Abstract: A method includes receiving a plurality of files where each file represents an object. The method further combines selected files of the plurality of files into an aggregation file and stores the aggregation file on low-latency storage media, such as disk storage media. The method represents the aggregation file with a single inode that is stored on the low-latency storage media. The method migrates the aggregate file to high-latency storage media, such as tape storage media, while leaving the single inode representing the aggregation file on the low-latency storage media. By collecting files into larger aggregation files, the number of files archived is significantly reduced and the files archived are of larger size. The reduced number of files and increased size of the files improves archival and retrieval performance due to reduced metadata operation overhead. A corresponding system and computer program product are also disclosed.

Abstract: In some examples, a system for decorating network traffic flows with outlier scores includes a processor and a memory device to store traffic flows received from a network. The processor is configured to receive a set of traffic flows from the memory device and generate a tree model to split the traffic flows into clusters of traffic flows. Each cluster corresponds with a leaf of the tree model. The processor is further configured to generate machine learning models for each of the clusters of traffic flows separately. For a new traffic flow, the processor is configured to identify a specific one of the machine learning models that corresponds with the new traffic flow, compute an outlier score for the new traffic flow using the identified specific one of the machine learning models, and decorate the new traffic flow with the outlier score.

Abstract: Insulated phase change memory devices are provided that include a first electrode; a second electrode; a phase change material disposed in an electrical path between the first electrode and the second electrode; and a porous dielectric configured to concentrate heat produced by a reset current carried through the phase change material between the first electrode and the second electrode to mitigate an amount of heat that escapes from the phase change material. The porous dielectric may be an inherently porous dielectric material or a dielectric material in which porous structures are induced during fabrication. Methods of fabrication of such devices are also provided.

Abstract: Provided are a computer program product, integrated cache manager, and method for maintaining data in a first level memory and buckets representing regions of memory devices to extend data cache. A plurality of buckets represent distinct regions of memory devices. The buckets are associated with different threshold access count ranges. Data having an access count is stored in one of the buckets associated with a threshold access count range including the access count of the data to store. Data evicted from a first level memory is copied to an initial bucket comprising one of the buckets. Data is moved from a source bucket comprising one of the buckets, including the initial bucket, to a target bucket of the buckets having a target threshold access count range including an access count of the data to move.

Abstract: A system for initiating web containers and isolating network traffic to web containers is described. The system initiates and uses web containers for associated web-based resources, such as domain addresses owned by a single entity. When a web container for a domain address has not been initiated, the system determines an owner of the domain address and any domain addresses associated with the owner and initiates a web container for the domain address and the associated domains. When a user navigates to the domain addresses associated with the determined owner, the system isolates the network activity to the initiated web container.

Abstract: In an approach for determining the date of planting of a crop growing in an agricultural field, a processor receives an aerial image of one or more agricultural fields in a pre-determined geographical region. A processor selects a plurality of points from the aerial image. A processor calculates a Vegetation Index of one or more crops growing at the plurality of points selected. A processor compares the Vegetation Index calculated for the one or more crops growing at the plurality of points selected to the Vegetation Index known for a plurality of historical reference signatures. A processor generates an actual signature. A processor cross-correlates the actual signature against the plurality of historical reference signatures to measure a degree of similarity. A processor identifies the one or more crops growing in the one or more agricultural fields in the pre-determined geographical region from the cross-correlation.

Abstract: A method includes: creating, by a computing device, a virtual consolidation layer; replicating, by the computing device, snippets that are results of a query; copying, by the computing device, the replicated snippets to the virtual consolidation layer; performing, by the computing device and by a machine learning model, matching of particular ones of the snippets in the virtual consolidation layer; and producing, by the computing device, a search result comprising the matched snippets.

Abstract: A phase change memory bridge cell comprising a dielectric layer located on top of a at least one electrode, wherein a trench is located in the dielectric layer. A first liner located at the bottom of the trench in the dielectric layer and the first liner is located on the sidewalls of the dielectric layer that forms the sidewalls of the trench. A phase change memory material located on top of the first liner, wherein a top surface of the phase change memory material is aligned with a top surface of the dielectric layer, wherein the dielectric layer is located adjacent to and surrounding the vertical sidewalls of the phase change memory material, wherein a top surface of the phase change memory material is flush with a top surface of the dielectric layer.

Abstract: A computer system includes a processor core and a memory system in signal communication with the processor core. The memory system includes a first cache and a second cache. The first cache is arranged at a first level of a hierarchy in the memory system and is configured to store a plurality of first-cache entries. The second cache is arranged at a second level of the hierarchy that is lower than the first level, and stores a plurality of second-cache entries. The first cache maintains a directory that contains information for each of the first-cache entries. The second cache maintains a shadow pointer directory (SPD) that includes one or more SPD entries that maps each of the first-cache entries to a corresponding second cache entry at a lower-level cache location.

Abstract: Aggregated log index-based log record identification is provided by maintaining log files of a database system on a plurality log storage devices, the log files including log records of changes to database objects, indexing the log files to obtain a collection of log indexes, where the indexing includes, for each log storage device, indexing log records stored on the log storage device to obtain a respective log index, then based on a request, identifying log storage devices that store log records for servicing the request, the log storage devices storing log indexes, of the collection of log indexes, that index an aggregate set of log records on the log storage devices, obtaining the log indexes from the log storage devices, and aggregating the log indexes to provide an aggregated log index that provides an index of the aggregate set of log records on the log storage devices.

Abstract: Monolithically stacked VTFET devices having source/drain contacts with increased contact area and dielectric isolation are provided. In one aspect, a stacked VTFET device includes: at least a bottom VTFET below a top VTFET, wherein the bottom VTFET and the top VTFET each includes source/drain regions interconnected by a vertical fin channel, and a gate stack alongside the vertical fin channel; and source/drain contacts to the source/drain regions, wherein at least one of the source/drain contacts is in direct contact with more than one surface of a given one of the source/drain regions. A stacked VTFET device having at least a bottom VTFET1 below a top VTFET1, and a bottom VTFET2 below a top VTFET2, and a method of forming a stacked VTFET device are also provided.

Abstract: Computer based methods, systems, and computer readable media are provided for intelligently sorting cells using machine learning. A biological cell analysis sorting machine, wherein the biological cell analysis sorting machine comprises a flow cytometry system and a cell analytics sorting system, may be configured to detect configuration issues by analyzing results of a sorting experiment performed by the biological cell analysis sorting machine. An analysis of a history of prior sorting experiments and associated configuration settings may be performed and a corpus of documents pertaining to the sorting experiment based on the detected configuration issues may be analyzed. Updated configuration settings for the biological cell analysis sorting machine based on the performed analysis may be determined, and the biological cell analysis sorting machine may be configured with the updated configuration settings to conduct a desired sorting experiment.

Abstract: Methods, computer program products, and/or systems are provided that perform the following operations: obtaining metadata associated with a content item, wherein the metadata includes at least one runtime metadata definition; identifying an environment for providing access to the content item based on a runtime metadata definition; implementing the environment based, at least in part, on the runtime metadata definition; and providing access to the content item in the environment.

Abstract: A system, having a memory that stores computer executable components, and a processor that executes the computer executable components, reduces data size in connection with training a neural network by exploiting spatial locality to weight matrices and effecting frequency transformation and compression. A receiving component receives neural network data in the form of a compressed frequency-domain weight matrix. A segmentation component segments the initial weight matrix into original sub-components, wherein respective original sub-components have spatial weights. A sampling component applies a generalized weight distribution to the respective original sub-components to generate respective normalized sub-components. A transform component applies a transform to the respective normalized sub-components.

Abstract: The network comprises at least one network layer in which a plurality of electronic oscillators, interconnected via programmable coupling elements storing respective network weights, generate oscillatory signals at time delays dependent on the input signal to propagate the input signal from an input to an output of that layer. The network is adapted to provide a network output signal dependent substantially linearly on phase of oscillatory signals in the last layer of the network. The method includes calculating a network error dependent on the output signal and a desired output for the training sample, and calculating updates for respective network weights by backpropagation of the error such that weight-updates for a network layer are dependent on a vector of time delays at the input to that layer and the calculated error at the output of that layer.

Abstract: Embodiments of a method are disclosed. The method includes performing distributed deep learning training on multiple batches of training data using corresponding learners. Additionally, the method includes determining training times wherein the learners perform the distributed deep learning training on the batches of training data. The method also includes modifying a processing aspect of the straggler to reduce a future training time of the straggler for performing the distributed deep learning training on a new batch of training data in response to identifying a straggler of the learners by a centralized control.

Abstract: A first set of features associated with a neural network are parameterized. A decision tree is generated from the first set of features. One or more adjustments for the neural network are received at the decision tree. A second set of features associated with the adjustments at the decision tree are parameterized. The parameterized first and second set of features are combined into a plurality of parameters. From the plurality, an adjusted neural network is generated.

Abstract: Demand-based deployment of a font server to an edge device is facilitated by identifying a need for font-related data that is currently unavailable at a user device. The process also includes deploying a font server to an edge device of the computing environment to facilitate distribution of the font-related data to the user device.

Abstract: A method, system and computer program product for a heuristic determination of in-process damage class control to manage expected output product category. The heuristic technique determines the predicted damages and their ranges while keeping the initial expected defects, the respective classes and range of defect and mitigation. The method dynamically computes a damage mitigation range of operation while being within the overall constraints and completes the computation in smaller number of loops being run at the edge computers so that the manufacturing equipment can operate at a higher velocity for higher quality of the output. The method includes a step of reducing error of the co-efficient and damage counts. An Internet of Things (IoT) based robot is used to mitigate the damages in the manufacturing steps to ensure that the output class of the product remains what was expected at the start despite damages and mitigation measures.

Abstract: A method, apparatus and computer program product to defend learning models that are vulnerable to adversarial example attack. It is assumed that data (a “dataset”) is available in multiple modalities (e.g., text and images, audio and images in video, etc.). The defense approach herein is premised on the recognition that the correlations between the different modalities for the same entity can be exploited to defend against such attacks, as it is not realistic for an adversary to attack multiple modalities. To this end, according to this technique, adversarial samples are identified and rejected if the features from one (the attacked) modality are determined to be sufficiently far away from those of another un-attacked modality for the same entity. In other words, the approach herein leverages the consistency between multiple modalities in the data to defend against adversarial attacks on one modality.