Abstract: Techniques for providing machine-learned (ML)-based artificial intelligence (AI) capabilities are described. In one technique, multiple AI capabilities are stored in a cloud environment. While the AI capabilities are stored, a request for a particular AI capability is received from a computing device of a user. Also, in response to receiving training data based on input from the user, the training data is stored in a tenancy, associated with the user, in the cloud environment. In response to receiving the request, the particular AI capability is accessed, a ML model is trained based on the particular AI capability and the training data to produce a trained ML model, and an endpoint, in the cloud environment, is generated that is associated with the trained ML model. The endpoint is provided to the tenancy associated with the user.

Abstract: The system for optimizing multi-domain machine learning models determines layer-specific precision levels for each neural network layer of a trained machine learning model based on activation values during forward propagation with validation data. The precision levels are assigned according to a predetermined priority order to create a quantized machine learning model. The system trains a meta-learning model to predict software library configurations using the trained machine learning model. The meta-learning model identifies required software libraries based on hardware specifications and specific tasks for the target device. The system dynamically links the software libraries with the quantized machine learning model for deployment on the target device, enabling autonomous operation.

Abstract: A method of analyzing the effectiveness of a machine learning model may include executing a test procedure of a machine learning model comprising a plurality of input features. The machine learning model may include a lag feature associated with one of the plurality of input features. Executing the test procedure may include querying one or more data sources for historical data records associated with the input features. The lag feature may restrict a query for historical data records from at least one of the data sources to a predetermined time prior to a time of the test procedure of the machine learning model. Executing the test procedure may include receiving the historical data records from the one or more data sources at the machine learning model. Executing the test procedure may include generating an output of the machine learning model.

Abstract: In some embodiments, reducing utilization of computational resources associated with segmenting datasets via a cluster-ensemble model may be facilitated. In some embodiments, the system may receive a raw dataset having a first dimension. The system may then embed the raw dataset into an embedded dataset having a second dimension, where the embedded dataset comprises a vector embedding. The system may then provide the embedded dataset to a set of clustering models to generate a set of clusters, where each cluster of the set of clusters corresponds to a respective clustering model of the set of clustering models. The system may provide the set of clusters to a cluster-ensemble model to generate a set of ensemble-clusters. Based on the set of ensemble-clusters, the system may generate a set of data segments corresponding to the set of ensemble-clusters indicating at least one characteristic of a respective ensemble-cluster of the set of ensemble-clusters.

Abstract: Systems and methods for machine learning architecture for out-of-distribution data detection. The system may include a processor and a memory storing processor-executable instructions that may, when executed, configure the processor to: receive an input data set; generate an out-of-distribution prediction based on the input data set and an auto-encoder, the auto-encoder trained based on a pretext task including a transformation of one or more training data sets for reconstruction, the trained auto-encoder trained for reducing a reconstruction error to encode semantic meaning of the training data sets; and generate a signal for providing an indication of whether the input data set is an out-of-distribution data set.

Abstract: For each corresponding configuration item type of a plurality of different configuration item types, a corresponding multi-variate machine learning model of a plurality of multi-variate machine learning models is trained to perform anomaly detection for a corresponding configuration item type of the plurality of different configuration item types. In response to detecting, via a univariate machine learning model, an anomaly associated with a specific configuration item type of the plurality of different configuration item types, an execution of a particular multi-variate machine learning model of the plurality of multi-variate machine learning models is initiated for the specific configuration item type. An output of the execution of the particular multi-variate machine learning model is evaluated to determine an anomaly detection result.

Abstract: A method is disclosed for managing transformed datasets in a compute cluster environment. The method includes identifying, based on one or more machine learning models to be executed on a compute cluster comprising a plurality of GPU servers, one or more transformations to apply to a dataset. The method further includes generating a transformed dataset based on the one or more transformations, storing the transformed dataset, receiving a request to transmit the transformed dataset to at least one GPU server of the plurality of GPU servers, and, responsive to the request, transmitting the stored transformed dataset to the at least one GPU server without re-performing the one or more transformations on the dataset.

Abstract: Methods and systems for selecting a target model for an unlabeled dataset of a dataset provider, the target model for generating labels for the unlabeled dataset. The method comprises acquiring the unlabeled dataset from the dataset provider; acquiring a first candidate model from a first model provider and a second candidate model from a second model provider, generating a first usefulness score for the first candidate model and a second usefulness score for the second candidate model using the unlabeled dataset, the first and second usefulness scores being indicative of likelihood that the first and second candidate models generate accurate labels for the unlabeled dataset respectively; selecting the first candidate model as the target model using the first usefulness score and the second usefulness score; and causing generation of the labels from the unlabeled dataset using the target model.

Abstract: A multi-stage selectable neural network noise suppression system has a first stage noise pattern selection neural network receptive to an input signal. An automatic noise classification is generated based upon an evaluation of the input signal. A noise suppression weight table stores one or more sets of automatic noise suppression weight values corresponding to the generated automatic noise classifications. A second stage noise pattern suppression neural network then selectively applies a specific automatic targeted noise suppression based upon the automatic noise suppression weight values. This approach balances performance quality and power/memory usage that efficiently tailors noise detection and suppression.

Abstract: A system may receive a request to generate an object. A system may generate, based on the request, a first portion of a prompt associated with a first layer of a hierarchy and a second portion of the prompt associated with a second layer of the hierarchy, where the second layer has a lower priority than the first layer. A system may generate the object based in part on providing the portions of the prompt as input to a machine learning model, where the object is formatted according to an object format and is internally consistent. To generate the object the machine learning model does not violate instructions associated with a layer of the hierarchy based on instructions associated with a layer of the hierarchy having a relatively lower priority.

Abstract: Methods, systems and computer program products are described herein that provide a serverless, multi-engine, multi-user data lake indexing subsystem and application programming interface. Indexes are defined as derived datasets and stored on the data lake in a universal format that enables disparate engines to create and/or discover indexes for workload optimization. Embodiment of indexes enable stateful control and management of an index via metadata included in the index and stored on the data lake.

Abstract: Provided is a recommendation method. The method includes: determining a target user identifier and a plurality of candidate item identifiers; acquiring the intermediate user feature vector corresponding to the target user identifier by searching for the target user identifier; acquiring the intermediate item feature vectors corresponding to the plurality of candidate item identifiers by searching for the plurality of candidate item identifiers; outputting degrees of matching of the plurality of candidate item identifiers and the target user identifier by predicting the intermediate user feature vector and the intermediate item feature vectors in the online prediction model; and determining a target item identifier to be recommended to the target user identifier from the plurality of candidate item identifiers.

Abstract: A method of performing a semantic textual similarity search between a target document and a set of source documents includes: selecting target textual data from the target document; identifying and classifying, name entities for each textual sequence of the target textual data; generating, a semantic textual similarity model and using the semantic textual similarity model to generate textual embedding of the identified name entities for each textual sequence of the target textual data; aggregating the textual embedding with the structural features of the target textual data to generate target document embedding; searching, by a similarity estimator, for similar documents by measuring similarities of the target document embedding with embeddings of each source document of the set of source documents; and computing explanatory information about a degree of similarity between the target document and any document of the set of the source documents.

Abstract: The present disclosure describes techniques for improving data subsampling for recommendation systems. A user-item graph associated with training data may be constructed. An importance of user-item interactions may be estimated via graph conductance based on the user-item graph. An importance of the training data may be measured via sample hardness using a pre-trained pilot model. A subsampling rate may be generated based on the importance estimated from the user-item graph and the importance measured by the pre-trained pilot model.

Abstract: A system allows users to define alternate scenarios in a database to allow users to analyze the data to determine what happens if certain values of some of the fields were different from the values stored in the database during its actual operation. The system stores a source dataset representing data of a source database, for example, a production database. The system receives an alternate scenario specification that includes instructions for determining alternate values for a portion of the database. The system determines an alternate dataset comprising alternate values for a portion of the database as defined by the alternate scenario specification. The system receives requests to process database queries for one or more alternate scenarios. The system combines data from the source dataset and the alternate datasets corresponding to the alternate scenarios specified by the request to process database queries.

Abstract: A method for determining relationships of entities associated with interactions may include receiving interaction data associated with interactions between first and second entities. The interaction data may include first entity identification data, second entity identification data, and relative timing data. A node may be generated for each second entity. A set of edges may be generated for each first entity to include an edge connecting the node associated with second entity identification data of each interaction to the node associated with second entity identification data of a next interaction based on the relative timing data. Sample data associated with a portion of the nodes/edges may be generated. A vector for each node of the sample may be generated. A distance between each vector and other vectors may be determined. A relationship between each second entity may be determined based on the distance. Systems and products are also disclosed.

Abstract: A method for training a control system model includes introducing a simulated fault into a software simulation of a physical system and generating emulated sensor data based on the simulated fault, where the emulated sensor data emulates output from one or more sensors of the physical system. The method further includes obtaining output data from a test control system provided with the emulated sensor data, where the test control system emulates a control system of the physical system and tagging the output data with the simulated fault to create training data. The method further includes utilizing the training data to train the control system model, where the control system model is a machine learning model for use with the control system of the physical system during operation of the physical system.

Abstract: An optimized fact checking system analyzes and determines the factual accuracy of information and/or characterizes the information by comparing the information with source information. The optimized fact checking system automatically monitors information, processes the information, fact checks the information in an optimized manner and/or provides a status of the information. In some embodiments, the optimized fact checking system generates, aggregates, and/or summarizes content.

Abstract: A storage medium storing a program that causes a computer to execute a process that includes acquiring a first model that is trained based on training data which indicates a first combination of constituent values of a target object and an environmental value in an experiment on the target object with associating with a characteristic value and that specifies a mean value and a deviation value of the characteristic value; acquiring a second model that is trained based on training data which indicates a second combination of the constituent values and an allowable condition for the experiment, and that specifies the allowable condition; and generating a solution set for the first combination by performing multi-objective optimization by a penalty term based on the allowable condition, a first objective function, and a second objective function.

Abstract: A computer-implemented method, a computer program product, and a computer system for automatic adaptive client selection in federated learning. A server sends parameters of a machine learning model to all of clients, where all of the clients compute respective gradients using the parameters. The server receives sketches of the respective gradients, where the sketches are computed by all of the clients. The server uses the sketches to compute similarity between all of the clients and clusters the all of the clients based on the similarity. The server optimizes a number of client clusters and a dimension of the sketches, subject to a constraint of memory consumption, a constraint of communication overhead, and a performance metric. The server determines a subset of the clients that send the respective gradients, by selecting the clients from the client clusters. The server aggregates the respective gradients sent by the subset of the clients.