Abstract: Embodiments for automatically optimizing and persisting database views by receiving queries made to a database, wherein each query generates a respective database view, and generating a set of database maintained views generated by the queries. The system obtains, for each generated view, certain telemetry information about a respective view including latency, memory space utilization, and processor utilization, among other factors. It then scores each view of the generated views based on an base score modified by the obtained information to determine which one or more of the generated views to make persistent, and maintains the one or more persistent views to produce an optimized persistent set of database views. It further adapts later queries to use the optimized persistent views.

Abstract: Methods and systems for managing storage of data are provided. To manage storage of data, images may be stored along with data usable to facilitate subsequent interpretation of the images. The data may provide information to a subsequent interpreter regarding how a previous outcome was made using the image. The data may allow the subsequent interpreter to understand how the previous interpreter viewed portions of the image, the order in which the portion of the images were viewed by the previous interpreter, etc. The subsequent interpreter may be provided with context regarding landmarks or other features of the image added by the previous interpreter and identified by the previous interpreter as being relevant to the outcome.

Abstract: Methods and systems for identifying areas of interest in an image are disclosed. To manage identification of areas of interest in an image, a combination of automated and subject matter expert driven processes may be used to identify the areas of interest. Some areas of interest may be highly relevant for a user of an image and may be classified as primary areas of interest. The image and subsequent use of the image may be enhanced through the areas of interest by improving storage and downstream usability of the image.

Abstract: Techniques are provided for evaluating one or more anomaly detection models using aggregated time-series signals.

Abstract: A system can train a neural network model at a first edge device regarding respective amounts of time to process data at the first edge device compared to corresponding amounts of time to process the data at cloud computing equipment that is connected to the first edge device via a communications network, wherein the data is generated at the first edge device. The system can update the neural network model to produce an updated neural network model based on information received from a second edge device regarding a performance of the cloud computing equipment in processing the data, wherein the first edge device and the second edge device having respective different processing capabilities. The system can determine whether to process first data, generated at the first edge device, locally at the first edge device.

Abstract: A search engine responding to a user query to find relevant data assets in a federation business data lake (FBDL) system by monitoring and recording interactions of known users interacting with data assets in the FBDL system. Predicted data usage for unknown or new users is derived by training a generative model that uses reconstructive self-supervised learning (SSL) techniques to generate possible values for missing data usage features of the unknown users. The predicted usage is then used to generate similarity scores that are combined for those of the known users to help inform the search engine processing to return relevant results to a target user.

Abstract: A search engine responding to a user query to find relevant data assets in a federation business data lake (FBDL) system based on interactions of known users interacting with data assets in the FBDL system. Data assets are optimally placed for minimal latency or maximal load. Data asset recommendations and past data asset access information are input as features to a time-series model for predicting future data access patterns. An expected latency and load risk is then determined and scored by a weighted mean of these values, and placement optimization is simulated using an optimization method (e.g., genetic algorithm). Using the scoring and simulation, a data asset placement engine is then used to move the locations of the data assets to minimize latency and/or to minimize maximal load.

Abstract: A search engine responding to a user query to find relevant data assets in a federation business data lake (FBDL) system based on interactions of known users interacting with data assets in the FBDL system. Data assets are optimally placed for minimal latency or maximal load. Data asset recommendations and past data asset access information are input as features to a time-series model for predicting future data access patterns. An expected latency and load risk is then determined and scored by a weighted mean of these values, and placement optimization is simulated using an optimization method (e.g., genetic algorithm). Using the scoring and simulation, a data asset placement engine is then used to move the locations of the data assets to minimize maximal load that comprises a load risk representing how close a current load is to a service level agreement (SLA) requirement set by a system provider.

Abstract: A search engine responding to a user query to find relevant data assets in a federation business data lake (FBDL) system. The search engine receives a search query from an unprivileged user or a user not having sufficient privileges to access the FBDL. It returns initial results to the unprivileged user including a first data asset recommendation responsive to the search query. It then determines a causal reason that the first data asset was recommended, and uses a similarity engine conditioned on the causal reason to return a replacement data asset in response to the search query.

Abstract: A search engine responding to a user query to find relevant data assets in a federation business data lake (FBDL) system based on interactions of known users interacting with data assets in the FBDL system. Predicted data usage for a population of possible users is derived by training a generative model that uses reconstructive self-supervised learning (SSL) techniques to generate possible values for missing data usage features of the possible users. The predicted usage is then used to generate similarity scores that are combined with those of the known users to return relevant results to a target user. The predicted usage for possible users is processed in a consensus mechanism to produce integrated recommendations that are iteratively optimized using a chosen usage by the target user an a optimizer using a genetic algorithm to change consensus parameters to derive a final recommendation to the target user.

Abstract: Techniques are provided for automatically detecting disruptive orders for a supply chain. One method comprises obtaining a quote for an order; extracting features from the quote; and applying the extracted features to a disruptive quote machine learning engine that generates an anomaly score indicating a likelihood that the quote will cause a disruption, based on one or more predefined disruption criteria. The disruptive quote machine learning engine may employ an isolation forest algorithm and/or a multi-dimensional anomaly detection algorithm. The disruptive quote machine learning engine may be trained using historical order information comprising part-level information from historical orders and/or a manufacturing production plan comprising an inventory forecast.

Abstract: Techniques are provided for decompression of compressed log data, such as for a real-time viewing of compressed log data. An exemplary method comprises: obtaining a compressed log file comprised of a plurality of compressed log messages, wherein a given compressed log message is comprised of one or more message variables and a message signature corresponding to a message template of the given compressed log message; and presenting a first subset of the compressed log file by translating, in memory, the message signatures of the compressed log messages within the first subset to corresponding message templates using a decompression index that maps a plurality of the message signatures to corresponding message templates. The first subset of the compressed log file may comprise a predefined number of lines surrounding a requested line of the compressed log file. In further variations, at least one additional subset of the compressed log file is precomputed using the disclosed decompression techniques.

Abstract: Techniques are provided for correcting sensor data in a multi-sensor environment. An exemplary method comprises obtaining sensor data from a first sensor; applying an anomaly detection technique to detect an anomaly in the sensor data from the first sensor based on additional sensor data from one or more of the first sensor and at least one additional sensor in proximity to the first sensor; and correcting the anomalous sensor data from the first sensor using additional sensor data from one or more of the first sensor and the at least one additional sensor. In some embodiments, additional sensor data from a plurality of neighboring sensors is used to predict the sensor data from the first sensor. The anomalous sensor data is optionally corrected substantially close in time to the detection of the anomaly in the sensor data.

Abstract: Techniques are provided for compression and decompression of log data. An exemplary method comprises: obtaining a log message, wherein the log message comprises a message template and one or more message variables; obtaining a compression index that maps a plurality of message templates to a corresponding message signature; and writing the one or more message variables and a message signature corresponding to the message template of the log message to a log file. A counter may be maintained for each of a plurality of distinct message templates, and a given message signature may be assigned to a particular message template based on a length of the given message signature and a frequency of occurrence of the particular message template. The compression index comprises, for example, a key/value database where the message templates are keys and the corresponding message signatures are values of the key/value database. A decompression index maps message signatures to corresponding message templates.

Abstract: Techniques are provided for automatically detecting disruptive orders for a supply chain. One method comprises obtaining a quote for an order; extracting features from the quote; and applying the extracted features to a disruptive quote machine learning engine that generates an anomaly score indicating a likelihood that the quote will cause a disruption, based on one or more predefined disruption criteria. The disruptive quote machine learning engine may employ an isolation forest algorithm and/or a multi-dimensional anomaly detection algorithm. The disruptive quote machine learning engine may be trained using historical order information comprising part-level information from historical orders and/or a manufacturing production plan comprising an inventory forecast.

Abstract: Techniques are provided for software license optimization using machine learning-based user clustering. One method comprises obtaining key performance indicators indicating individual usage by a plurality of users of a software product; applying at least one function to the key performance indicators to obtain a plurality of time dependent features; processing the time dependent features using a machine learning model to cluster the users into a plurality of persona clusters; and determining a number of each available license type for the software product for the plurality of users based on the persona clusters. The key performance indicators comprise, for example, user behavioral data with respect to usage of the software product and/or performance data with respect to usage of the software product. One or more policies can be determined for managing an allocation of the available license types for the software product to the plurality of users.

Abstract: Methods and systems for managing the utilization of cloud computing resources are described. The system monitors cloud computing resource utilization for a first set of active jobs to determine real-time utilization data. The system compares the real-time utilization data with historic utilization data to generate a utilization pattern and determines future cloud computing resource utilization for at least one future time period based on the utilization pattern and a second set of scheduled jobs. The system further generates a pricing matrix for utilizing the cloud computing resources during a future time period based on the determined future cloud computing resource utilization. The pricing matrix includes prices associated with utilization of the cloud computing resources for each of the at least one future time period. The system transmits the pricing matrix to one or more devices requesting utilization of the cloud computing resources.

Abstract: Flexible scheduling for backup jobs includes backup policies that allow for flexibility in scheduling execution time, and an automated data driven backup job execution scheduler. The scheduler balances load on the customer's and data protection system's resources using a dynamic pricing calculator that takes into account the real-time and predicted near-future status of the protection environment. The backup job scheduler includes a user interface that enables users to specify user-customized backup policies that enable flexibility in scheduling execution times and transparency in pricing.

Abstract: Techniques are provided for sensor data compression in a multi-sensor Internet of Things environment. An exemplary method comprises obtaining sensor data from a plurality of sensors satisfying one or more of predefined sensor proximity criteria and predefined similar sensor type criteria; applying an image-based compression technique to the sensor data to generate compressed sensor data; and providing the compressed sensor data to a data center. The image-based compression technique comprises a discrete cosine transform technique, a video compression technique, and/or an auto-encoder deep learning technique that utilizes one or more over-fitted bidirectional recurrent convolutional neural networks. The sensor data is optionally normalized prior to being applied to the image-based compression technique.

Abstract: Techniques are provided for software license optimization using machine learning-based user clustering. One method comprises obtaining key performance indicators indicating individual usage by a plurality of users of a software product; applying at least one function to the key performance indicators to obtain a plurality of time dependent features; processing the time dependent features using a machine learning model to cluster the users into a plurality of persona clusters; and determining a number of each available license type for the software product for the plurality of users based on the persona clusters. The key performance indicators comprise, for example, user behavioral data with respect to usage of the software product and/or performance data with respect to usage of the software product. One or more policies can be determined for managing an allocation of the available license types for the software product to the plurality of users.