Monitoring focus engine based on machine learning
By integrating text and numerical features in deep neural networks and using word embedding technology, the problem of insufficient flexibility and richness of existing models when processing text features is solved, achieving more efficient state prediction and lower memory usage .
Patent Information
- Application Number
- CN202380032556.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-04-26
- Filing Date
- 2023-02-09
- Publication Date
- 2024-11-26
AI Technical Summary
Existing machine learning models lack flexibility and richness in processing text features and are difficult to provide effective word embedding and interpretability, often relying on simple classification models and term-document matrix, resulting in large memory occupancy and prediction efficiency. Low.
Using a machine learning-based monitoring focus engine, the use of word embedding technology to reduce the dimensions and provide an interpretability layer, allowing users to bring their own data and models to achieve flexible mapping of different inputs and integration.
It improves the representation ability of text features, reduces memory usage, enhances the accuracy and efficiency of prediction, provides a specific explanation of state prediction, and is suitable for system state monitoring in different scenarios.
Smart Images

Figure CN119032360A_ABST
Abstract
Description
Background Art
[0001] Custom rule-based heuristics and machine learning models are used to predict the results of specific scenarios based on building or training such models using specific training data for the analyzed scenarios. In some solutions, simple classification models are trained and utilized to determine the likelihood of results for specific scenarios in a term document matrix format for text features. Additionally, typical custom rule-based heuristics and machine learning models are built from those associated with specific scenarios, and these solutions are implemented separately. All current solutions are custom machine learning models written for specific purposes. Summary of the invention
[0002] This Summary is provided to introduce some concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter.
[0003] The methods, systems, devices, and computer-readable storage media described herein are configured for a machine learning-based monitoring focus engine. Numerical and textual features are collected from (multiple) computing systems (such as different nodes in a cluster) and are used to determine whether (multiple) systems will continue to operate without problems or failures. In other words, external feature information corresponding to a predicted likelihood of a state associated with a processing system is received, and a textual portion and a numerical portion of the external feature information are mapped to inputs of a neural network. Word embedding is performed on the textual portion to generate embedded textual features, and a plurality of inputs are provided to the neural network, wherein the plurality of inputs include at least the embedded textual features, numerical features based on the numerical portion, and local features based on the local feature information. Thus, the predicted likelihood of a state is determined based at least on the output of the neural network from the plurality of inputs. In other words, a machine learning solution is described that allows any number of different users / customers of a machine learning-based monitoring focus engine host (including internal host users / teams) to leverage the same scalable machine learning model for determining predicted likelihood, and the described solution provides a data abstraction layer, enabling users / customers to provide their own data; provides a native model layer, enabling users / customers to provide their own machine learning models as input; enables integration of existing local data on the host platform as well as external data; and provides new text feature processing as word embeddings into a deep neural network (DNN), with an interpretability layer that details the incremental contribution to the predicted likelihood as a unique application.
[0004] Other features and advantages as well as the structure and operation of various example embodiments are described in detail below with reference to the accompanying drawings. Note that example implementations are not limited to the specific embodiments described herein. Such example embodiments are presented herein for illustrative purposes only. Based on the teachings contained herein, additional implementations will be apparent to those skilled in the art(s). BRIEF DESCRIPTION OF THE DRAWINGS
[0005] The accompanying drawings, which are incorporated herein and form a part of the specification, illustrate example embodiments of the present application and, together with the description, also serve to explain the principles of the example embodiments and to enable those skilled in the relevant art to make and use the example embodiments.
[0006] Figure 1 A block diagram of an example network-based computing system configured for a machine learning-based monitoring focus engine is shown according to an example embodiment.
[0007] Figure 2 A flow chart of a method for a machine learning based monitoring focus engine is shown according to an example embodiment.
[0008] Figure 3 Depicted is a system flow diagram showing a sequence of actions performed with respect to a machine learning based monitoring focus engine according to an example embodiment.
[0009] Figure 4 A flow chart of a method for a machine learning based monitoring focus engine is shown according to an example embodiment.
[0010] Figure 5 A flow chart of a method for a machine learning based monitoring focus engine is shown according to an example embodiment.
[0011] Figure 6 Depicted is a system flow diagram showing a sequence of actions performed with respect to a machine learning based monitoring focus engine according to an example embodiment.
[0012] Figure 7 A flow chart of a method for a machine learning based monitoring focus engine is shown according to an example embodiment.
[0013] Figure 8 is a block diagram of an example processor-based computer system that may be used to implement various embodiments.
[0014] Features and advantages of the implementations described herein will become more apparent from the detailed description set forth below when taken in conjunction with the accompanying drawings, wherein like reference numerals identify corresponding elements throughout. In the accompanying drawings, like reference numerals generally indicate identical, functionally similar, and / or structurally similar elements. The drawing in which an element first appears is indicated by the leftmost digit(s) in the corresponding reference numeral. DETAILED DESCRIPTION
[0015] I. Introduction
[0016] This specification and the accompanying drawings disclose many example implementations. The scope of this application is not limited to the disclosed implementations, but also covers combinations of the disclosed implementations and modifications to the disclosed implementations. References to "an implementation", "implementation", "example embodiment", "example implementation" and the like in the specification indicate that the described implementation may include specific features, structures or characteristics, but each implementation may not necessarily include the specific features, structures or characteristics. In addition, such phrases do not necessarily refer to the same implementation. In addition, when specific features, structures or characteristics are described in conjunction with an implementation, whether or not explicitly described, it is claimed that it is within the knowledge of a technician in (multiple) related fields to implement such features, structures or characteristics in conjunction with other implementations.
[0017] In the discussion, unless otherwise stated, adjectives such as "substantially" and "approximately" that modify a condition or relationship characteristic of one or more features of an implementation of the present disclosure should be understood to mean that the condition or characteristic is defined to be within an acceptable tolerance for operation of the implementation for its intended application.
[0018] In addition, it should be understood that the spatial descriptions used herein (such as "above", "below", "upward", "left", "right", "downward", "top", "bottom", "vertical", "horizontal", etc.) are for illustrative purposes only, and actual implementations of the structures described herein may be spatially arranged in any orientation or manner.
[0019] If the performance of an operation is described herein as being "based on" one or more factors, it is understood that the performance of the operation may be based only on such factor(s), or may be based on such factor(s) and one or more additional factors. Thus, as used herein, the term "based on" should be understood to be equivalent to the term "based at least on."
[0020] Many example embodiments are described below. It should be noted that any section / subsection headings provided herein are not intended to be limiting. Implementations are described throughout this document, and any type of implementation may be included under any section / subsection. In addition, implementations disclosed in any section / subsection may be combined in any manner with any other implementations described in the same section / subsection and / or in different sections / subsections.
[0021] II. Example Implementation
[0022] Various aspects described herein are directed to a monitoring focus engine based on machine learning (ML). For example, numerical and textual features are collected from (multiple) computing systems (such as different nodes in a cluster) and are used to determine the predicted likelihood of system states, such as whether (multiple) systems will continue to operate without problems or failures, via a neural network such as a deep neural network (DNN). According to various aspects, the monitoring focus engine based on machine learning is configured to utilize local and external textual data and numerical data and external ML models as inputs to the neural network, where the textual data is words embedded before being input to the neural network. In addition, various aspects provide an interpretability layer that determines and provides incremental contributions of neural network variables to the predicted likelihood. Based on the output of the neural network, actions can be taken for the predicted state.
[0023] Monitoring Focus Engine (MFE) is an end-to-end state-of-the-art machine learning technology that is applicable to different types of systems, services, data, etc. For example, one aspect of the machine learning-based MFE is an application running locally, while another aspect of this MFE is an application running in the cloud (e.g., Microsoft Corp. It collects numerical and textual features from different nodes in a server cluster and uses that information to determine whether the load on the system allows the cluster to operate without issues or failures.
[0024] In general, embodiments of MFEs herein utilize the following aspects: a data abstraction / mapping layer that allows users to "bring their own data" (e.g., third-party or external data that was not considered when developing the MFE), the ability of neural networks in the MFE to incorporate any of the user's custom ML models as features, the use of word embeddings for text features input to the neural network, and the ability of the neural network to utilize an interpretability layer that provides explanations about the incremental contributions of neural network variables to a determined likelihood prediction for a given state.
[0025] The data abstraction / mapping layer allows users to establish data contracts to determine the input pipeline into the MFE, and this applies to text, images, numerical features, etc. For the server cluster workload application example mentioned in this article, this is the "bring your own data" aspect for users to map their specific cluster and node characteristics to feature buckets utilized by the MFE application.
[0026] Thus, the approach to integrating textual and numerical features into DNNs (with an interpretability layer and embeddings for textual features) is a unique application of the described MFE and embeddings to a priori solution problems dominated by traditional term-document matrices. In other words, the MFE solution provided herein is a combination of a data abstraction layer and the ability to bring your own data and / or models and the integration of textual features as embeddings into DNNs, including a novel way of computing embedding size (e.g., by reducing the memory and processing required for larger dimensions) that is more accurate and performs more efficiently than traditional term-document matrices. Additionally, the integrated interpretability layer provides focus on specific variables that affect the state, and this enables specific actions to be taken in response to state predictions.
[0027] Existing solutions utilize simple rule-based heuristics and simple ML classification models that are customized to the specific data to determine the future state. In contrast, various aspects of this document provide a mapping layer that allows any user to map their own tools / data to features suitable for the described MFE model, and allows the use of word embeddings to achieve representation of text data that is multi-dimensional and richer than the term-document matrix model, while also avoiding the memory footprint issues caused by the large dimensionality inherent in other models (such as the "curse of dimensionality" problem), and provides an interpretability layer for DNNs.
[0028] It should be noted herein that the embodiments are generally directed to a machine learning based monitoring focus engine, but the described MFE aspects are applicable to different scenarios for state prediction, including but not limited to uptime prediction for a server cluster or other computing system, prediction of the state of automation components in different systems, and prediction of the state. Therefore, even though the embodiments herein exemplarily describe state prediction for a server cluster, such illustrative example is not limiting, and various aspects herein also contemplate other predicted possibilities as mentioned herein and equivalent and / or similar scenarios that may also utilize the predicted possibilities, such as to take actions based thereon.
[0029] Thus, various aspects herein provide solutions to problems associated with state prediction via a machine learning based monitoring focus engine, and also enable the ability for different users to provide their own data and ML models to an adaptable MFE that is configured to map and merge their unique information. These and other aspects for a machine learning based monitoring focus engine will be described in further detail herein in association with the accompanying figures and the sections / subsections of the description that follow below.
[0030] For example, Figure 1 1 is a block diagram of an example network-based computing system 100 configured for a machine learning-based monitoring focus engine according to an example embodiment. Figure 1 As shown in FIG. 1 , system 100 includes a plurality of clusters 102A, 102B, and 102N and a storage cluster 124. Each of clusters 102A, 102B, and 102N and storage cluster 124 are communicatively coupled to each other via a network 116. Network 116 may include one or more networks, such as a local area network (LAN), a wide area network (WAN), an enterprise network, the Internet, etc., and may include one or more of wired and / or wireless portions.
[0031] Clusters 102A, 102B, and 102N and / or storage cluster 124 may form a network-accessible collection of servers (e.g., a distributed or cloud-based environment or service platform (e.g., an environment / platform hosting resources, services, and / or applications). Each of clusters 102A, 102B, and 102N may include a set of one or more nodes (also referred to as computing nodes) and / or a set of one or more storage nodes. For example, Figure 1 As shown in , cluster 102A includes nodes 108A to 108N, cluster 102B includes nodes 112A to 112N, and cluster 102N includes nodes 114A to 114N. Each of nodes 108A to 108N, nodes 112A to 112N, and / or nodes 114A to 114N is accessible via network 116 (e.g., in a "cloud-based" embodiment) to build, deploy, and manage applications and services. Storage cluster 124 includes one or more storage nodes 110A to 110N. Each of the (multiple) storage nodes 110A to 110N includes multiple physical storage disks accessible via network 116 and is configured to store data associated with applications and services managed by nodes 108A to 108N, nodes 112A to 112N, and / or nodes 114A to 114N.
[0032] As mentioned above, in an embodiment, the system 100 may include one or more distributed or "cloud-based" servers. In other words, the system 100 may be a network or "cloud" implementation for applications and / or services associated with hosted databases and / or data warehouses in a network architecture / cloud platform. According to an embodiment, the cloud platform includes a collection of networked computing resources, including servers, routers, etc., which are configurable, shareable, provide data security, and are accessible via a network such as the Internet. Cloud applications / services for hosting databases, data warehouses, etc. are configured to run on these computing resources, typically running on an operating system running on the resources, for entities accessing the applications / services locally and / or over a network.
[0033] The cloud platform is configured to support multi-tenancy as mentioned above, where the cloud platform-based software services serve multiple tenants, each tenant including one or more users who share common access to certain software services and applications of the cloud platform, as mentioned herein. In addition, the cloud platform is configured to support hypervisors implemented as hardware, software and / or firmware that run virtual machines (emulated computer systems, including operating systems) for tenants. The hypervisor presents a virtual operating platform to the tenants in the cloud platform.
[0034] In an embodiment, one or more of clusters 102A, cluster 102B, and cluster 102N and / or storage cluster 124 may be co-located (e.g., housed in one or more nearby buildings with associated components such as backup power, redundant data communications, environmental control, etc.) to form a data center, or may be arranged in other ways. Thus, in an embodiment, one or more of clusters 102A, cluster 102B, and cluster 102N and / or storage cluster 124 may be a data center in a distributed collection of data centers.
[0035] Each of the node(s) 108A to 108N, the node(s) 112A to 112N, and the node(s) 114A to 114N may include one or more server computers, server systems, and / or computing devices. Each of the node(s) 108A to 108N, the node(s) 112A to 112N, and the node(s) 114A to 114N may be configured to execute one or more software applications (or “applications”) and / or services and / or manage hardware resources (e.g., processors, memory, etc.) that may be utilized by users of the network-accessible collection of servers (e.g., customers or tenants in a cloud-based platform). The node(s) 108A to 108N, the node(s) 112A to 112N, and the node(s) 114A to 114N may also be configured for specific uses. For example, Figure 1As shown in , node 108A can be configured to execute monitoring focus engine (MFE) 118, node 108B can be configured to execute (multiple) monitors 132, node 112B can be configured to execute identity service 128, and node 114N can be configured to execute management tool 120. It is noted that instances of MFE 118, (multiple) monitors 132, management tool 120 and / or identity service 128 can be executed on (multiple) other nodes (e.g., (multiple) nodes 108B to 108N, (multiple) nodes 112A to 112N and / or (multiple) nodes 114A to 114N) to replace or in addition to the above-mentioned nodes, respectively. It is also noted that one or more of these components can be merged with each other.
[0036] The identity service 128 may be configured to maintain multiple user identities by which associated users may be utilized to access one or more devices, applications, and / or services maintained by the system 100 and / or associated with the identity service 128 (e.g., web applications and / or services hosted and / or executed by any of the node(s) 108A to 108N, the node(s) 112A to 112N, and / or the node(s) 114A to 114N). Similarly, the identity service 128 may be configured to maintain multiple workload identities and associated credentials that may be used for authentication and access. In response to successful authentication, access to the device, application, and / or service is provided to the account / workload identity. Examples of the identity service 128 may include a directory such as, but not limited to, the Microsoft® Directory Directory published by Microsoft Corporation of Redmond, Washington. Azure Active Directory TM . User data associated with their identity may be used herein as local data for input to the neural network of the MFE 118, which local data may be stored as the DNN 130 in the storage node 110N.
[0037] The management tool 120 may be configured to provide applications and / or services by which user data is collected. In other words, the management tool 120 may monitor user accounts and utilized applications / services to collect data thereof, which may be used herein as local data for input to the neural network of the MFE 118. The local data collected by the management tool 120 may be stored within the system 100 at one or more storage nodes of the storage cluster 124. It is also contemplated herein that external data (e.g., third-party or user data) may be used as input to the MFE 118 in various aspects.
[0038] Monitor(s) 132 are configured to monitor the performance of a server, server cluster, node, etc. Monitor(s) 132 may detect and / or track characteristics and information such as, but not limited to, data bandwidth, latency, throughput, memory occupancy / usage, processor usage, hard drive usage, utilization percentage for memory and / or processor, type of server, central processing unit (CPU) and / or graphics processing unit (GPU), GPU load, number of running processes, etc. Monitor(s) 132 may also be configured to monitor access to and / or actions taken with respect to applications, services, and / or data objects (e.g., managed by such applications and / or services). Additionally, monitor(s) 132 and / or each individual node may collect / generate data for Figure 1 , which includes status and operational information of the node, as understood by those skilled in the relevant art(s) having the benefit of instant disclosure. Monitor(s) 132 may be implemented at one or more clusters / nodes, including in each cluster / node of system 100. The text log file may be stored in a storage node, as exemplarily shown for storage node 110B or elsewhere in various embodiments, and the detected characteristics and information may be stored as performance data 106, as exemplarily shown for storage device 110a or elsewhere in various embodiments.
[0039] In some aspects, text log files 104 and performance data 106 may be information and data external to system 100, such as information and data collected by a user and provided to MFE 118 for predictive analysis.
[0040] Now refer to Figure 2 , a flowchart 200 of a method for monitoring a focus engine based on machine learning is shown according to an example embodiment. In various aspects, the flowchart 200 is composed of Figure 1 118 of the system 100 shown in FIG. 100 , but the method is not limited to this implementation. Figure 1 The flowchart 200 is described as an example. Figure 1
[0036] Given the discussion of flowchart 200 and system 100, other structural and operational embodiments will be apparent to persons skilled in the relevant art(s).
[0041] Flowchart 200 begins at step 202. In step 202, input is provided to a neural network, including external numerical and textual features mapped from a computing system(s), the textual features including word embeddings. For example, in various aspects, textual log files 104 and performance data 106 and external data / ML models may be provided to MFE 118 for input to its neural network, such as DNN 130, wherein the text of textual log files 104 undergoes word embeddings prior to consumption by the neural network, and the external data features are mapped to DNN 130 so that the features can be utilized by DNN 130 (otherwise they would not be utilized).
[0042] In step 204, a predicted likelihood is determined by the neural network based at least on the provided input. For example, the neural network of MFE 118 is configured to determine a predicted likelihood based on the input in step 202, such as a predicted likelihood related to health or uptime for a server / server cluster.
[0043] Further details about MFE 118 and flow chart 200 are provided below with reference to the described figures. For example, the following will now be described in this context. Figure 3 and Figure 4 .
[0044] Figure 3 A system flow diagram 300 is depicted showing a sequence of actions performed by a machine learning based monitoring focus engine according to an example embodiment. The system flow diagram 300 is based on a monitoring focus engine (MFE) 318, which is Figure 1 Embodiment of the MFE 118. Figure 4 A flowchart 400 is shown for a method for a machine learning based monitoring focus engine according to an example embodiment. In various aspects, the flowchart 400 is composed of Figure 3 The method is implemented by MFE 318, but the method is not limited to this implementation. Figure 1 and Figure 3 The flowchart 400 is described as an example. Figure 1
[0043] Given the discussion of system 100, system flow diagram 300, and flow diagram 400, other structural and operational embodiments will be apparent to those skilled in the relevant art(s).
[0045] As shown in the system flow diagram 300, the MFE 318 includes a mapping layer 302, a word embedding layer 304, and a deep neural network (DNN) 306 (which can be Figure 1 130 ). The MFE 318 receives various inputs described below to generate outputs as state probabilities 308 representing the predicted likelihood of a state.
[0046] MFE 318 is shown receiving textual data 310, numerical data 312, and external data 314, which are employed by mapping layer 302 and which represent, for example, characteristics of the system for which a predicted likelihood is to be determined.
[0047] Now also refer to Figure 4 , flowchart 400 begins at step 402, where local and external characteristic information corresponding to a predicted likelihood of a state associated with a processing system is received. Figure 3 In the context of system flow diagram 300, text data 310 may include, among other text data, Figure 1 The text log data 104; the numerical data 312 may include, among other numerical data, Figure 1 performance data 106; and external data may include one or both of these data types provided by a user from outside the system (ie, the user brings his own data to the MFE 318).
[0048] The mapping layer 302 is configured to map the provided data and information to features that can be utilized by the MFE 318. For example, referring again to Figure 4 , in step 404, the text portion and the numerical portion of the feature information are mapped to the corresponding inputs of the neural network. The mapping layer 302 is configured to map the text portion 320a of the text data 310 (including the combined text portion of the data) and the numerical portion 324a of the numerical data 312 to the input of the DNN 306. Additionally, the mapping layer 302 is configured to map the text portion 320b (including the combined text portion of the data) and the numerical portion 324b of the external data 314 to the input of the DNN 306. In other words, the mapping layer 302 implements mapping the available features of the user's external data to the features to be used by the model.
[0049] The word embedding layer 304 is configured to perform word embedding on the text portion of the input data. Figure 4 In step 406, word embedding is performed on the mapped text portion to generate embedded text features. According to various aspects, the word embedding layer 304 is configured to receive the mapped text portion 320a and the mapped text portion 320b, and is configured to generate or derive embedded text features 322, which are rich multi-dimensional representations of the text input.
[0050] As mentioned above, a traditional way of representing text features is to use a term-document matrix (TDM), where each term in the vocabulary is represented in a column. Traditional methods have all the processing text using the TDM and are also referred to as "dummy" variables. In contrast and as exploited by various aspects herein, word embeddings map m terms to n embeddings, where n << m. This provides benefits compared to previous TDM solutions. First, it reduces the dimensionality, helping to avoid the "curse of dimensionality" problem, which is a common pitfall in ML applications. Second, it helps to learn a richer representation of the text input. In other words, for a TDN model, each text term (e.g., a word) is mapped to a column / feature, while in word embeddings, each word is mapped to the entire embedding size, thus learning a richer representation.
[0051] Various aspects herein also provide the ability to input an external model 316 into the DNN 306. In other words, the "bring your own model" ability allows a user / customer to bring their external ML model. In some aspects, such an external ML model may be allowed provided that: their object or container type matches the DNN 306, they can be called using an application programming interface (API) endpoint when implementing a call to such a model, and / or the time period for which they are used for training and inference is the same as or substantially the same as the time period for the model used for the DNN 306 to avoid any unintentional leakage.
[0052] Although not shown for illustrative clarity, the MFE 318 may, in various aspects, include a featureizer that is configured to generate one or more feature vectors for input into the DNN 306 based on the text data 310, numerical data 312, external data 314, and / or external model 316. The feature vectors generated by such a featureizer may take any form, such as numerical, visual, and / or text representations, or may include any other form suitable for representing audit data. The featureizer may operate in several ways to characterize or generate feature vectors for a given data set. For example but not limited to, the featureizer may be configured to characterize a given data set by time series analysis, keyword characterization, semantic-based characterization, digital count characterization, n-gram TFIDF characterization, etc.
[0053] The DNN 306 is configured to receive multiple inputs and perform operations on them. For example, again referring to Figure 4 , in step 408, multiple inputs are provided to the neural network, the multiple inputs including at least embedded text features and numerical features based on the numerical portion. As Figure 3As shown in the system flow diagram 300, the DNN 306 of the MFE 318 takes the embedded text features 322, the numerical portion 324a, and the numerical portion 324b as inputs. Therefore, the DNN 306 operates on these inputs and uses the target variable as the subsequent N cycles of feature creation points. The cycle can be in units of days, weeks, months, etc. or any other time window.
[0054] In various aspects, DNN 306 includes one or more layers, wherein each layer in DNN 306 performs an affine transformation (W.X+b) followed by a nonlinear operation ReLU, wherein
[0055] ReLU(x)=0 if x≤0 and x if x>0. (Equation 1)
[0056] It should be noted that W is the set of unknown weights that the DNN will seek to learn, X is the input, and b is the bias associated with the layer. The first layer of the DNN 306 has the numerical features as input and the text features that hit the embedded text features. The embedded text features can be mathematically represented as X e < <X text Therefore, word embedding is a compressed high-dimensional representation of the input text. If X numeric represents a numerical feature, then the first layer of DNN 306 will be after affine transformation
[0057] W1*(X numeric +W e (X text )) (Equation 2)
[0058] It should be noted that this will be followed by a non-linear operator such as the ReLU function.
[0059] Each subsequent layer performs the affine transformation and nonlinear operation steps mentioned above. For example, in a scenario where DNN 306 includes three hidden layers, DNN 306 will calculate the following form:
[0060] σ(W3*(ReLU(W2*(ReLU(W1-(X numeric +W e (X text ))+b1))+b2)+b3). (Equation 3)
[0061] Therefore, DNN 306 is configured to generate state probabilities 308. For example, and again referring to Figure 4 , in step 410, a predicted likelihood of a state is determined based at least on an output of the neural network from a plurality of inputs. Figure 3The state probability 308 in represents the predicted likelihood of the state to be determined by the DNN 306.
[0062] Generating state probabilities 308 includes multiple sub-steps in various aspects. For example, DNN 306 finds unknown weights by a process of stochastic gradient descent, including but not limited to the following: Initializing the unknowns to random values, where the unknowns in the above equation are weights (W i , where i is the layer and i=e means the embedding in the first layer) and the bias (b i , where i is a layer); computing predicted values for observations through the forward pass described above, e.g., as mentioned for Equation 3; computing a loss by comparing the output of the forward pass with the actual output; backpropagating the loss through each layer; and repeating these substeps until convergence is achieved by DNN 306.
[0063] The state probability 308 may indicate a prediction by the DNN 306 that a state is likely or unlikely. Figure 4 and flowchart 400, in step 412, performing a mitigation action based at least on the predicted likelihood of the state indicating an unfavorable state, the mitigation action comprising one or more of predictive maintenance, load balancing, changing the schedule, upgrading, or increasing resource capacity. For example, in an exemplary scenario of the health / availability of a server or server cluster, the state probability 308 may indicate that the server or server cluster is likely to be up and running after a period of time, or that the server is likely to fail / crash within such a time period, or have issues that affect / degrade performance. In the event that a failure or performance degradation is predicted, the MFE 318 may trigger or provide communication to a system component so that a remedial action is performed to attempt to prevent the failure / degradation (e.g., by providing the state probability 308 or an indicia thereof to the system component taking the remedial action).
[0064] Additionally, in some aspects, the state probability 308 or an indicia thereof can be provided to a user interface of a display via an alert, email, text message, etc., or in some aspects to a prediction log file for an administrative user to be aware of the state probability 308. The administrative user can then take manual proactive steps and / or provide support to users whose systems are predicted to be adversely impacted during the time period.
[0065] Thus, the machine learning based monitoring focus engine aspects described herein improve system availability and operation by at least enabling proactive predictive maintenance, load balancing and / or scheduling improvements, ease of use and determination of hardware and software upgrades, capacity additions for system resources, enabling proactive support, etc. In this manner, a user or organization can respond to future failure events by having the machine learning based monitoring focus engine monitor logs and performance characteristics and additionally monitor external user data, and take proactive and preventative actions accordingly (automatically or manually, e.g., based on provided alerts).
[0066] Figure 5 A flowchart 500 is shown for a method for a machine learning based monitoring focus engine according to an example embodiment. In various aspects, the flowchart 500 is composed of Figure 3 The MFE 318 is implemented, for example, via the mapping layer 302, but the method is not limited to this implementation. Therefore, reference will continue to be made to Figure 1 and Figure 3 The flowchart 500 is described as an example. Figure 1 With reference to the discussion of the system 100 and the system flow chart 300, other structural and operational embodiments will be apparent to those skilled in the relevant art(s). In various aspects, the flow chart 500 is Figure 4 An embodiment of flowchart 400 (such as step 404).
[0067] In step 502, a mapping is performed via an abstraction layer, wherein the textual portion and the numerical portion of the external information are incompatible inputs to the neural network without the mapping. As described herein, the mapping herein is performed by Figure 3 According to step 502, the mapping layer 302 is configured to utilize the abstraction layer in the mapping to link external information to the Figure 3 In other words, the mapping layer 302 implements mapping the available features of the user's external data to the features to be used by the model to allow the bring your own data aspect described herein.
[0068] The flowchart 500 may further include an additional sub-step 504 for step 502. In sub-step 504, available features of the text portion and the numerical portion of the external information are mapped to embedded text features and numerical features, respectively.
[0069] We will now describe the explainability layer for the machine learning-based monitoring focus engine. Figure 6 and Figure 7 . Figure 6A system flow diagram 600 is depicted that illustrates a sequence of actions performed by a machine learning based monitoring focus engine according to an example embodiment. Figure 3 Part of the monitoring focus engine (MFE) 318. Figure 7 A flowchart 700 is shown for a method for a machine learning based monitoring focus engine according to an example embodiment. In various aspects, the flowchart 700 is composed of Figure 3 The MFE 318 implementation is described above, but the method is not limited to this implementation. Figure 1 and Figure 3 And in Figure 6 Flowchart 700 is exemplarily described in the context of . Based on the discussion of systems, flowcharts, and flow diagrams herein, other structural and operational embodiments will be apparent to those skilled in the relevant art(s).
[0070] As shown, the system flow diagram 600 includes the DNN 306 and the output (state probability 308), as described above for Figure 3 Also shown for the system flow diagram 600 is an interpretability layer 602 and an incremental contribution 604. The DNN 306 can provide variable information 606 to the interpretability layer 602. In some aspects, the variable information 606 can include state probabilities 308.
[0071] Also refer to Figure 7 , in step 702, the incremental contributions of the neural network variables to the predicted likelihood of the state are determined via the interpretability layer, respectively. For example, the interpretability layer 602 is configured to generate the incremental contributions 604 of the neural network variables to the predicted likelihood of the state. In various aspects, the incremental contributions 604 are based on at least the variable information 606, wherein the interpretability layer 602 determines the incremental contributions 604 as a profile of local importance, in other words, why a particular predicted likelihood has a higher or lower score than a baseline predicted likelihood or another predicted likelihood.
[0072] In various aspects, such as Figure 6 As shown in , the MFE 318 is configured to provide incremental contributions 604 for enhancing decision making and / or remedial actions. Figure 7and flowchart 700, in step 704, each incremental contribution is provided to at least one of a user interface or a prediction log file. For example, in some aspects, the MFE 318 is configured to provide the incremental contributions 604 and / or indicia thereof to a user interface or a prediction log file (not shown). The incremental contributions 604 and / or indicia thereof may be provided to a user interface of a display via an alert, email, text message, etc., or to a predictive log file in some aspects for an administrative user to understand the contributions provided by different neural network variables to a predictive likelihood output (e.g., state probabilities 308). The administrative user can then take manual proactive steps and / or provide support to users whose systems are predicted to be adversely affected by system features linked to influential variable values.
[0073] In some aspects, local interpretability for neural network variables is determined by treating each prediction as a game played between game participants (in this case, feature values) and using the Shapley technique to determine the incremental contribution of each of the variables.
[0074] Thus, a system that performs a method for a machine learning based monitoring focus engine may be implemented in a variety of ways.
[0075] III. Example Computer System Implementation
[0076] Refer to above Figures 1 to 7 The described systems and methods may be implemented in hardware or in hardware in combination with one or both of software and / or firmware. Figure 8 The system 800 may be used to implement Figure 1 nodes 108A to 108N, nodes 112A to 112N and / or nodes 114A to 114N, storage node(s) 110A to 110N, monitoring focus engine 118, monitor(s) 132, management tool 120, and identity service 128, Figure 3 Monitoring focus engine 318, mapping layer 302, embedding layer 304, characterizer and / or DNN 306, Figure 6 Any of the interpretability layers 602 and any of its components and / or subcomponents and the flowcharts / flowcharts described herein (including parts thereof) and / or other examples described herein may be implemented as computer program codes / instructions, respectively, which are configured to be executed in one or more processors and stored in a computer-readable storage medium. Alternatively, Figure 1 nodes 108A to 108N, nodes 112A to 112N and / or nodes 114A to 114N, storage node(s) 110A to 110N, monitoring focus engine 118, monitor(s) 132, management tool 120, and identity service 128, Figure 3 Monitoring focus engine 318, mapping layer 302, embedding layer 304, characterizer and / or DNN 306, Figure 6 Any of the interpretability layers 602 and any of its components and / or subcomponents and the flowcharts / flowcharts described herein (including portions thereof) and / or other examples described herein may be implemented as hardware logic / electronic circuit devices. In an embodiment, Figure 1 nodes 108A to 108N, nodes 112A to 112N and / or nodes 114A to 114N, storage node(s) 110A to 110N, monitoring focus engine 118, monitor(s) 132, management tool 120, and identity service 128, Figure 3 Monitoring focus engine 318, mapping layer 302, embedding layer 304, characterizer and / or DNN 306, Figure 6 Any of the interpretability layer 602 and any of its components and / or subcomponents and the flowcharts / flowcharts described herein (including portions thereof) and / or other examples described herein may be implemented in one or more SoCs (systems on a chip). A SoC may include an integrated circuit chip that includes a processor (e.g., a central processing unit (CPU), a microcontroller, a microprocessor, a digital signal processor (DSP), etc.), a memory, one or more communication interfaces, and / or other circuits, and may optionally execute received program code and / or include embedded firmware to perform functions.
[0077] Figure 8 An exemplary implementation of a computing device 800 in which embodiments may be implemented is depicted, including Figure 1 nodes 108A to 108N, nodes 112A to 112N and / or nodes 114A to 114N, storage node(s) 110A to 110N, monitoring focus engine 118, monitor(s) 132, management tool 120, and identity service 128, Figure 3 Monitoring focus engine 318, mapping layer 302, embedding layer 304, characterizer and / or DNN 306, Figure 6 Any of the interpretability layers 602 and any of its components and / or subcomponents and the flowcharts / flowcharts described herein (including portions thereof) and / or other examples described herein. The description of the computing device 800 provided herein is provided for illustrative purposes and is not intended to be limiting. As known to those skilled in the relevant art(s), the embodiments may be implemented in other types of computer systems.
[0078] like Figure 8As shown in FIG. 8 , computing device 800 includes one or more processors (referred to as processor circuit 802), system memory 804, and bus 806, which couples various system components including system memory 804 to processor circuit 802. Processor circuit 802 is an electrical and / or optical circuit implemented as a central processing unit (CPU), microcontroller, microprocessor, and / or other physical hardware processor circuit in one or more physical hardware electrical circuit device elements and / or integrated circuit devices (semiconductor material chips or dies). Processor circuit 802 can execute program codes stored in computer-readable media, such as program codes of operating system 830, application programs 832, other programs 834, etc. Bus 806 represents one or more of any of several types of bus structures, including memory buses or memory controllers, peripheral buses, accelerated graphics ports, and processor or local buses using any of various bus architectures. System memory 804 includes read-only memory (ROM) 808 and random access memory (RAM) 810. Basic input / output system 812 (BIOS) is stored in ROM 808.
[0079] The computing device 800 also has one or more of the following drives: a hard disk drive 814 for reading from and writing to a hard disk, a disk drive 816 for reading from or writing to a removable disk 818, and an optical drive 820 (such as a CDROM, DVD ROM, or other optical media) for reading from or writing to a removable optical disk 822. The hard disk drive 814, the disk drive 816, and the optical drive 820 are connected to the bus 806 by a hard disk drive interface 824, a disk drive interface 826, and an optical drive interface 828, respectively. The drives and their associated computer-readable media provide non-volatile storage of computer-readable instructions, data structures, program modules, and other data for the computer. Although a hard disk, a removable disk, and a removable optical disk are described, other types of hardware-based computer-readable storage media may be used to store data, such as flash memory cards, digital video disks, RAM, ROM, and other hardware storage media.
[0080] Several program modules may be stored on a hard disk, a magnetic disk, an optical disk, a ROM, or a RAM. These programs include an operating system 830, one or more application programs 832, other programs 834, and program data 836. Application programs 832 or other programs 834 may include, for example, a system for implementing the system described above (including the system described above with reference to Figures 1 to 7 The computer program logic (eg, computer program code or instructions) of the described embodiments.
[0081] A user may enter commands and information into the computing device 800 through input devices such as a keyboard 838 and a pointing device 840. Other input devices (not shown) may include a microphone, a joystick, a game pad, a satellite dish, a scanner, a touch screen and / or a touch pad, a voice recognition system to receive voice input, a gesture recognition system to receive gesture input, etc. These and other input devices are typically connected to the processor circuit 802 through a serial port interface 842 coupled to the bus 806, but may be connected by other interfaces such as a parallel port, a game port, or a universal serial bus (USB).
[0082] The display screen 844 is also connected to the bus 806 via an interface such as a video adapter 846. The display screen 844 may be external to the computing device 800 or incorporated into the computing device 800. The display screen 844 may display information, as well as be a user interface for receiving user commands and / or other information (e.g., by touch, finger gestures, a virtual keyboard, by providing a tap input (in which the user lightly presses and quickly releases the display screen 844), by providing a "touch and hold" input (in which the user touches and holds his finger (or touch instrument) on the display screen 844 for a predetermined period of time), by providing a touch input that exceeds a predetermined pressure threshold, etc.). In addition to the display screen 844, the computing device 800 may include other peripheral output devices (not shown), such as speakers and printers.
[0083] The computing device 800 is connected to a network 848 (eg, the Internet) through an adapter or network interface 850, a modem 852, or other means for establishing communications over the network. Figure 8 As shown, the modem 852, which may be internal or external, may be connected to the bus 806 via the serial port interface 842, or may be connected to the bus 806 using another type of interface including a parallel interface.
[0084] As used herein, the terms "computer program medium", "computer readable medium" and "computer readable storage medium" and the like are generally used to refer to physical hardware media, such as a hard disk associated with hard disk drive 814, a removable disk 818, a removable optical disk 822, other physical hardware media, such as RAM, ROM, flash memory cards, digital video disks, zip disks, MEMs, nanotechnology-based storage devices, and other types of physical / tangible hardware storage media (including Figure 8The system memory 804 of the present invention. Such computer-readable media, computer-readable storage media, etc. are distinguished from and do not overlap with (do not include) communication media and propagated signals. Communication media typically implement computer-readable instructions, data structures, program modules, or other data in a modulated data signal (such as a carrier wave). The term "modulated data signal" means a signal having one or more of its characteristics set or changed in such a manner as to encode information in the signal. By way of example and not limitation, communication media include wireless media (such as acoustic, RF, infrared, and other wireless media) and wired media. Embodiments are also directed to such communication media, which are separate from and do not overlap with embodiments directed to computer-readable storage media.
[0085] As mentioned above, computer programs and modules (including application programs 832 and other programs 834) can be stored on a hard disk, disk, optical disk, ROM, RAM or other hardware storage medium. Such computer programs can also be received via a network interface 850, a serial port interface 852 or any other interface type. Such computer programs, when executed or loaded by an application, enable the computing device 800 to implement the features of the embodiments discussed herein. Therefore, such computer programs represent a controller of the computing device 800.
[0086] Embodiments are also directed to computer program products including computer code or instructions stored on any computer readable medium, including hard drives, optical drives, memory device packages, portable memory sticks, memory cards, and other types of physical storage hardware.
[0087] IV. Other Example Embodiments
[0088] As described, systems and devices implementing the techniques herein can be configured and implemented in various ways to perform corresponding functions for a machine learning-based monitoring focus engine. In an embodiment, one or more of the steps or operations of any flowchart and / or flow chart described herein may not be performed. In addition, these steps or operations may be performed in addition to or instead of the steps or operations in any flowchart and / or flow chart described herein. In addition, in examples, one or more operations of any flowchart and / or flow chart described herein may be performed out of order, in an alternative sequence, or partially (or completely) concurrently with each other or with other operations.
[0089] As described herein, systems, devices, components and the like of the embodiments configured to perform the functions and / or operations are also contemplated as performing such functions and / or operations.
[0090] Prior solutions fail to adequately address the flexibility and richness of inputs into neural networks, and do not adequately provide word embeddings and interpretability in neural network predictions. Prior solutions focus on custom ML models for specific data and purposes, which still rely on simple classification to achieve predictions using term-document matrices to process text features. In contrast, various aspects of this paper utilize a scalable ML-based monitoring focus engine that allows different users of the cloud platform (e.g., the platform's internal users and its customers) to utilize a single solution that can map and accept different inputs, as well as include word embeddings for richer text features that are balanced for dimensionality (which reduces memory usage).
[0091] In other words, the mapping / data abstraction layer for external data, the ability to input external models, and the ability to integrate a user's local data already on the cloud platform into a deep neural network (DNN) with an interpretability layer have no analogs in prior solutions. These and other improvements to the scalable ML-based monitoring focus engine described herein were not previously available to software-based platforms and networks, let alone the specific system implementation, organization, and architecture in the embodiments described herein.
[0092] Aspects of this article also apply to features within management tools / software, e.g. or Dynamics A customer relationship management (CRM) tool from Microsoft Corp., Redmond, Wash., to again leverage text and numeric features to determine the likelihood of an opportunity closing, such as the status of an opportunity. As mentioned in the article, the data abstraction / mapping layer allows users to establish data contracts to determine the input pipeline into the MFE, and this applies to text, image, and numeric features alike.
[0093] For management tools / software applications, this is done on top of existing tools (e.g. ) that allows any of the users / clients to run a chance-possible-close model that is an application on top of the base CRM tool. In other words, the user / client is able to bring their own additional data on top of the data already in the platform hosting the tool for their specific instance and still use the provided MFE model as described herein.
[0094] The mapping layer allows any user / organization utilizing a CRM tool to map its own features, including the text features described herein. In other words, it is not easy for each user / organization to analyze text features, for example due to lack of budget to hire data scientists and ML engineers. Some of these text features considered according to various aspects of this article include free-form text in various fields, such as those used by sellers to include information about sales opportunities. These include text information about budgets, competitor solutions, reasons for high / low closing opportunities, dissatisfaction, etc.
[0095] This article may also include additional features from the opportunity, such as sales information (order history, product purchase history, sales by product, etc.), company statistics (industry vertical, number of employees, etc.), marketing response, clickstream / digital features, seller characteristics, sales progress inputs, opportunity milestone fields in dynamic tools, third-party data sets available in CRM tools, customer satisfaction (Net Promoter Score) data, etc.
[0096] Therefore, desired business outcomes, such as a sales quadrant visualization, i.e., a 2x2 matrix of seller confidence versus predicted likelihood (low-low, low-high, high-low, and high-high), likelihood of opportunity win, inferences about opportunity scores, early deal quality assessments, and improved sales staff assignments, which make the predicted likelihood more actionable and valuable to the user / client. As mentioned above, the 2x2 sales quadrant visualization includes four cases. If the model predicted likelihood and the seller closing prediction are consistent (high-high or low-low), no action is required based on the agreement. However, in these inconsistent cases (e.g., high prediction scores and low seller closing scores, or vice versa), further exploration of opportunities for action is required. With respect to the likelihood of opportunity win, more personnel can be assigned to opportunities that are more likely to be closed.
[0097] The additional examples and embodiments described in this section may be applicable to the examples disclosed in any other sections or subsections of this disclosure.
[0098] Embodiments in the present description provide systems, devices, and methods for a machine learning-based monitoring focus engine. For example, such a system is described herein. The system includes at least one memory storing program code and a processing system including one or more processors, the processing system being configured to receive the program code from the memory and to perform operations in response to at least receiving the program code. The operations include receiving feature information, including local and external information corresponding to predicted probabilities associated with a state of the processing system, and mapping a text portion and a numerical portion of the feature information to corresponding inputs of a neural network. The operations also include performing word embedding on the mapped text portion to generate embedded text features, providing a plurality of inputs to the neural network, the plurality of inputs including at least embedded text features and numerical features based on the numerical portion, and determining the predicted probabilities of the state based at least on the output of the neural network from the plurality of inputs.
[0099] In an embodiment of the system, the processing system is configured in response to at least receiving the program code to perform a mitigation action based at least on the predicted likelihood of the state indicating an unfavorable state, the mitigation action comprising one or more of predictive maintenance, load balancing, changing scheduling, upgrading, or increasing resource capacity.
[0100] In an embodiment of the system, at least one of the plurality of inputs to the neural network further comprises an external machine learning model.
[0101] In an embodiment of the system, the embedded text has a reduced input dimensionality and has a mapping range that is larger than the text portion via the embedding size.
[0102] In an embodiment of the system, mapping the text portion and the numerical portion of the feature information to inputs of the neural network includes mapping via an abstraction layer, wherein without the mapping, the text portion and the numerical portion are incompatible inputs to the neural network.
[0103] In an embodiment of the system, the abstraction layer is configured to map available features of the text portion and the numerical portion to embedded text features and numerical features, respectively.
[0104] In an embodiment of the system, the processing system is configured in response to at least receiving the program code to: determine, via the interpretability layer, incremental contributions of the neural network variables to the predicted likelihood of the state, respectively, and provide each incremental contribution to at least one of a user interface or a prediction log file.
[0105] A method performed by a computing system is also described. The method includes receiving feature information, including local and external information corresponding to a predicted likelihood of a state associated with a processing system, and mapping a text portion and a numerical portion of the external feature information to corresponding inputs of a neural network. The method also includes performing word embedding on the text portion to generate embedded text features, providing a plurality of inputs to the neural network, the plurality of inputs including at least the embedded text features and numerical features based on the numerical portion, and determining the predicted likelihood of the state based at least on an output of the neural network from the plurality of inputs.
[0106] In an embodiment, the method further comprises performing a mitigation action based at least on the predicted likelihood of the state indicating an unfavorable state, the mitigation action comprising one or more of predictive maintenance, load balancing, changing scheduling, upgrading, or increasing resource capacity.
[0107] In an embodiment of the method, at least one of the plurality of inputs to the neural network further comprises an external machine learning model.
[0108] In an embodiment of the method, the embedded text has a reduced input dimensionality and has a mapping range larger than the text portion via the embedding size.
[0109] In an embodiment of the method, mapping the textual portion and the numerical portion of the feature information to an input of the neural network comprises mapping via an abstraction layer, wherein without said mapping, the textual portion and the numerical portion are incompatible inputs to the neural network.
[0110] In an embodiment of the method, the abstraction layer is configured to map available features of the text part and the numerical part to embedded text features and numerical features, respectively.
[0111] In an embodiment, the method further comprises determining, via the interpretability layer, incremental contributions of the neural network variables to the predicted likelihood of the state, respectively, and providing each incremental contribution to at least one of a user interface or a prediction log file.
[0112] A computer readable storage medium is also described. The computer readable storage medium has program instructions recorded thereon that, when executed by at least one processor of a computing system, perform a method. The method includes receiving feature information, including local and external information corresponding to a predicted likelihood of a state associated with a processing system, and mapping a text portion and a numerical portion of the external feature information to corresponding inputs of a neural network. The method also includes performing word embedding on the text portion to generate embedded text features, providing a plurality of inputs to the neural network, the plurality of inputs including at least the embedded text features and numerical features based on the numerical portion, and determining a predicted likelihood of the state based at least on an output of the neural network from the plurality of inputs.
[0113] In an embodiment of the computer-readable storage medium, the method further comprises at least one of: performing a mitigation action based at least on the predicted likelihood of the state indicating an unfavorable state, the mitigation action comprising one or more of predictive maintenance, load balancing, scheduling changes, upgrades, or resource capacity increases; or determining, via an interpretability layer, incremental contributions of neural network variables to the predicted likelihood of the state, respectively, and providing each incremental contribution to at least one of a user interface or a prediction log file.
[0114] In an embodiment of the computer-readable storage medium, at least one of the plurality of inputs to the neural network further comprises an external machine learning model.
[0115] In an embodiment of the computer readable storage medium, the embedded text has a reduced input dimensionality and has a mapping range larger than the text portion via the embedding size.
[0116] In an embodiment of the computer-readable storage medium, mapping the text portion and the numerical portion of the feature information to an input of the neural network includes mapping via an abstraction layer, wherein without the mapping, the text portion and the numerical portion are incompatible inputs to the neural network.
[0117] In an embodiment of the computer-readable storage medium, the abstraction layer is configured to map available features of the text portion and the numerical portion to embedded text features and numerical features, respectively.
[0118] V. Conclusion
[0119] Although various example embodiments have been described above, it should be understood that they are presented by way of example only and not limitation. It will be appreciated by those skilled in the relevant art(s) that various changes in form and detail may be made therein without departing from the spirit and scope of the embodiments defined in the appended claims. Therefore, the breadth and scope of the present disclosure should not be limited by any of the example embodiments described above, but should be defined only in accordance with the following claims and their equivalents.
Claims
1. A computing system comprising: at least one memory storing program code; as well as a processing system comprising one or more processors, the processing system being configured to receive the program code from the at least one memory, and in response to at least receiving the program code, the processing system being configured to: receiving characteristic information including local and external information, the characteristic information corresponding to a predicted likelihood associated with a state of the processing system; Mapping the textual part and the numerical part of the feature information to corresponding inputs of a neural network; performing word embedding on the mapped text portion to generate embedded text features; providing a plurality of inputs to the neural network, the plurality of inputs comprising at least the embedded text features and a numerical feature based on the numerical portion; as well as The predicted likelihood of the state is determined based at least on an output of the neural network from the plurality of inputs.
2. The computing system of claim 1 , wherein in response to at least receiving the program code, the processing system is configured to: A mitigation action is performed based at least on the predicted likelihood of the state indicating an unfavorable state, the mitigation action comprising one or more of predictive maintenance, load balancing, changing scheduling, upgrading, or increasing resource capacity.
3. The computing system of claim 1 , wherein at least one of the plurality of inputs to the neural network further comprises an external machine learning model.
4. The computing system of claim 1, wherein the embedded text has a reduced input dimensionality and has a mapping range greater than the text portion via an embedding size.
5. The computing system of claim 1 , wherein to map the textual portion and the numerical portion of the characteristic information to the input of the neural network comprises mapping via an abstraction layer, wherein without the mapping, the textual portion and the numerical portion are incompatible inputs to the neural network. 6 . The computing system of claim 5 , wherein the abstraction layer is configured to map available features of the text portion and the numerical portion to the embedded text features and the numerical features, respectively.
7. The computing system of claim 1 , wherein in response to at least receiving the program code, the processing system is configured to: determining, via an interpretability layer, respectively, incremental contributions of neural network variables to the predicted likelihood of the state; and Each incremental contribution is provided to at least one item in the user interface or the forecast log file.
8. A method performed by a computing system, the method comprising: receiving characteristic information including local and external information, the characteristic information corresponding to a predicted likelihood associated with a state of the processing system; Mapping the text part and the numerical part of the external feature information to corresponding inputs of a neural network; performing word embedding on the text portion to generate embedded text features; providing a plurality of inputs to the neural network, the plurality of inputs comprising at least the embedded text features and a numerical feature based on the numerical portion; as well as The predicted likelihood of the state is determined based at least on an output of the neural network from the plurality of inputs.
9. The method according to claim 8, further comprising: A mitigation action is performed based at least on the predicted likelihood of the state indicating an unfavorable state, the mitigation action comprising one or more of predictive maintenance, load balancing, changing scheduling, upgrading, or increasing resource capacity.
10. The method of claim 8, wherein at least one of the plurality of inputs to the neural network further comprises an external machine learning model.
11. The method of claim 8, wherein the embedded text has a reduced input dimensionality and has a mapping range larger than the text portion via an embedding size.
12. The method of claim 8, wherein mapping the textual portion and the numerical portion of the feature information to the input of the neural network comprises mapping via an abstraction layer, wherein without the mapping, the textual portion and the numerical portion are incompatible inputs to the neural network.
13. The method of claim 12, wherein the abstraction layer is configured to map available features of the text portion and the numerical portion to the embedded text features and the numerical features, respectively.
14. The method according to claim 8, further comprising: determining, via an interpretability layer, incremental contributions of neural network variables to the predicted likelihood of the state, respectively; as well as Each incremental contribution is provided to at least one item in the user interface or the forecast log file.
15. A computer-readable storage medium having program instructions recorded thereon, the program instructions, when executed by at least one processor of a computing system, performing the method according to any one of claims 8 to 14.
