A ) and local models C (246 A ) with a separate local server. In this release, maintenance and storage of local models is minimized.

[0039] The hierarchical structure shown and the associated number and structure of nodes and related items represented in the hierarchy are examples only and should not be considered limiting. The hierarchical structure (210) is a structure for organizing and classifying neural models, intended recipients, and related content. As shown and described herein, content and associated atomic units can be classified or identified with allowed characteristics in response to the hierarchical structure. By definition and illustration, assigning one or more allowed characteristics is not equivalent to organizing notification categories into a hierarchy. Similarly, the use of the hierarchical structure does not necessarily assign a priority to any content. Therefore, the classification model is shown herein as a hierarchical structure with a representation of a category for which the content and associated atomic units are processed.

[0040] As shown, in various embodiments, the computer network (105) can include local network connections and remote connections, so that the AI ​​platform (150) can operate in environments of any size, including local and global, such as the Internet. In addition, the AI ​​platform (150) serves as a front-end system that can make available various knowledge extracted or represented from network-accessible sources and / or structured data sources. In this way, some processes populate the AI ​​platform (150), and the artificial intelligence platform (150) also includes an input interface for receiving requests and responding accordingly.

[0041] The knowledge base (160) is configured with logically grouped domains (162 A )-(162 C ) and the corresponding model (164A )-(164 C ) for use by the AI ​​platform (150). In one embodiment, the knowledge base (160) may be configured with other or additional input sources, and therefore, the input sources shown and described herein should not be considered limiting. Similarly, in one embodiment, the knowledge base (160) includes structured, semi-structured and / or unstructured content related to activities and tasks. Various computing devices (180)-(190) in communication with the computer network (105) may include access points for logically grouped domains and models. Some computing devices may include devices for a database that stores a data corpus that is a body of information used by the AI ​​platform (150) to generate a response output (172) and communicate the response output to a corresponding network device, such as a visual display (170) operably coupled to one or more computing devices (180)-(190) on the server (110) or the entire computer network (105).

[0042] In various embodiments, the computer network (105) may include local network connections and remote connections so that the artificial intelligence platform (150) can operate in environments of any size, including local and global (e.g., the Internet). In addition, the artificial intelligence platform (150) serves as a front-end system that makes available various knowledge extracted or represented from network-accessible sources and / or structured data sources. In this manner, some processes populate the AI ​​platform (150), wherein the AI ​​platform (150) also includes one or more input interfaces or portals to receive requests and respond accordingly.

[0043] An AI platform (150) connected to a computer network (105) via a network connection or the Internet is configured to detect and manage network activity and task data related to online learning. The AI ​​platform (150) can effectively choreograph or optimize action sequences for relevant activity data by utilizing a knowledge base (160), which in one embodiment can be operably coupled to a server (110) via the computer network (105).

[0044] In some illustrative embodiments, the server (110) may be an IBM Watson system available from International Business Machines Corporation of Armonk, New York, having the mechanisms of the illustrative embodiments described below. Managers (152)-(156), hereinafter collectively referred to as AI tools, are shown as an AI platform (150) embodied in or integrated with the server (110). The AI ​​tools may be implemented in a separate computing system (e.g., 190), or in one embodiment, they may be implemented in one or more systems connected to the server (110) across a computer network (105). Regardless of where embodied, the AI ​​tools may dynamically optimize activities to minimize or mitigate risk.

[0045] The types of devices and corresponding systems that can utilize the artificial intelligence platform (150) range from small handheld devices, such as handheld computers / mobile phones (180) to large mainframe systems, such as mainframe computers (182). Examples of handheld computers (180) include personal digital assistants (PDAs), personal entertainment devices (such as MP4 players), portable televisions, and compact disc players. Other examples of information processing systems include pen or tablet computers (184), laptop or notebook computers (186), personal computer systems (188), and servers (190). As shown, a computer network (105) can be used to network various devices and systems together. Types of computer networks (105) that can be used to interconnect various devices and systems include local area networks (LANs), wireless local area networks (WLANs), the Internet, the public switched telephone network (PSTN), other wireless networks, and any other network topology that can be used to interconnect devices and systems. Many devices and systems include non-volatile data storage, such as hard drives and / or non-volatile memory. Some devices and systems may use a separate non-volatile data store (e.g., the server (190) utilizes a non-volatile data store (190) A ), and large computers (182) utilize non-volatile data storage (182 A ). Non-volatile data storage (182 A ) can be a component external to the various devices and systems or can be a component internal to one of the devices and systems.

[0046] Devices and systems for supporting the artificial intelligence platform (150) can take many forms, some of which are Figure 1 For example, an information handling system may take the form of a desktop, server, portable computer, laptop computer, notebook computer, or other form factor computer or data processing system. In addition, devices and systems may take other form factors such as a personal digital assistant (PDA), a gaming device, an ATM machine, a portable telephone device, a communication device, or other devices that include a processor and memory.

[0047] In the art, an application program interface (API) is understood as a software intermediary between two or more applications. Figure 1 The AI ​​platform (150) shown and described in the accompanying drawings may utilize one or more APIs to support one or more of the tools (152)-(156) and their associated functionality. Figure 3 , a block diagram (300) is provided, which shows at least some of the tools (352)-(356) and their related APIs. As shown, multiple tools are embedded in the AI ​​platform (305), wherein these tools include APIs 0 (312) is associated with a hierarchy manager (152) shown here as (352), with an API 1 (322) associated with an MLM manager (154) shown here as (354), and an API 2 (332) is associated with a training manager (156) shown here as (356). Each API may be implemented in one or more languages ​​and interface specifications. API 0 (312) Provides functional support to collect and organize activity data across two or more domains. API 1 (322) Provide functional support for MLM corresponding to the collected and organized activity data; and API 2 (332) provides functional support to dynamically optimize, orchestrate and update models and corresponding FLF groups. As shown, each API (312), (322) and (332) is operably coupled to an API orchestrator (360), or orchestration layer, which is understood in the art to be used as an abstraction layer to transparently string together separate APIs. In one embodiment, the functionality of the separate APIs can be combined or combined. Thus, the configuration of the APIs shown herein should not be considered restrictive. Therefore, as shown herein, the functionality of the tools can be embodied or supported by their respective APIs.

[0048] As shown and described herein, one or more local models are located within the FLF. In one embodiment, a local model can represent a set of nodes and corresponding data. For example, Figure 2 As shown, the local model B (244 B ) indicates that the node 2,2 (276), Node 2,3 (278) and nodes 2,4(280) Formed groups. Groups can be created in FLF using clustering techniques. Similarly, created groups can be adjusted, which in one embodiment can include changing the members of the group. With respect to clustering, there are different mechanisms to form new groups. One mechanism is to naturally form new groups based on an existing hierarchy. For example, such groups can be composed of clients working on the same project. Clustering techniques can also be used to form new groups based on shared similar physical characteristics (e.g., language, job, title, etc.). Clients can contribute to all local models of their member groups. With respect to group adjustment, new groups can be created by splitting existing groups or merging two or more groups. Prior to splitting, analysis is required to determine the performance of existing local models in predicting group behavior. Examples of performance characteristics that may contribute to splitting include, but are not limited to, low prediction accuracy. Merging existing groups can be applicable to groups that do not have sufficient training data. Therefore, different techniques can be applied to form new groups in FLF.

[0049] refer to Figure 4 A flow chart (400) is provided, illustrating a process for creating a new secondary model in FLF. As shown, a federated learning framework (FLF) is constructed or a process (402) is performed thereon, supported by the functionality of a hierarchy manager (152). Figure 2 An example of a constructed FLF is shown and described in . The FLF is consulted and a secondary model, such as a local model, is identified in the FLF (404). A new secondary model can be created, synchronized, or trained (410). A new secondary model is created by splitting an existing secondary model (412), forming a new secondary model (414), or merging two or more existing secondary models (416). There are multiple ways to merge information from different secondary models. A weighted average can be determined by scoring and measuring accuracy based on the confidence of each group associated with the model. Typically, the model is trained and compared to the prediction. If sufficiently correlated, the groups or secondary models can be combined. Alternatively, predictions are made by the closest groups. Typically, the group with the highest proximity score is used, and behavior is better predicted by having closer delta scores on active channels. After the new secondary model (410) is created in one or more of steps (412), (414) or (416), the new secondary model created is initialized using the global model and the corresponding weights (420), and the framework and weights of the global model are applied to the created secondary model. Thus, one or more secondary models are created and initialized.

[0050] After step (404) or step (410), the existing secondary model is synchronized or trained. As shown, if there is not enough training data, the created secondary model (410) is initialized (418), including synchronization with the primary model (e.g., global model) (420), or if there is enough training data, it is trained (422) based on model updates from only clients belonging to the created secondary group. In one embodiment, the global model is used to make decisions or predictions until the secondary model has enough training data to capture local patterns. Therefore, the secondary MLM is operably associated with and synchronized with the global model.

[0051] Reference Figure 5 A flow chart (500) is provided, illustrating a process for controlled training of a secondary model. As shown and described above, the framework and weights of the global model are used to create or establish an initial framework of the secondary model, and the secondary model is trained based on local customer data in a federated learning manner. Prior to training the secondary model, the number of iterations N and the data threshold B for calculating model updates are initialized or established in response to a training query for the secondary model. threshold (502). In one embodiment, until the secondary model has reached a threshold, the global model or a different secondary model will be used. Establishing a termination condition As a metric for evaluating model performance (504), the batch size (508) for each training iteration from centralized neural network training or adapted from distributed machine learning (506) is set. The model framework begins by using a threshold amount of data (e.g., B threshold ) performs controlled training on the local model (510). The F1 score is an indicator for evaluating model performance. The F1 score is calculated for the current secondary model (512). After the calculation in step (512), it is determined whether the performance metric of the model has improved sufficiently relative to the previous average performance. In one embodiment, the determination of step (514) is mathematically expressed as:

[0052]

[0053] where ρ is the precision improvement factor and ρ>1 is a given constant, k is the current iteration, is the F1 score of the local model at iteration k, which ranges from 0 to 1, and is the average F1 score from iteration k / 2 to iteration k. thresholdThe increase in percentage c (516) indicates a positive response to the determination of step (514). However, a negative response is made to the determination of step (514) or a determination is made after the increase in step (516) to evaluate whether the performance metric of the secondary model exceeds the target performance metric (518). In one embodiment, the determination in step (518) is mathematically represented as:

[0054]

[0055] in is the target F1 score and N is the iteration limit. After a negative response is made to the determination at step (518), a return is made to step (510), and an affirmative response ends the training of the secondary model. Thus, as shown herein, the secondary model is trained based on a threshold amount of data and a target performance metric.

[0056] Above Figure 1 An example online learning algorithm is described in the description of the system. It should be understood that this is an example learning algorithm and therefore should not be considered limiting. In one embodiment, another reinforcement learning algorithm can be used to optimize the arrangement list or feature data. The output of the reinforcement learning algorithm learns the values ​​of different states and generates a list of actions to be performed for training. One or more actions from the generated list are selectively performed to obtain the desired level of compliance. Therefore, the reinforcement learning shown and described herein dynamically learns the values ​​of different states from primary and secondary data sources, which can then be applied to create local training modules.

[0057] The embodiments shown and described herein may be in the form of a computer system for use with an intelligent computer platform to provide orchestration of activities across one or more domains to minimize risk. Aspects of tools (152)-(156) and their associated functionality may be embodied in a computer system / server at a single location, or in one embodiment, may be configured in a cloud-based system that shares computing resources. Figure 6 , a block diagram (600) is provided, which shows an example of a computer system / server (602) (hereinafter referred to as a host (602)) in communication with a cloud-based support system to implement the above Figures 1 to 5The host computer (602) may be operable with many other general or special purpose computing system environments or configurations. Examples of well-known computing systems, environments, and / or configurations that may be suitable for use with the host computer (602) include, but are not limited to, personal computer systems, server computer systems, thin clients, fat clients, handheld or laptop devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputer systems, mainframe computer systems, and file systems including any of the above systems, devices, and their equivalents (e.g., distributed storage environments and distributed cloud computing environments).

[0058] The host (602) may be described in the general context of computer system executable instructions (e.g., program modules) executed by a computer system. In general, program modules may include routines, programs, objects, components, logic, data structures, etc. that perform specific tasks or implement specific abstract data types. The host (602) may be practiced in a distributed cloud computing environment, where tasks are performed by remote processing devices linked through a communications network. In a distributed cloud computing environment, program modules may be located in local and remote computer system storage media including memory storage devices.

[0059] like Figure 6 As shown, the host (602) is shown in the form of a general computing device such as a computer system and / or server. The components of the host (602) may include, but are not limited to, one or more processors or processing units (604), such as a hardware processor, a system memory (606), and a bus (608) that couples various system components including the system memory (606) to the processor (604). The bus (608) represents any one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, and a processor or local bus using any of a variety of bus architectures. As an example and not limitation, such architectures include an industry standard architecture (ISA) bus, a microchannel architecture (MCA) bus, an enhanced ISA (EISA) bus, a video electronics standard association (VESA) local bus, and a peripheral component interconnect (PCI) bus. The host (602) typically includes various computer system readable media. Such media can be any available media accessible to the host (602), and it includes volatile and non-volatile media, removable and non-removable media.

[0060] The memory (606) may include computer system readable media in the form of volatile memory, such as random access memory (RAM) (630) and / or cache memory (632). By way of example only, a storage system (634) may be provided for reading and writing to a non-removable, non-volatile magnetic medium (not shown and commonly referred to as a "hard drive"). Although not shown, a magnetic disk drive for reading and writing to a removable, non-volatile magnetic disk (e.g., a "floppy disk"), and an optical disk drive for reading or writing to a removable, non-volatile optical disk may be provided, such as a CD-ROM, DVD-ROM, or other optical media. In such a case, each may be connected to the bus (608) via one or more data media interfaces.

[0061] A program / utility (640) having a set (at least one) of program modules (642) may be stored in memory (606) by way of example and not limitation, along with an operating system or more application programs, other program modules, and program data. Each operating system, one or more application programs, other program modules, and program data, or some combination thereof, may include an implementation of a networked environment. The program modules (642) generally perform the functions and / or methods of the embodiments to dynamically orchestrate activities across one or more domains to minimize risk. For example, a set of program modules (642) may include, for example, Figure 1 Tools (152)-(156) described in.

[0062] The host (602) may also communicate with one or more external devices (614), such as a keyboard, a pointing device, etc., a display (624), one or more devices that enable a user to interact with the host (602), and / or any device that enables the host (602) to communicate with one or more other computing devices (e.g., a network card, a modem, etc.). Such communication may occur via an input / output (I / O) interface (622). Still, the host (602) may communicate with one or more networks, such as a local area network (LAN), a general wide area network (WAN), and / or a public network (e.g., the Internet) via a network adapter (620). As shown, the network adapter (620) communicates with other components of the host (602) via a bus (608). In one embodiment, multiple nodes of a distributed file system (not shown) communicate with the host (602) via an I / O interface (622) or via a network adapter (620). It should be understood that, although not shown, other hardware and / or software components may be used in conjunction with the host (602). Examples include, but are not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems.

[0063] In this document, the terms "computer program medium", "computer usable medium" and "computer readable medium" are generally used to refer to media such as main memory (606), including RAM (630), cache (632) and storage systems (634), such as removable storage drives and hard disks installed in hard drives.

[0064] A computer program (also referred to as computer control logic) is stored in the memory (606). The computer program may also be received via a communication interface such as a network adapter (620). Such a computer program, when run, enables the computer system to perform the features of the present embodiments discussed herein. Specifically, the computer program, when run, enables the processing unit (604) to perform the features of the computer system. Thus, such a computer program represents a controller of the computer system.

[0065] A computer-readable storage medium may be a tangible device that can retain and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but 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. A non-exhaustive list of more specific examples of computer-readable storage media includes the following: a portable computer floppy disk, a hard disk, a dynamic or static random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a magnetic storage device, a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device such as a punch card or a raised structure in a groove having instructions recorded thereon, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium should not be understood as being a transient signal in itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse passing through a fiber optic cable), or an electrical signal transmitted through a wire.

[0066] The computer-readable program instructions described herein may be downloaded from a computer-readable storage medium to a corresponding computing / processing device, or downloaded to an external computer or external storage device via a network (e.g., the Internet, a local area network, a wide area network, and / or a wireless network). The network may include copper transmission cables, optical transmission fibers, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions to be stored in a computer-readable storage medium within the corresponding computing / processing device.

[0067] The computer-readable program instructions for performing the operation of the present embodiment can be assembly instructions, instruction set architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, and the programming language includes object-oriented programming languages ​​(such as Smalltalk, C++, etc.) and conventional procedural programming languages ​​(such as "C" programming languages ​​or similar programming languages). The computer-readable program instructions can be performed completely on the user's computer, partly on the user's computer, as an independent software package, partly on the user's computer and partly on a remote computer, or completely on a remote computer or server. In the latter case, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider through the Internet). In certain embodiments, in order to perform the various aspects of the present embodiment, the electronic circuit including, for example, a programmable logic circuit, a field programmable gate array (FPGA) or a programmable logic array (PLA) can perform computer-readable program instructions to personalize the electronic circuit by utilizing the state information of the computer-readable program instructions.

[0068] In one embodiment, the host (602) is a node of a cloud computing environment. As is known in the art, cloud computing is a service delivery model for enabling configurable computing resources (e.g., networks, network bandwidth, servers, processing, memory, storage, applications, virtual machines, and services) to be quickly configured and released with minimal management effort or interaction with service providers. The cloud model can include at least five features, at least three service models, and at least four deployment models. Examples of such features are as follows:

[0069] On-demand self-service: Cloud consumers can unilaterally and automatically deploy computing capabilities such as server time and network storage on demand without human interaction with the service provider.

[0070] Broad network access: Computing power can be accessed over the network through standard mechanisms that facilitate the use of the cloud through different types of thin-client or thick-client platforms (e.g., mobile phones, laptops, personal digital assistants (PDAs)).

[0071] Resource pool: The computing resources of the provider are grouped into resource pools and serve multiple consumers through a multi-tenant model, where different physical and virtual resources are dynamically allocated and reallocated on demand. Generally, consumers cannot control or even know the exact location of the provided resources, but can specify the location (such as country, state or data center) at a higher level of abstraction, so it is location-independent.

[0072] Rapid elasticity: The ability to quickly and elastically (sometimes automatically) deploy computing power to scale up quickly, and quickly release it to scale down quickly. To consumers, the computing power available for deployment often appears to be unlimited, and any amount of computing power can be accessed at any time.

[0073] Measurable services: Cloud systems automatically control and optimize resource usage by leveraging metering capabilities at a level of abstraction appropriate to the type of service (e.g., storage, processing, bandwidth, and active user accounts). Resource usage can be monitored, controlled, and reported, providing transparency to both service providers and consumers.

[0074] The service model is as follows:

[0075] Software as a Service (SaaS): The capability provided to consumers is to use the provider's applications running on a cloud infrastructure. Applications can be accessed from a variety of client devices through a thin client interface such as a web browser (e.g., web-based email). Aside from limited user-specific application configuration settings, consumers neither manage nor control the underlying cloud infrastructure including networks, servers, operating systems, storage, or even individual application capabilities.

[0076] Platform as a Service (PaaS): The capability provided to consumers is to deploy applications created or acquired by consumers on the cloud infrastructure, which are created using programming languages ​​and tools supported by the provider. Consumers neither manage nor control the underlying cloud infrastructure including networks, servers, operating systems or storage, but have control over the applications they deploy and may also have control over the configuration of the application hosting environment.

[0077] Infrastructure as a Service (IaaS): The capabilities provided to consumers are processing, storage, network, and other basic computing resources on which consumers can deploy and run arbitrary software including operating systems and applications. Consumers neither manage nor control the underlying cloud infrastructure, but have control over the operating system, storage, and applications they deploy, and may have limited control over selected network components (such as host firewalls).

[0078] The deployment model is as follows:

[0079] Private Cloud: Cloud infrastructure is run solely for an organization. The cloud infrastructure can be managed by the organization or a third party and can exist inside or outside the organization.

[0080] Community Cloud: A cloud infrastructure is shared by several organizations and supports a specific community with common interests (e.g., mission, security requirements, policies, and compliance considerations). A community cloud can be managed by multiple organizations within the community or by a third party and can exist inside or outside the community.

[0081] Public cloud: Cloud infrastructure is provided to the public or to large industry groups and is owned by an organization that sells cloud services.

[0082] Hybrid cloud: A cloud infrastructure consisting of two or more clouds (private, community or public) in a deployment model that remain distinct entities but are bound together by standardized or proprietary technologies (such as cloud burst sharing for load balancing between clouds) that enable data and application portability.

[0083] The cloud computing environment is service-oriented, with characteristics centered on statelessness, low coupling, modularity, and semantic interoperability. The core of cloud computing is the infrastructure that includes a network of interconnected nodes.

[0084] Now refer to Figure 7 , schematic cloud computing network (700). As shown, the cloud computing network (700) includes a cloud computing environment (750) having one or more cloud computing nodes (710), and local computing devices used by cloud consumers can communicate with the cloud computing nodes (710). Examples of these local computing devices include, but are not limited to, personal digital assistants (PDAs) or cellular phones (754A), desktop computers (754B), laptop computers (754C), and / or automotive computer systems (754N). Individual nodes within a node (710) can further communicate with each other. They can be physically or virtually grouped (not shown) in one or more networks (such as private, community, public, or hybrid clouds or combinations thereof as described above). This allows the cloud computing environment (700) to provide infrastructure, platforms, and / or software as a service, for which cloud consumers do not need to maintain resources on local computing devices. It should be understood that Figure 7 The types of computing devices (754A-N) shown in are intended to be illustrative only, and the cloud computing environment (750) may communicate with any type of computerized device over any type of network and / or network addressable connection (eg, using a web browser).

[0085] Now refer to Figure 8 , shown by Figure 7 A set of functional abstraction layers (800) provided by the cloud computing network. It should be understood in advance that Figure 8The components, layers and functions shown in are intended to be illustrative only, and the embodiments are not limited thereto. As shown, the following layers and corresponding functions are provided: hardware and software layer (810), virtualization layer (820), management layer (830) and workload layer (840).

[0086] The hardware and software layer (810) includes hardware and software components. Examples of hardware components include: a host computer, e.g. System; Servers based on RISC (Reduced Instruction Set Computer) architecture, such as IBM System; IBM System; IBM systems; storage devices; networks and network components. Examples of software components include: network application server software, such as IBM Application server software; database software, such as IBM Database software. (IBM, zSeries, pSeries, xSeries, BladeCenter, WebSphere and DB2 are registered trademarks of International Business Machines Corporation throughout the world).

[0087] The virtualization layer (820) provides an abstraction layer that can provide examples of the following virtual entities: virtual servers, virtual storage, virtual networks (including virtual private networks), virtual applications and operating systems, and virtual clients.

[0088] In one example, the management layer (830) may provide the following functions: Resource provisioning function: providing dynamic acquisition of computing resources and other resources for performing tasks in a cloud computing environment; Metering and pricing function: tracking the cost of resource usage within the cloud computing environment and providing bills and invoices for this purpose. In one example, the resources may include application software licenses. Security function: providing identity authentication for cloud consumers and tasks, and providing protection for data and other resources. User portal function: providing access to the cloud computing environment for consumers and system administrators. Service level management function: providing allocation and management of cloud computing resources to meet the required service levels. Service level agreement (SLA) planning and fulfillment function: providing pre-scheduling and provisioning for future demand for cloud computing resources predicted according to the SLA.

[0089] The workload layer (840) provides examples of functions that may be implemented in a cloud computing environment. In this layer, examples of workloads or functions that may be provided include: mapping and navigation; software development and lifecycle management; virtual classroom teaching delivery; data analysis processing; transaction processing; and hierarchy management of neural models.

[0090] It should be appreciated that a system, method, apparatus, and computer program product are disclosed herein for evaluating natural language input, detecting interrogative sentences in corresponding communications, and resolving detected interrogative sentences with answers and / or supporting content.

[0091] Although specific embodiments of the present embodiment have been shown and described, it is obvious to those skilled in the art that changes and modifications can be made without departing from the embodiments and their broader aspects based on the teachings herein. Therefore, the appended claims will include within their scope all changes and modifications within the true spirit and scope of the embodiments. In addition, it should be understood that the embodiments are limited only by the appended claims. It will be understood by those skilled in the art that if a specific number of claim elements are intended to be introduced, such intention will be explicitly stated in the claims, and in the absence of such a statement, there is no such limitation. As a non-limiting example, to aid understanding, the following appended claims contain the use of introductory phrases "at least one" and "one or more" to introduce claim elements. However, the use of such phrases should not be interpreted as implying that the introduction of the indefinite article "a" or "an" to the claim element will limit any specific claim containing the introduced claim element to an embodiment containing only one such element, even if the same claim includes the introductory phrases "one or more" or "at least one" and indefinite articles such as "a" or "an", in the claim of the definite article.

[0092] The present embodiment may be a system, method and / or computer program product. In addition, selected aspects of the present embodiment may take the form of a fully hardware embodiment, a fully software embodiment (including firmware, resident software, microcode, etc.), or an embodiment combining software and / or hardware aspects, generally referred to herein as a "circuit", "module" or "system". In addition, aspects of the present embodiment may take the form of a computer program product embodied in a computer-readable storage medium having computer-readable program instructions thereon for causing a processor to perform aspects of the present embodiment. As embodied, the disclosed systems, methods and / or computer program products may be operable to improve the functionality and operation of an artificial intelligence platform to build a federated learning framework.

[0093] Various aspects of the present embodiment are described herein with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments. It will be understood that each frame of the flowchart illustration and / or block diagram and the combination of frames in the flowchart illustration and / or block diagram can be implemented by computer-readable program instructions.

[0094] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine, so that the instructions are executed by a computer or other processor to execute a programmable data processing device to create a device for implementing the functions / actions specified in the flowchart and / or block diagram box. These computer-readable program instructions can also be stored in a computer-readable storage medium, which can instruct a computer, a programmable data processing device, and / or other equipment to function in a specific manner, so that the computer-readable storage medium having the instructions stored therein includes an article of manufacture, which includes instructions for implementing various aspects of the functions / actions specified in the flowchart and / or block diagram or multiple boxes.

[0095] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing device or other device to cause a series of operating steps to be performed on the computer, other programmable device or other device to produce implemented computer processing, so that the instructions are executed on the computer, other programmable device or other device to implement the functions / actions specified in the flowchart and / or block diagram blocks.

[0096] The flow charts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the system, method and computer program product according to the various embodiments of the present embodiment. In this regard, each box in the flow chart or block diagram can represent a module, a fragment or a part of an instruction, which includes one or more executable instructions for realizing a specific logical function. In some alternative embodiments, the functions indicated in the box may not occur in the order indicated in the figure. For example, the two boxes shown in succession can actually be executed substantially at the same time, or these boxes can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box of the block diagram and / or flow chart and the combination of the boxes of the block diagram and / or flow chart can be implemented by a system based on dedicated hardware that performs a specified function or action or realizes a combination of hardware and computer instructions for special purposes.

[0097] It should be understood that although specific embodiments have been described herein for illustrative purposes, various modifications may be made without departing from the spirit and scope of the embodiments. In one embodiment, the FLF is a logical hierarchy of models that are operably coupled across layers. Therefore, the scope of protection of the embodiments is limited only by the attached claims and their equivalents.

Claims

1. A computer system for building a joint learning framework, comprising: a processor operatively coupled to the memory; An artificial intelligence (AI) platform in communication with the processor and having a natural language processing tool, including: A hierarchy manager, configured to construct a federated learning framework including a hierarchy, wherein the hierarchy manager creates a hierarchy of machine learning models MLM, wherein the hierarchy of MLMs includes a first-level MLM in a first-level layer of the hierarchy, and the federated learning framework includes a hierarchy of nodes and one or more corresponding MLMs; a training manager for training the first-level MLM, including capturing contribution model updates via at least one communication channel; the hierarchy manager for creating a secondary MLM logically located in a secondary level of the hierarchy, the secondary MLM being operatively coupled to the primary MLM via the at least one communication channel; The training manager is used to initialize the created secondary MLM, including cloning the weights and framework of the primary MLM into the secondary MLM, and populating the created secondary MLM with secondary data, wherein the populated data includes model updates to the created secondary MLM; An MLM manager, for logically placing the secondary MLM locally in the secondary layer and restricting access to the secondary MLM within the secondary layer; The MLM manager is used to globally store primary MLM data, wherein the primary MLM is accessible by the secondary MLM; and The MLM manager is further used to manage the federated learning framework, including using clustering technology to group nodes of the federated learning framework according to one or more attributes to form the secondary MLM, and managing one or more group memberships within one or more formed groups.

2. The system according to claim 1 further includes the MLM manager for synchronizing the secondary MLM with the primary MLM, including aggregating weight parameters under the secondary MLM setting, and updating the primary MLM with the aggregated local neural network model weights included in the captured contribution data.

3. The system of claim 1, further comprising the MLM manager forming a new secondary MLM, which comprises the MLM manager applying a clustering technique within the secondary layer, the applied clustering technique being used to identify new clusters.

4. The system of claim 3, wherein applying the clustering technique within the secondary layer further comprises: The MLM manager splits the secondary MLM, the splitting forms at least one new secondary MLM, and trains the at least one new secondary MLM.

5. The system of claim 3, wherein applying the clustering technique further comprises: The MLM manager combines two or more existing secondary MLMs into a merged secondary MLM.

6. The system of claim 1, further comprising the training manager updating the secondary MLM using model updates local to the secondary MLM.

7. A computer program product for constructing a federated learning framework, the computer program product comprising a computer-readable storage medium having program code contained therein, the program code being executable by a processor to: Establishing a federated learning framework, including program code to create a hierarchy of machine learning models MLM, wherein the hierarchy of MLM includes a first-level MLM in a first-level layer of the hierarchy; Training the primary MLM includes capturing contribution model updates in at least one communication channel; creating a secondary MLM in a secondary layer of the hierarchy, the secondary MLM being logically coupled to the primary MLM via at least one communication channel, the federated learning framework comprising a hierarchy of nodes and one or more corresponding MLMs; Initializing the created secondary MLM, including cloning the weight and framework of the primary MLM to the secondary MLM, and filling the created secondary MLM with secondary data, wherein the filled data includes model updates to the created secondary MLM; storing the secondary MLM locally in the secondary layer and restricting access to the secondary MLM to members of the secondary group locally in the secondary layer; storing the primary MLM globally, wherein the primary MLM data is accessible to the secondary MLM; and Managing the federated learning framework includes grouping nodes of the federated learning framework according to one or more attributes to form the secondary MLM using clustering techniques, and managing one or more group memberships within one or more formed groups.

8. The computer program product of claim 7, further comprising program code for synchronizing the secondary MLM with the primary MLM, including program code for aggregating local data and weight parameters in a secondary MLM setting and updating the primary MLM using aggregated local neural network model weights contained in the captured contribution data.

9. The computer program product of claim 7, further comprising program code for forming new secondary MLMs, including program code for applying a clustering technique within the secondary layer, the applied clustering technique being used to identify new secondary MLMs.

10. The computer program product of claim 9, wherein applying the clustering technique within the secondary layer further comprises program code for splitting the secondary MLM and training the resulting new secondary MLM.

11. The computer program product of claim 9, wherein applying the clustering technique further comprises program code for combining two or more existing two-level MLMs into a merged two-level MLM.

12. The computer program product of claim 7, further comprising program code for updating the secondary MLM using model updates local to the secondary MLM.

13. A method for constructing a joint learning framework, comprising: Establishing a federated learning framework, including creating a hierarchy of machine learning models MLM, wherein the hierarchy of MLM includes a first-level MLM in a first-level layer; training the primary MLM, including capturing model contribution updates via at least one communication channel; creating a secondary MLM in a secondary layer of the hierarchy, the secondary layer MLM being operably coupled to the primary MLM via the at least one communication channel, the federated learning framework comprising a hierarchy of nodes and one or more corresponding MLMs; Initializing the created secondary MLM, including cloning the weight and framework of the primary MLM into the secondary MLM, and filling the created secondary MLM with secondary data, wherein the filled data includes model updates for the created secondary MLM; storing the secondary MLM locally in the secondary layer and restricting access to the secondary MLM to the secondary layer; as well as Managing the federated learning framework includes grouping nodes of the federated learning framework according to one or more attributes to form the secondary MLM using clustering techniques, and managing one or more group memberships within one or more formed groups.

14. The method of claim 13, further comprising synchronizing the secondary MLM with the primary MLM, including aggregating weight parameters in a secondary MLM setting, and updating the primary MLM using aggregated local neural network model weights included in the captured contribution data.

15. The method of claim 13, further comprising forming a new secondary MLM, comprising applying a clustering technique within the secondary layer, the applied clustering technique identifying a new secondary MLM.

16. The method of claim 15, wherein applying a clustering technique within the secondary layer further comprises splitting the secondary MLM, the splitting forming at least one new secondary MLM, and training the at least one new secondary MLM.

17. The method of claim 15, wherein applying the clustering technique further comprises combining two or more existing secondary MLMs into a merged secondary MLM.

18. The method of claim 15, further comprising storing the primary MLM data globally, wherein the primary MLM is accessible to the secondary MLM.

19. The method of claim 15, further comprising updating the secondary MLM using a model update local to the secondary MLM.

20. A system for constructing a joint learning framework, the system comprising modules for respectively executing each step of the method according to any one of claims 13-19.

Images