Abstract: Various embodiments described herein provide for performing garbage collection with reduced buffer usage on a memory system. In particular, various embodiments use program suspend and program resume during garbage collection to facilitate performance of the garbage collection using a buffer that is smaller in size than buffers used during conventional garbage collection processes.

Abstract: System and methods are disclosed relating media intelligence for a company relating to a topic and/or theme. In some examples, media intelligence parameter data and company historical data can be received, which can be used to generate a subject search parameter. The subject search parameter can include one or more phrases, words, sentences, and/or categories for the topic and/or theme. Data for the topic and/or theme from a number of private and/or media data sources can be queried based on the subject search parameter. The queried data can be aggregated to provide aggregated data. The aggregated data can be filtered to provide filtered data. The filtered data can indicate a position of the private and/or media data sources on the topic and/or theme. A recommendation can be provided for the topic and/or theme using a machine learning model.

Abstract: Techniques can include performing sector aligned data deduplication for a sequence of pages including: performing a search phase and identifying a target set of one or more target pages, wherein a first target page of the target set is included in a second sequence of previously written pages; and performing an expansion phase including: based on the first target page, determining matches between i) pages of the first sequence, and ii) corresponding generated sector aligned pages each constructed from two logically consecutive pages of the second sequence given an offset; and determining, based on the expansion phase and the first target page, a set of sector aligned duplicates within the sequence of pages, wherein each sector aligned duplicate identifies a respective page of the sequence having its content from two stored pages which are aligned on the page boundary and are logically consecutive pages of the second sequence.

Abstract: From many features and many multidimensional points, a computer generates exploratory training configurations. Each point contains a value for each of the features. Each exploratory training configuration identifies a random subset of the features and a random subset of the points. A performance score is generated for each of the exploratory training configurations. A feature weight is generated for each of the features that is based on the performance scores of the exploratory training configurations whose random subset of features contains the feature. A point weight is generated for each of the points that is based on the performance scores of the exploratory training configurations whose random subset of the many points contains the point. A machine learning model is trained using an optimized training corpus that consists of a subset of the many features based on feature weight and a subset of the many points based on point weight.

Abstract: A method of automated database element processing includes training a wide machine learning model with historical feature vector inputs to generate a wide ranked element output. The method includes training a deep machine learning model with the historical feature vector inputs to generate a deep ranked element output. The method includes generating a set of inputs specific to an individual entity, obtaining a set of current article database elements, and creating a feature vector input according to the set of inputs and the set of current article database elements. The method includes processing the feature vector input with the wide machine learning model to generate a wide ranked element list, processing the feature vector input with the deep machine learning model to generate a deep ranked element list, and merging database elements of the wide and deep ranked element lists to generate a ranked element recommendation output.

Abstract: Methods, systems, apparatuses, devices, and computer program products are described. A system may obtain a set of documents associated with a knowledge base for retrieval-augmented generation (RAG). The system may generate multiple representations of the information included in the documents using multiple knowledge extraction pipelines. For example, the system may generate a set of metadata-based vector embeddings based on the documents, a set of knowledge graphs based on the documents, and a set of hierarchical tree representations based on the documents. The system may receive a user query and may retrieve contextual information from the set of vector embeddings, the set of knowledge graphs, and the set of hierarchical tree representations to augment the user query for a large language model (LLM) prompt. The system may input the prompt to the LLM, and the LLM may output a response based on the user query and the contextual information.

Abstract: One embodiment sets forth a technique for performing matrix operations. The technique includes traversing a tree structure to access one or more non-empty regions within a matrix. The tree structure includes a first plurality of nodes and a second plurality of nodes corresponding to non-empty regions in the matrix. The first plurality of nodes includes a first node representing a first region and one or more second nodes that are children of the first node and represent second region(s) with an equal size formed within the first region. The second plurality of nodes include a third node representing a third region and one or more fourth nodes that are children of the third node and represent fourth region(s) with substantially equal numbers of non-zero matrix values formed within the third region. The technique also includes performing matrix operation(s) based on the non-empty region(s) to generate a matrix operation result.

Abstract: Context-based prompt generation techniques for generative machine-learning models are described. In one or more examples, a user interface (UI) selection of a UI element of first digital content is received. A processing device captures contextual information associated with the UI selection. A machine-learning model generates a prompt based on the UI selection and the contextual information. Generative artificial intelligence (AI) generates second digital content based on the prompt. The generative AI is implemented using one or more machine-learning models. The processing device then presents the second digital content for display in a user interface.

Abstract: A system can be provided for controlling resource transfers based on unique digital identifier data. For example, the system can receive a resource request from a user device. The resource request can be a request to transfer a resource from a first user account to a second user account based on a document. The system can then identify a unique digital identifier for the document. The system can further query a centralized database using the unique digital identifier to determine, based on a status indicator stored in the centralized database and associated with the unique digital identifier, a status of the document. Additionally, based on determining that the status of the document is a pre-processing status, the system can authorize the transfer of the resource and update the status indicator to change the status of the document from the pre-processing status to the processing status.

Abstract: A user of a network-based system may correspond to a user profile that describes the user. The user profile may describe the user using one or more descriptors of items that correspond to the user (e.g., items owned by the user, items liked by the user, or items rated by the user). In some situations, such a user profile may be characterized as a “taste profile” that describes an array or distribution of one or more tastes, preferences, or habits of the user. Accordingly, the user profile machine within the network-based system may generate the user profile by accessing descriptors of items that correspond to the user, clustering one or more of the descriptors, and generating the user profile based on one or more clusters of the descriptors.

Abstract: The technology described herein is directed to generating prompts for a language model, such as a large language model. In this regard, an enterprise interpretation model may receive a request for assistance with an enterprise application record. In response to receiving the request for assistance, the enterprise interpretation model may determine data corresponding to the record. Based on the determined data, the enterprise interpretation model may generate a query corresponding to the record. The query may include at least a subset of the data corresponding to the record. The query may then be provided to the language model for a response to be generated.

Abstract: A hybrid artificial intelligence (AI) agent balances use of a plurality of predefined decision tree flows and one or more large language model (LLM) chatbots. A query is received from a user client device that is associated with a user. A routing LLM of the hybrid AI agent determines an intent of the user based in part on context associated with the query. The routing LLM selects a response model from the plurality of predefined decision tree flows and the one or more LLM chatbots based in part on the intent. The selected response model is used to determine a response to the query. The response is provided to the user client device.

Abstract: A data management system may maintain a total-size counter for the file system. The total-size counter represents a sum of data size related to snapshots backed up to the file system. The data management system may perform data operation cycles to the file system. Each data operation cycle corresponds to a snapshot that includes files. Each data operation cycle may include incrementing the total-size counter by the data size of the files in the snapshot exchanged with the file system and adding, to a snapshot record, the amount of increment in incrementing the total-size counter as an increment-size counter. A data management system may perform a correction operation to correct the total-size counter. The correction operation may change the total-size counter by a difference between the total of the increment-size counters in the snapshot records and the total data size of file data exchanged with the file system.

Abstract: Automated content recommendation is used in conferencing. In one embodiment, a system receives a list of content recommendation actions. The system receives a number of utterances associated with the participants in real time. For each utterance, the system determines whether a prediction of relatedness is present between the utterance and one or more trigger phrases associated with a content recommendation action. Upon determining that a prediction of relatedness is present, the system performs the associated content recommendation action by transmitting, to one or more client devices, one or more pieces of content to be recommended.

Abstract: In one aspect, a query management module executing on a processor receives, from a large language model (LLM), a graph query generated by the LLM based on a natural language query (NLQ). A validation module identifies an error in the graph query. The query management module provides an indication of the error to the LLM. The query management module receives a modified graph query from the LLM. The validation module validates the modified graph query. Based on the validation of the modified graph query, the query management module executes the modified graph query against a knowledge graph to return a result as a response to the NLQ.

Abstract: A request for information (RFI) system is provided for use in communications networks including broadcast networks and the Internet. In one implementation, a code identifying an item of media content of interest (e.g. television, newspaper, magazines, billboards, radio) is captured and input to an RFI system that includes stored media tags and a search tool for matching inputs to the stored media tags. Upon receipt of the captured code, the RFI system matches the captured code with the stored media tags and provides a response to the user based on the match. The response may include or relate to follow-on or premium information relating to the content of interest. Using this information, an RFI data center or an RFI platform may credit value to a rewards account established for the network user based on the user's verified consumption of assets and/or data requests.

Abstract: A method of latent intent clustering is provided. The method comprises encoding identified features in a number of electronic user reports in a database. A binary matrix is created, wherein each row of the binary matric represents a different report and each column represents a different available feature. A 1 is placed in each cell of the matrix that matches a feature present in a user report. Cosine similarities are calculated for the user reports, and a similarity matrix is created, wherein each row and column of the binary matrix represents a different report, and wherein the cosine similarities of the reports are placed in corresponding cells of the matrix. The reports are clustered according to the cosine similarities. Features common to reports in each cluster are identified, and an intent of each report cluster is labeled according to the common features.

Abstract: Disclosed is an approach to autonomously abort a lower priority transaction that is blocking a higher priority transaction, where the lower priority transaction is holding a lock on a resource for which there is contention. Some approaches also provide an efficient way to implement identification of a blocker transaction, which drastically reduces the response time to abort the blocker transaction.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for a source independent query language. One of the methods includes receiving, by a computer system, a statement to perform operations on data stored in a data storing application, the data including fields. The method includes identifying a query language based on the data store. The method includes transforming the received statement into a query language specific statement. The method includes identifying at least one operation of the operations that is not supported by the query language specific statement. The method includes obtaining a function to replace the unsupported operation. The method includes applying the function to the query language specific statement. The method also includes providing the query language specific statement to the data storing application for processing.

Abstract: Name caching can be used external to a distributed file system (DFS), such as a network file system (NFS), to reduce the complexity of facilitating a file operation at the DFS. A system that can use the name caching and request the file operation of the DFS can comprise a processor and a memory that stores executable instructions that, when executed by the processor, facilitate performance of operations. The operations can comprise, in response to a lookup operation or a file operation performed by a DFS, storing, using the processor, a name cache entry for each directory element accessed during the lookup operation or the file operation. The operations further can comprise sending, to the DFS, a single request that comprises a compound command comprising requests for a set of lookup operations and for a file operation.