Abstract: Embodiments for training a neural network using sequential tasks are provided. A plurality of sequential tasks are received. For each task in the plurality of tasks a copy of the neural network that includes a plurality of layers is generated. From the copy of the neural network a task specific neural network is generated by performing an architectural search on the plurality of layers in the copy of the neural network. The architectural search identifies a plurality of candidate choices in the layers of the task specific neural network. Parameters in the task specific neural network that correspond to the plurality of candidate choices and that maximize architectural weights at each layer are identified. The parameters are retrained and merged with the neural network. The neural network trained on the plurality of sequential tasks is a trained neural network.

Abstract: Disclosed are some implementations of systems, apparatus, methods and computer program products for implementing a software development tool that enables a set of computer-readable instructions to be requested and obtained in association with an element of a process flow represented in a layout. A set of specifications can be submitted in association with a request, where the set of specifications includes input-output value pairs. In response, the system can automatically generate a first set of computer-readable instructions according to the set of specifications, where the first set of computer-readable instructions implements a mapping between the input-output value pairs. In addition, the system can obtain a second set of computer-readable instructions according to the set of specifications, either from pre-existing software code or from individual(s) (e.g., software developer(s)) to whom the system has sent a request for computer-readable instructions.

Abstract: Systems and methods are provided for near-zero-cost (NZC) query framework or approach for differentially private deep learning. To protect the privacy of training data during learning, the near-zero-cost query framework transfers knowledge from an ensemble of teacher models trained on partitions of the data to a student model. Privacy guarantees may be understood intuitively and expressed rigorously in terms of differential privacy. Other features are also provided.

Abstract: A method is provided for processing a query. In some embodiments, the method includes creating a historical schema defining an ordered set of fields based on a current schema of an object and a field modification of the current schema. The method includes inserting, based on the field modification, a new field at an end of the historical schema and archiving records into a second storage device. The method includes receiving a query, which includes a command specifying a subset of fields and which indicates a non-current field. The method includes searching in the second storage device for archive records having a second schema including the subset of fields and the non-current field. The first and second schemas follow the field order in accordance with the historical schema. The method further includes accessing the subset of fields and the non-current field in accordance with the command.

Abstract: Embodiments described herein provide systems and methods for an Explicit Memory Tracker (EMT) that tracks each rule sentence to perform decision making and to generate follow-up clarifying questions. Specifically, the EMT first segments the regulation text into several rule sentences and allocates the segmented rule sentences into memory modules, and then feeds information regarding the user scenario and dialogue history into the EMT sequentially to update each memory module separately. At each dialogue turn, the EMT makes a decision among based on current memory status of the memory modules whether further clarification is needed to come up with an answer to a user question. The EMT determines that further clarification is needed by identifying an underspecified rule sentence span by modulating token-level span distributions with sentence-level selection scores. The EMT extracts the underspecified rule sentence span and rephrases the underspecified rule sentence span to generate a follow-up question.

Abstract: Systems and methods are described for applying a plurality of data points of a time series data set representing values of a metric measuring performance of a cloud computing service to a machine learning model to predict a forecast of a most likely value of the metric at a selected future time. The method includes determining whether the plurality of data points of the time series data set are anomalies according to the machine learning model and the forecast and generating a collective anomaly from the anomalies when the plurality of data points is determined to be anomalies. The method further includes determining whether the collective anomaly does not meet one or more cloud computing service level objective (SLO) threshold requirements and sending an alert when the collective anomaly does not meet one or more cloud computing SLO threshold requirements.

Abstract: Methods and systems are provided for generating an interactive simulation representing one or more assets based on one or more asset records. Based on information from asset records stored at a database system of a cloud-based computing system, an asset simulator module, executed at a cloud-based computing system, can generate one or more simulated representations of the assets. A simulator application executed at the cloud-based computing system can augment the simulated representations of the assets with (at least) additional information from the asset records stored in the database system, and generate a user interface that presents an interactive simulation of the assets. The user interface can include the simulated representations of the assets with the additional information from the asset records stored in the database system.

Abstract: A self-healing build pipeline architecture for a software application build job across a distributed computer platform comprises a public API that receives configuration data describing the build job, stores the configuration data in a decentralized database, serves requests to/from a pipeline tracker API, and stores states of build pipelines during the build job. The decentralized database stores the configuration data and a project identifier for the build, and metadata regarding states of the build pipelines collected across the distributed computer platform. The pipeline tracker API runs local to the build environment in the distributed computer platform and sends a build status to public API for updating the decentralized database. For any failures in any of the build pipelines, the state is retrieved from the decentralize database and a new build pipeline is triggered locally that resumes from the failed state to provide a self-healing build pipeline architecture.

Abstract: Techniques and structures to prevent exhaustion of a database connection pool, including receiving a plurality of events from a plurality of application servers operating as a database connection pool, categorizing each of the plurality of events, including generating a categorization message for each of the plurality events, wherein each categorization message includes one or more fields having contextual information to identify an event type to which an event is associated, analyzing the categorization messages to generate an event record; and generating a report based on the event record including one or more performance measurements to indicate an effectiveness of management of the database connection pool.

Abstract: System and methods are described for automated recovery from errors occurring during replication of a database. The method includes getting text from one or more log files generated during database replication processing in a cloud computing environment, transforming the text into a structured language form represented by vectors, and identifying patterns in the vectors. The method further includes classifying one or more errors based on the identified patterns using supervised learning as either a recoverable error or an unrecoverable error, analyzing the one or more errors to determine one or more recovery jobs associated with database replication processing in the cloud computing environment for each of the recoverable errors, and invoking the one or more recovery jobs.

Abstract: Approaches for multi-hop knowledge graph reasoning with reward shaping include a system and method of training a system to search relational paths in a knowledge graph. The method includes identifying, using an reasoning module, a plurality of first outgoing links from a current node in a knowledge graph, masking, using the reasoning module, one or more links from the plurality of first outgoing links to form a plurality of second outgoing links, rewarding the reasoning module with a reward of one when a node corresponding to an observed answer is reached, and rewarding the reasoning module with a reward identified by a reward shaping network when a node not corresponding to an observed answer is reached. In some embodiments, the reward shaping network is pre-trained.

Abstract: Disclosed herein are system, method, and computer program product embodiments for facilitating access to and incorporation of APIs within an application during development, while ensuring that the API satisfies quality constraints. These quality constraints are controlled by the application of rules, which result in a maturity score for the API, in accordance with embodiments. These rules may be provided by a variety of sources for use in scoring the API at various stages. For example, a developer group may control access to developers within the group by applying a ruleset that restricts APIs for applications written by the developer group to a threshold score. This threshold score can limit or restrict APIs that are offered to a developer by an API exchange. Additionally, a ruleset may be applied to enforce a minimum score for submission of an API for inclusion within an API exchange.

Abstract: A method for generating a model for recommendations from an item data set for a target data set includes embedding a set of targets from the target data set in a shared coordinate space using a first embedding function, embedding a first set of items from the item data set in the shared coordinate space using a second embedding function, selecting at least one target from the set of targets, and identifying a second set of items from the first set of items that are proximate to the at least one target as candidates from the recommendations.

Abstract: Methods and systems are provided for declaratively configuring a process and a page for a particular service provider. User interface(s) can be displayed at a user system and include options for declaratively configuring different log in processes and log in pages for the particular service provider. After receiving a selection of one of the options to configure the log in process and the log in page, inputs can be received that specify a type of identifiers associated with a user to be verified as part of the log in process, and an identity verification process to define how the user will be verified as part of the log in process for the service provider.

Abstract: Disclosed are some implementations of systems, apparatus, methods and computer program products for implementing a shim driver configurable to provide a database driver. A shim driver class is loaded from a class path and instantiated such that a shim driver is generated. A directory path and database driver class name are obtained from a configuration file. An application transmits a request for a database driver via an application programming interface (API) of the shim driver, where the request includes the directory path and database driver class name. The shim driver loads a database driver class identified by the database driver class name using the directory path, instantiates the database driver class such that a first database driver is generated, and provides the first database driver responsive to the request.

Abstract: The type system of a functional programming language, such as DataWeave and/or the like, may be extended to add valuable semantic information to data types (e.g., attributes of data that inform a compiler/interpreter how the data should be used, etc.) that is automatically generated and/or hidden from a developer/programmer. For example, the type system may be configured for flow typing, based on a graph of type dependencies between expressions, without modifying an existing implementation, except at places where extra information can be obtained. Different parts of the type system implementation may be modified, for example, based on one or more injected retyper nodes, to provide more and better type inference.

Abstract: Techniques are disclosed relating to efficiently managing skip list data structures. In some embodiments, a computing system stores a skip list including a plurality of key-value records that include one or more pointers to others of the plurality of key-value records. The computing system scans the plurality of key-value records in key order to identify key-value records to be purged from the skip list. The scanning includes maintaining a list of key-value records that include pointers that point to key-value records that have not yet been scanned by the scanning. In response to identifying a key-value record for purging, the computing system purges the key-value record by substituting the pointers included the key-value records of the list with pointers included in the key-value record being purged.

Abstract: Embodiments described herein provide a query autocompletion (QAC) framework at subword level. Specifically, the QAC framework employs a subword encoder that encodes or converts the sequence of input alphabet letters into a sequence of output subwords. The generated subword candidate sequences from the subword encoder is then for the n-gram language model to perform beam search on. For example, as user queries for search engines are in general short, e.g., ranging from 10 to 30 characters. The n-gram language model at subword level may be used for modeling such short contexts and outperforms the traditional language model in both completion accuracy and runtime speed. Furthermore, key computations are performed prior to the runtime to prepare segmentation candidates in support of the subword encoder to generate subword candidate sequences, thus eliminating significant computational overhead.