Abstract: This disclosure provides a filtering mechanism to manage anomalous security data items. An anomalous security data item is provided to an analysis engine (such as a Large Language Model (LLM) or another form of generative language model) for interpretation. By curating a selection of one or more relevant non-anomalous security data items to provide with the anomalous data item, the filtering mechanism enables the analysis engine to perform with increased accuracy, without requiring the analyst engine to process large numbers of data items to ascertain their relevance to the anomalous security data item.

Abstract: A computerized method for processing packets in parallel using a reorder queue is described. As packets are received, it is determined which packet is the first packet in the flow. Once identified, the first packet in the flow, along with all other packets in the flow, is sent to the reorder queue while only a copy of the first packet in the flow is sent to a packet scheduler for processing on one of a plurality of processing cores. After the copy of the first packet in the flow is processed, the reorder queue releases each of the packets for the flow from the reorder queue in an order in which they were received. Thereafter, each of the packets released from the reorder queue are processed based on the processing of the copy of the first packet.

Abstract: A computerized method performs fusion of segmentation masks to generate a refined segmentation mask. A plurality of segmentation masks, each including a group of segments, is obtained for a geographic area. For each segmentation mask, a subgroup of segments is filtered from the group of segments of the segmentation mask based on areas of the subgroup of segments. Pairs of segments of the filtered subgroup of segments are matched to form matched segment groups in the geographic area. Representative segments of the geographic area are selected from the matched segment groups to generate a refined segmentation mask including the selected representative segments. A segment-specific action is performed on at least one of the selected representative segments of the generated refined segmentation mask of the geographic area. In some examples, the segment-specific action is for precision farming in agriculture domain.

Abstract: A model that includes a generative network mapping a latent vector to a feature vector, wherein weights in the generative network are modelled as probabilistic distributions, is trained on observed data points with incomplete observations of the feature vector to learn values of the weights. From a plurality of potential next features to observe, a target feature is selected that maximizes a measure of expected reduction in uncertainty in the distribution of the weights given the observed data points so far. A target data point comprising at least the target feature is then requested, and the generative network is further trained based on the target data point received in response to the request.

Abstract: A computer program predictor is described which has a processor configured to access a program attribute predictor; and a memory storing a search component configured to search a space of possible programs, to find a program which, given an input data instance and an output data instance, will compute the output data instance from the input data instance, the search being guided by attributes predicted by the attribute predictor given the input data instance and the output data instance.

Abstract: Independent refresh rates may be enabled for multiple monitors. Any displays that are presenting content in a full screen mode are refreshed at a rate set by the application presenting the content. All other devices are refreshed at a rate that is determined by a window manager based on various factors including the type of application(s) executing, the size of the applications presenting content, the location of the applications presenting content, hardware configurations, and so forth.

Abstract: A source enclave of a source application includes: at least one process thread; a respective at least one process stack memory; a heap memory; and a thread context area. An interrupt is sent to the source enclave which causes the process thread to exit. A migrator thread is sent to the source enclave to save to an external memory, using a migrator stack memory, the thread context area, the at least one process stack memory, and the heap memory, but not the migrator stack memory. A destination enclave is instantiated at a destination application. An initiator thread is sent to the destination enclave to clone, using an initiator stack memory, the state of the source enclave from the external memory.

Abstract: Systems and methods for characterizing and forecasting evolving query workloads. The method includes receiving a query, the received query including a parameter value and an arrival time; identifying the query as a recurrent query; extracting a query template from the received query by parsing the received query; based at least on the identifying, generating a feature vector for the received query, the feature vector generated based on the extracted template and the parameter value; and forecasting a future query based on the generated feature vector by applying a neural network.

Abstract: The disclosure herein describes training a document recommendation model using loss data generated from a linear score difference vector. A training data entry is provided including a query and a set of candidate documents. A document recommendation model generates a set of document prediction scores indicative of a likelihood that the candidate documents are responses to the query and a pairwise score difference matrix is generated using the set of document prediction scores. The pairwise score difference matrix is transformed into a score difference vector using a correct document vector that indicates a correct document among the set of candidate documents. Loss data of the document recommendation model is generated using the score difference vector and the document recommendation model is adjusted using the calculated loss data. Training the document recommendation model based on the linear score difference vector reduces resource usage when compared to training with a difference matrix.

Abstract: A method for streaming videos with internally-variable frame quality is implemented via a computing system including a processor. The method includes accessing a video including and evenly sampling the video frames. The method includes, for each sampled video frame, analyzing the video frame using a CNN model to determine whether the video frame includes high-relevance region(s). The method also includes, for each sampled video frame including high-relevance region(s), extracting coordinates of the high-relevance region(s) using the CNN model, and for each sampled video frame including high-relevance region(s) and each intervening video frame between the sampled video frame and a next sampled video frame, setting a minimum frame quality for the extracted coordinates.

Abstract: An experiment design is determined for a plurality of physical experiment entities, based on a training loss that is dependent on a critic function and an action parameter individually associated with each physical experiment entity, with the aim of increasing (e.g., optimizing) information gain with respect to a plurality of test entities. The training loss encodes a predicted information gain between a predicted experiment outcome and a predicted test quantity. The predicted experiment outcome associated therewith is sampled from a joint probability distribution based on an entity context. A numerical output is computed using the critic function applied to the predicted experiment outcome and the predicted test quantity.

Abstract: An imaging apparatus comprises two actuators, such as an autofocus actuator and optical image stabilizer. The actuators are nested, wherein the outer actuator is suspended from the device body and the inner actuator is suspended from the outer actuator. A suspension element may be a flexure bearing, allowing a flat actuator design.

Abstract: A click-to-script service enables developers of big-data job scripts to quickly see the underlying script operations from optimized execution plans. Once a big-data job is received, the disclosed examples compile it and generate tokens that are associated with each operation of the big-data job. These tokens include may include the file name of the job, the line number of the operation, and/or an Abstract Syntax Tree (AST) node for the given operations. An original execution plan is optimized into an optimized execution plan, and the tokens for the original operations of the job script are assigned to the optimized operations of the optimized execution plan. The optimized execution plan is graphically displayed in an interactive manner such that users may view the optimized execution plan and click on its optimized operations to find the original operations of the job script.

Abstract: A computer-implemented method for performing a document search action includes receiving a search request identifying a sample document. A plurality of terms occurring within the sample document is identified. Each term is scored based on a global term frequency of that term within a collection of documents. Each term is then weight based on proximity, within the sample document, of that term to at least one of the other terms of the first plurality of terms. A document search query is created to include a proximity-limiting clause restricting results to include documents that have a highest-weighted term of the first plurality of terms within a threshold distance of another term of the first plurality of terms. The document search query is performed, thereby resulting in identification of a first resulting document.

Abstract: A method, computer program product, and computing system for enhancement of audio signals received from a plurality of microphones. A multichannel audio signal is received from a plurality of microphones and is processed with a short-time discrete cosine transform (STDCT) to generate a real-valued spectral representation of the multichannel signal encoding both magnitude and phase information. Magnitude- and phase-dependent weights are generated, and an enhanced single-channel signal is produced based upon, at least in part, the spectral representation of the multichannel signal and the magnitude- and phase-dependent weights.

Abstract: A distillation system extracts knowledge from a large pre-trained sequence-to-sequence neural transformer model into a smaller bi-encoder. The pre-trained sequence-to-sequence neural transformer model is trained to translate data from a first domain into a second domain on a large corpus. A teacher model is generated from the pre-trained model by fine-tuning the pre-trained neural transformer model on a smaller translation task with true translation pairs. The fine-tuned model is then used to generate augmented data values which are used with the true translation pairs to train the bi-encoder. The bi-encoder is used for perform cross-domain searches.

Abstract: A computer-implemented method includes receiving an incorrect prediction output by a trained machine learning model, which has been trained using training data items. The method includes identifying a training data item used to train the model that is a cause of the incorrect prediction, by determining an impact on performance of the trained machine learning model associated with removing the training data item from the plurality of training data. The trained model can then be updated to remove the effect of the identified training data item, allowing the model to be automatically corrected in view of poor quality training data.

Abstract: A tool analyzes source code of a program that targets a parallel processing system, and searches for parallelism unstructured behavior values that potentially inhibit parallelism efficiency. Example parallelism unstructured behavior values include particular memory addresses, memory masks, control divergences, and instruction predicates, which are identified according to their context and use in the program. The tool also locates program operations that contribute to these values, and determines a source of parallelism structure information in the program. In some scenarios, the tool populates a pattern data structure which is suitable to help guide code generation. Patterns detected include addressing patterns, mask patterns, and thread control patterns. Programs analyzed include single instruction multiple data programs and single instruction multiple thread programs.

Abstract: Example solutions for reducing the likelihood of hallucinations by large language models (LLMs) and improving the relevance of generated text are disclosed. A set of multiple of machine learning (ML) models, each custom-trained to identify a different specific topic within text, dynamically generate multiple topic-specific prompts from a received email in an email thread. A user, responding to the prompts, produces material that is used in conjunction with customer relations management (CRM) data and enterprise suite data (e.g., calendar/schedule information) to dynamically generate an LLM prompt. An LLM, using the prompt, generates output text suitable for a business correspondence, such as a reply email, to the sender of the email. The combination of the responses to the multiple topic-specific prompts and the CRM and enterprise suite data in the LLM prompt both ensures relevance of the reply correspondence and minimizes the risk of hallucinations.

Abstract: Bi-directional tunnel probing in a network may be extended to multi-realm networks, providing bi-directional probing in a multi-realm network. Bi-directional probing uses probe packets with a forward tunnel label, a reverse tunnel label, and an IP packet header. Bi-directional probing in a multi-realm network uses a forward tunnel label, an SID, a reverse tunnel label, and an IP packet header. Penultimate hop popping strips the outermost labels in a specific order. The SID gets the probe packet back into the originating realm, and the reverse tunnel label returns the probe packet along the reverse direction of the tunnel. No SID is needed for intra-realm probing, and the IP packet header is used to return the probe packet if the reverse tunnel label is absent.