Abstract: An immersion cooling system includes a catch pan, a heat-generating electronic device, a housing, and a fluid pump. The housing is positioned around the heat-generating electronic device, and at least part of the housing is positioned above the catch pan. The fluid pump is configured to circulate a working fluid from the catch pan to the housing.

Abstract: The technology relates to systems and methods for generating advanced feedback for a draft message. The operations may include receive text for a message being drafted in a messaging application; upon an analysis condition being satisfied, analyze the message by applying at least one of a message-analysis model or heuristic to generate a feedback score for the message; and based on the feedback score crossing a feedback threshold, trigger generation of advanced feedback for the message. The operations may also or alternatively include receive an initial sent message from a messaging application; analyze the message by applying at least one of a message-analysis model or heuristic to generate a feedback score for the message; based on the feedback score crossing a feedback threshold, transmit a feedback alert message for surfacing in the messaging application; and based on receiving an interaction, trigger generation of advanced feedback for the message.

Abstract: Systems and methods for determining predicted olfactory perception are provided. In particular, a method comprises receiving an input indicating an odorant, generating an odorant vector representing the odorant, generating an olfactory receptor vector, and determining one or more predicted olfactory percepts associated with the odorant based on the odorant vector and the olfactory receptor vector.

Abstract: Systems and methods for processing a stream of input images are provided. An example method includes receiving a stream of input images and a pointing angle associated with the stream of input images, wherein each input image in the stream of input images comprises a plurality of pixels; interpolating an effective analytical projection, for each input image of the stream of input images, from a grid of predetermined analytical projections, based on the respective pointing angle and plurality of pixels of each of the input images of the stream of input images, wherein the grid of predetermined analytical projections comprises a plurality of spaces that each correspond to respective predetermined pointing angles; generating a modified stream of input images, by mapping pixels of the input stream of images to projected pixels of the modified stream of images, using the effective analytical projection; and displaying the modified stream of images.

Abstract: Systems and methods are provided for implementing an ultra-scalable high-performance computing (“HPC”) network using dense wavelength-division multiplexing (“DWDM”). The HPC system includes an interconnection of GPU devices, multiplexer/demultiplexer (“mux/demux”) devices, amplifiers, wavelength selective switches (“WSSs”), and optical circuit switches (“OCSs”). Each OCS includes a plurality of micro-electromechanical systems (“MEMS”) mirrors and a plurality of input/output (“I/O”) ports each communicatively coupled to one WSS mux/demux device one WSS. Each WSS mux/demux device is either communicatively coupled to one of the I/O ports of an OCS or one of a plurality of GPU mux/demux devices via an amplifier. Each GPU mux/demux device is communicatively coupled to a number of GPU devices, each including another number GPUs and one or more optoelectronic devices.

Abstract: The description relates to cameras, and camera calibration for enhancing user experiences. One example can receive a first image of a user at a first location relative to a camera. The first image can include the user's upper body but does not include the user from head to toe. The example can receive a second image of the user at a second location relative to a camera. The second image can include the user's upper body but does not include the user from head to toe. The example can estimate a distance of the second location from the first location relative to the camera and calibrate a height and tilt angle of the camera from the first image, the second image, and the estimated distance and without a full body image of the user.

Abstract: A data processing system for providing a service to extract information from a resource includes: a network interface for communicating over a computer network; a scraper tool to receive user instruction specifying a target resource and to extract content from the specified resource, wherein the user instruction further specifies a desired restructuring of the extracted content; and a prompt generator to structure the extracted content into a prompt for an Artificial Intelligence (AI) model, the prompt further directing the AI model to restructure the extracted content based on the user instruction. The prompt generator is to call the AI model with the generated prompt. The service is to receive restructured content from the AI model and provide the restructured content to a workstation submitting the user instruction, the restructured content presenting the content of the target resource in a form according to the user instruction.

Abstract: Systems and methods for generating custom art fonts with consistent style include receiving user input that identifies a base font style for a custom font and includes descriptive text that defies one or more text effects to use for the custom font. Depth maps are selected for characters to be included in the custom font. The depth maps are preprocessed to add noise to the depth maps. A generative model generates custom font images conditioned with the text prompt and the depth maps. The custom font images are then used to render text on a display screen of a computing device.

Abstract: A confidence estimation tool uses a calibrated confidence mapping model to estimate confidence for a model-generated candidate root cause. The tool uses a generative artificial intelligence (“AI”) model to determine, based on a description of a current event, a candidate root cause of the current event. The tool determines a description-based confidence score using the description of the current event and descriptions of a set of relevant historical events in a target domain. The tool also determines a cause-based confidence score using the candidate root cause of the current event and root causes of the set of relevant historical events. Finally, the tool determines a final confidence score using the description-based and cause-based confidence scores. Even if the generative AI model is configured for general-domain applications, by referencing relevant historical events, the tool can accurately estimate confidence for a model-generated candidate root cause within the target domain.

Abstract: Systems and methods for representing two-dimensional representations as three-dimensional avatars are provided herein. In some examples, one or more input video streams are received. A first subject, within the one or more input video streams, is identified. Based on the one or more input video streams, a first view of the first subject is identified. Based on the one or more input video streams, a second view of the first subject is identified. The first subject is segmented into a plurality of planar object. The plurality of planar objects are transformed with respect to each other. The plurality of planar objects are based on the first and second views of the first subject. The plurality of planar objects are output in an output video stream. The plurality of planar objects provide perspective of the first subject to one or more viewers.

Abstract: A technique for interacting with map-related information integrates the use of a machine-trained language model. Upon submission of a query, the technique uses the machine-trained language model to assess at least one intent associated with the query. The technique then invokes an intent-specific processing flow to provide an output result. Each processing flow invokes the use of at least one processing engine to perform an engine-specific task, such as geocoding, route finding, or image retrieval. A processing flow can also call on the machine-trained language model one or more additional times. In some cases, the technique includes a feedback mechanism for soliciting additional information from a user.

Abstract: A computing system for monitoring language model compliance with a rubric of one or more output characteristics. The computing system includes processing circuitry configured to interface with a trained generative language model that receives input of a prompt including natural language text input and, in response, generates an output that includes natural language text output. The processing circuitry is further configured to monitor compliance of the generative language model with the rubric, by feeding the output of the generative language model to a rubric classifier configured to generate a predicted classification for an output characteristic in the rubric, and output the predicted classification.

Abstract: A multimode optical waveguide network comprises a parent waveguide and a plurality of child waveguides. Each waveguide is a multimode optical waveguide having a first surface region, multiple second surface regions, and at least one guiding element attached to a surface of the waveguide or embedded within the waveguide, each second surface region of the parent waveguide optically coupled to the first surface region of a corresponding child waveguide. The guiding element(s) of the parent waveguide is arranged to guide a beam, from or to its first surface region, to or from any selected second surface region of its multiple second surface regions. The guiding element(s) of each of the waveguides is configurable for selecting the second surface region of that waveguide and/or responsive to at least one beam characteristic for selecting the second surface region of that waveguide via modulation of the at least one beam characteristic.

Abstract: This document relates to training of machine learning models such as neural networks. One example method involves providing a machine learning model having one or more layers and associated parameters and performing a pretraining stage on the parameters of the machine learning model to obtain pretrained parameters. The example method also involves performing a tuning stage on the machine learning model by using labeled training samples to tune the pretrained parameters. The tuning stage can include performing noise adjustment of the labeled training examples to obtain noise-adjusted training samples. The tuning stage can also include adjusting the pretrained parameters based at least on the labeled training examples and the noise-adjusted training examples to obtain adapted parameters. The example method can also include outputting a tuned machine learning model having the adapted parameters.

Abstract: A “Content Optimizer” applies a machine-learned relevancy model to predict levels of interest for segments of arbitrary content. Arbitrary content includes, but is not limited to, any combination of documents including text, charts, images, speech, etc. Various automated reports and suggestions for “reformatting” segments to modify the predicted levels of interest may then be presented. Similarly, the Content Optimizer applies a machine-learned comprehension model to predict what a human audience is likely to understand (e.g., a “comprehension prediction”) from the arbitrary content. Various automated reports and suggestions for “reformatting” segments to modify the comprehension prediction may then be presented.

Abstract: Systems and methods are provided for implementing confidence enhancement for responses by document-based large language models (“LLMs”) or other AI/ML systems. A first prompt is generated based on data items that are previously received or accessed. The first prompt is used by a first LLM or AI/ML system to extract requested information from the data items. One or more citations are generated and presented within a structured object together with a representation of the extracted information, in some cases, as output from a second LLM or AI/ML system. In some cases, the citations and/or the representation may be verified by a third LLM or AI/ML system, and reliability indicators may be generated for the citations and/or the representation based on determined accuracy of the citations and/or the representation. In this manner, the common issue of hallucinations may be mitigated.

Abstract: One example provides a computing device comprising a display, a touch sensor operatively coupled to the display, a non-touch sensor configured to provide an output indicative of user engagement between the computing device and a user of the computing device, a logic machine, and a storage machine. The storage machine comprises instructions executable by the logic machine to determine a touch usage probability based at least upon the output from the non-touch sensor. The instructions are further executable to change the operation of the touch sensor between an idle mode and a scanning mode based at least on the touch usage probability meeting a probability threshold condition.

Abstract: Aspects of the present disclosure relate to engagement-based communication session management. In examples, an interaction intent metric is generated for a user according to a variety of factors relating to the user (e.g., the semantic content and tone of the user's speech, a direction of the user's gaze, and historical user characteristics) and the meeting (e.g., whether the user's name was mentioned by another communication participant or whether another communication participant is soliciting input), among other examples. Accordingly, if a positive interaction intent is identified and the user is currently muted, an action can be recommended to address the mismatch between the positive interaction intent and the muted status of the user. Similarly, if a negative interaction intent is identified and the user is currently unmuted, an action can be performed to address the mismatch between the negative interaction intent and the unmuted status of the user.

Abstract: A method and system for intelligently organizing one or more groups of relevant files may include retrieving a user data signal including user-specific data, analyzing the user data signal to identify a parameter relating to file relevance, identifying one or more relevant files in a storage medium based on the parameter at least one of a user category property, a lifecycle stage property, a relevant activity property, or an activity level property of one or more files in the storage medium, organizing the one or more relevant file into the one or more groups of relevant files, and providing for display data relating to the one or more groups of relevant files.

Abstract: The description relates to executing an inference query relative to a database management system, such as a relational database management system. In one example a trained machine learning model can be stored within the database management system. An inference query can be received that applies the trained machine learning model on data local to the database management system. Analysis can be performed on the inference query and the trained machine learning model to generate a unified intermediate representation of the inference query and the trained model. Cross optimization can be performed on the unified intermediate representation. Based upon the cross-optimization, a first portion of the unified intermediate representation to be executed by a database engine of the database management system can be determined, and, a second portion of the unified intermediate representation to be executed by a machine learning runtime can be determined.

Abstract: A method for backfilling records dropped by a distributed node system during a transfer to an external database includes initially determining whether an attempt to transfer a data record failed or succeeded. In response to determining that the transfer succeeded, the data record is appended at its original location include a first transfer status identifier. In response to determining that the transfer failed, the data record is appended to include a second transfer status identifier and queued copy of the data record is created by copying the data record to a delayed re-try queue. The method further includes verifying that the transfer status identifier of the queued copy is identical to the transfer status identifier of a newest version of the data record within the distributed node system and, responsive to the verification, attempting a subsequent transfer of the queued copy of the data record to the external database.

Abstract: Aspects of the disclosure include methods and systems for providing a theme based font search in a manner that bypasses the latency limitations inherent to large language models and ensures domain-relevant answers. An exemplary method can include inputting a font category to a compact model to generate a font tag embedding and inputting a search query to the compact model to generate a query embedding. A distance measure is determined between the query embedding and the font tag embedding. The font category for the font tag embedding is provided as a suggested font category responsive to the distance measure satisfying a predetermined threshold.

Abstract: A pre-trained neural code generation model generates repair code for a method containing a performance bug given a prompt including a code transformation instruction. The code transformation instruction guides the model on how to predict the repair code when the model has not been fine-tuned for the repair code task. The code transformation instruction is retrieved from abstract bug patterns derived from historical performance bug fixes found in commits to a source code repository. The augmentation of the code transformation instruction in the prompt to the pre-trained neural code generation model provides the model with a hint on how the repair code may be generated based on similar performance bug fixes.

Abstract: A computer implemented method processes audio streams recorded during a meeting by a plurality of distributed devices.

Abstract: Some embodiments automatically and proactively adjust network device configuration settings during network operation, based on correlations between device performance and device configuration. Correlations are computed using statistics routines or computed by a machine learning module. Some embodiments share adjusted configuration values via a cache, and some persist adjusted values through an application restart. In some embodiments, the cache is hierarchical and different kinds of reconfiguration data are shared at different levels. In some embodiments, the configuration value is shared only between application instances that have sufficiently similar contexts. Some embodiments detect a correlation loss and fall back to a known good configuration setting or a default configuration setting. Some embodiments optimize network internode communications by making dynamic adjustments which are not available from static configuration settings or from static configuration rules.

Abstract: Systems and methods for opening files on a client device include receiving a request to open a file and identifying a file type pertaining to the file. A list of software applications is identified that are capable of opening the identified file type, and metadata pertaining to each of the software applications on the list is retrieved. User preference data pertaining to a user requesting to open the file is also retrieved. Scores for the software applications on the list are then determined based at least on the metadata and the user preference data. At least one software application on the list is then selected to open the file on the client device based on the determined scores.

Abstract: Information is provided to a service mesh by a Kubernetes (K8s) controller. The information enables the service mesh to determine an IP address of application pods to which a monitoring request is destined. The K8s controller detects which of the application pods are scheduled to be monitored by the monitoring service. The K8s controller creates dummy headless services that match the application pods scheduled to be monitored. The service mesh is programmed information from the dummy headless services to cause the service mesh to intercept the monitoring request and identify that the monitoring request is destined to the dummy headless service based on the addresses. In response the service mesh encrypts the monitoring request.

Abstract: Techniques are disclosed for providing method for providing an event timer for event synchronization across Kubernetes clusters. The event timer is configured to provide event synchronization on behalf of microservice instances in the cloud computing environment. In response to a request for an event timer for a timed event, it is determined whether the requested event timer has been started for a second microservice instance. If the requested event timer has been started, a state of the requested event timer is sent to the first microservice instance If the requested event timer has not been started, the requested event timer is instantiated, and a state of the instantiated event timer is stored in a database. The instantiated event timer is independent of the first and second microservice instances. In response to an expiration of the event timer, a single callback for processing of the event is generated.

Abstract: A projection system includes an illumination light source configured to emit an illumination light beam, a monitor light source configured to emit a monitor light beam, and a projector configured to project both the illumination light beam and the monitor light beam into a projected combined light beam. A first portion of the projected combined light beam is propagated over a first beam path in a first direction, causing an eye of a user to see a display image. A second portion of the projected combined light beam is propagated over a second beam path in a second direction, causing a monitor camera to capture a monitor image. The monitor image is analyzed to determine an orientation or a position of the monitor image. In response to determining that the monitor image is not properly oriented or positioned, an orientation or position of the projector or the illumination image is adjusted.

Abstract: In an example embodiment, tensor train decompositions are used to create large, personalized layers that are efficient for segment modeling. More particularly, rather than performing learning on an input matrix of training data that contains all segments, and then crossing this matrix with a vector for a particular segment, the matrix is mapped to an N-dimensional tensor, where each of the dimensions corresponds to one of the properties used to compose the segment, which can then be approximated by tensor train decomposition to enable efficient training and scoring.

Abstract: Filesystem driver software can receive a file access request indicating that an application process is requesting to access a target file in a filesystem, Network filter driver software can receive a connection establishment request indicating that the application process running on the processing apparatus is requesting to establish a connection over a network with a target endpoint. According to the present disclosure, one or both of: a) the filesystem driver software is configured to grant or deny the file access request in dependence on state information from the network filter driver software, and/or b) the network filter driver software is configured to grant or deny the connection establishment request in dependence on state information from the filesystem driver software.

Abstract: A speech module is joint trained with a knowledge module by transforming a first knowledge graph into an acoustic knowledge graph. The knowledge module is trained on the acoustic knowledge graph. Then, the knowledge module is integrated with the speech module to generate an integrated knowledge-speech module. In some instances, the speech module included in the integrated knowledge-speech module is aligned with a language module to generate an optimized speech model configured to leverage acoustic information and acoustic-based knowledge information, along with language information.

Abstract: Eye and hand tracking systems in head-mounted display (HMD) devices are arranged with lensless camera systems using optical masks as encoding elements that apply convolutions to optical images of body parts (e.g., eyes or hands) of HMD device users. The convolved body images are scrambled or coded representations that are captured by a sensor in the system, but are not human-recognizable. A machine learning system such as a neural network is configured to extract body features directly from the coded representation without performance of deconvolutions conventionally utilized to reconstruct the original body images in human-recognizable form. The extracted body features are utilized by the respective eye or hand tracking systems to output relevant tracking data for the user's eyes or hands which may be utilized by the HMD device to support various applications and user experiences. The lensless camera and machine learning system are jointly optimizable on an end-to-end basis.

Abstract: A method and system for identifying enhancement suggestions for a document and displaying the identified suggestions in a user-friendly manner may include examining the document to identify an enhancement fora portion of content in the document, enabling display of a first user interface element for previewing the identified enhancement as applied to the content, receiving a request to select the identified enhancement, and upon receiving the request enabling display of a second user interface element for accepting the identified enhancement. The first user interface element may provide a preview of the document post enhancement.

Abstract: A method for authorizing access to one or more secured computer resources includes obfuscating a reference biometric vector into an obfuscated reference biometric vector using a similarity-preserving obfuscation. An authentication biometric vector is obfuscated into an obfuscated authentication biometric vector using the similarity-preserving obfuscation. A similarity of the obfuscated authentication biometric vector and the obfuscated reference biometric vector is tested. Based on the similarity being within an authentication threshold, access to the one or more secured computer resources is authorized.

Abstract: A centralized skills management server, a computer-readable storage medium, and a computer-implemented method for skills inference are described herein. The method includes executing a web-based application on a remote computing system operated by a user associated with a tenant and extracting skills-related terms associated with the execution of the web-based application. The method includes interfacing with the global skills graph via an API and importing standardized skill tags relating to the extracted skills-related terms.

Abstract: Lightweight mechanisms provide a way to assert provenance when live streaming media content and establish provenance upon playback. For example, a provenance claim generator generates a key pair including a live-stream private key and live-stream public key. The claim generator signs, with a long-term private key reliably associated with a sender, manifest metadata including the live-stream public key, thereby producing a manifest signature. During live streaming, the claim generator signs respective portions of media content with the live-stream private key, producing portion signatures for the respective portions. A provenance claim validator receives the manifest signature and manifest metadata. The claim validator verifies the manifest metadata using a long-term public key (reliably associated with the sender) and the manifest signature.

Abstract: Methods, systems, and computer storage media for providing predicted task-actions for tasks from a plurality of task applications associated with a task management system. In operation, a task associated with a task application is accessed at a task-action prediction engine. An intent of the task is determined. The intent of the task is determined based on task features and an intent prediction machine learning model that is trained to predict intents associated with natural language text. Based on the intent of the task, a predicted task-action is selected based on a task-action computation model that is associated with predefined task-actions and predefined intent-task-categories. The predicted task-action can be communicated to a task management client that causes display of the predicted task-action in combination with a predicted task-action interface element associated with executing the predicted task-action.

Abstract: A method, computer program product, and computing system for processing event data associated with a plurality of known operational impact events on a business service and operational data associated with the business service using a supervised machine learning model conditioned on an operational impact parameter associated with the business service. A detection threshold is generated using the supervised machine learning model.

Abstract: A data processing system implements a method for training machine learning modes, including receiving a set of one or more unlabeled documents associated one or more first categories of documents to be used to train machine learning models to analyze the one or more unlabeled documents, and fine-tuning a first machine learning model and a second machine learning model based on the one or more unlabeled document to enable the first machine learning model to determine a semantic representation of the one or more first categories of document, and to enable the second machine learning model to classify the semantic representations according to the one or more first categories of documents, the first machine learning model and the second machine learning model having been trained using first unlabeled training data including a second plurality of categories of documents that do not include the one or more first categories of documents.

Abstract: A connector assembly including a first intermediate connector member associated with a first device and having a first connector interface and a second intermediate connector having a second connector interface. The first intermediate connector and the second intermediate connector are movable between an interfaced state in which the first connector interface is in contact with the second connector, and a tethered state in which the first connector interface is spaced apart from the second connector interface. The connector assembly further includes a tether member connecting the first intermediate connector and the second intermediate connector in the tethered state. The tether may include a flexible body configured to control a path along at least an axis of either the first intermediate connector or the second intermediate connector during movement from the interfaced state to the tethered state.

Abstract: Systems and methods are provided for determining faces and bodies of people in an image by adaptively scaling images and by iteratively using a deep neural network for inferencing. A camera captures an image including faces and bodies of people. A face/body determiner determines faces and bodies of people appearing in the image by resizing the image into a predetermined pixel dimension as input to the deep neural network. A region cropper determines a crop region associated with a low level of confidence in detecting faces and bodies that are too small to determine with an acceptable level of confidence. The region cropper resizes the crop region into the predetermined pixel dimension as input to the deep neural network. The face and body determiner determines other faces and bodies appearing in the resized crop region. An aggregator aggregates locations of the determined faces and bodies in the image.

Abstract: A system for adding persistence to SPAD imagery is configurable to capture, using a SPAD array, a plurality of image frames. The system is configurable to capture, using an IMU, pose data associated with the plurality of image frames. The pose data includes at least respective pose data associated with each of the plurality of image frames. The system is configurable to determine a persistence term based on the pose data. The system is also configurable to generate a composite image based on the plurality of image frames, the respective pose data associated with each of the plurality of image frames, and the persistence term. The persistence term defines a contribution of each of the plurality of image frames to the composite image.

Abstract: An ontology matching system performs operations to refine a natural language processing (NLP) model that encodes terms of a first hierarchical ontology and of a second hierarchical ontology as embeddings in a latent space. The operations include performing at least a first round of triplet loss training to decrease separation between select pairs of the embeddings sampled from the different ontologies that satisfy a first hierarchical relation while increasing separation between other pairs of the embeddings that do not satisfy the first hierarchical relation. The system then determines, from the refined NLP model, a stable matching scheme that matches each term in the first hierarchical ontology with a corresponding term of the second hierarchical ontology. Responsive to receiving terms of the first hierarchical ontology from an application, the system uses the stable matching scheme to map each of the terms to corresponding terms of the second hierarchical ontology.

Abstract: A method, computer program product, and computing system for obtaining one or more speech signals from a first device, thus defining one or more first device speech signals. One or more speech signals may be obtained from a second device, thus defining one or more second device speech signals. A noise component model may be selected from a plurality of noise component models based upon, at least in part, the one or more first device speech signals and the one or more second device speech signals. The one or more second device speech signals may be augmented, at run-time, based upon, at least in part, the noise component model.

Abstract: Innovations in encoder-side options for intra block copy (“BC”) prediction mode facilitate intra BC prediction that is more effective in terms of rate-distortion performance and/or computational efficiency of encoding. For example, some of the innovations relate to concurrently performing block vector (“BV”) estimation and making block splitting decisions for a block. Other innovations relate to selectively merging blocks into a larger block during BV estimation.

Abstract: A distributed computing system is provided, including a source computing device implementing a source hypervisor executing a virtual machine, a source-side router coupled to the source hypervisor, a destination computing device implementing a destination hypervisor, and a destination-side router coupled to the destination hypervisor. The source hypervisor is connected to the destination hypervisor through a network. During a migration of the virtual machine from the source hypervisor to the destination hypervisor, the source-side router duplicates packets ingressing into the source-side router. The duplicated packets include a first set of packets and a second set of packets, where the first set of packets is routed to the virtual machine through the source hypervisor on the network, and the second set of packets is routed to the virtual machine through the destination hypervisor on the network.

Abstract: Methods and apparatuses for performing context completion to messages in a session are provided in the present disclosure. A message may be obtained. It may be detected that there exists context ellipsis in the message. It may be determined whether the message is retained in the current domain of the session. In response to determining that the message is retained in the current domain, a complementary text for recovering the context ellipsis may be selected in the current domain. A completed message may be generated based on the message and the complementary text.

Abstract: A system to detect an abnormally permissive role definition, which can include an abnormally permissive custom role definition, and take action is described. The system receives a role definition for a security principal over a scope of resources in which the role definition includes a built-in role and a custom role. Permissions of the role definition and a creation event of the role definition are analyzed. A security score based on the role definition and creation event for the scope of resources is determined. An action is taken based on the security score and the creation event analysis.

Abstract: Described herein is a software update distribution service that leverages reinforcement learning—a specific type machine learning algorithm—to discover or learn optimal times (e.g., a schedule) to download software updates and to install software updates for software applications installed on a group of client computing devices of a specific enterprise, in order to achieve one of several predefined goals or objectives selected for the specific enterprise, or for the specific group of client computing devices. Using reinforcement learning, a software-based agent learns to perform activities relating to software updates at specific times that minimize a penalty, wherein the penalty is derived based on a weighted combination of metrics, some of which relate to the impact of software update timing on network traffic and power consumption.