Abstract: Technology is disclosed for selecting at least one icebreaker people highlight, with respect to a person of interest in a target group, to be shown on one or more user devices. The target group is determined, along with a set of feature values in common among the target group. Similarly, a reference group related to the target group is determined, along with a set of feature values for the reference group. For each feature value in the set of common target group feature values, a frequency of the feature value with respect to the determined reference group is determined. Based on the determined frequency, at least one icebreaker people highlight from the determined set of feature values in common among the target group is caused to be presented on the user device.

Abstract: The principles described herein relate to the training and implementation of a model designed to estimate the probability of new security incidents being true incidents. This occurs in an environment where a service such as a SIEM monitors a network of computing systems and other resources and detects a variety of incidents that could be security threats. These incidents are reported to the SOC for investigation and the SOC will take appropriate action to mitigate potential threats of true security breaches. As part of the investigation process, the SOC can label whether a security incident is true, false or benign. After labeling enough security incidents a model can be produced to estimate the probability that new security incidents are true incidents. This would help the SOC filter through security incidents more efficiently and allow for quicker response of the most likely security breaches.

Abstract: Methods, systems, apparatuses, and computer-readable storage mediums are described for machine learning-based techniques for identifying a deployment environment in which computing resources (e.g., servers, virtual machines, databases, etc.) reside and for enhancing security for the identified deployment environment. For instance, usage data is collected from the computing resources. The usage data is featurized and provided to a machine learning-based classification model that determines a deployment environment in which the computing resources reside based on the featurized usage data. Once the deployment environment is identified, a security policy that is applicable for the identified deployment environment is determined. The security policy specifies a plurality of recommended security settings that should be applied to the computing resources included in the identified deployment environment. The recommended security settings may be provided to the user (e.g.

Abstract: A plurality of computing devices are provisioned configured to communicate on a mobile communications network operated, in part, by an edge computing network. The edge computing network is associated with a customer of a computing service provider. The edge computing network comprises computing and storage devices configured to extend computing resources of the computing service provider to the customer of the computing service provider. A selection is received of a SIM provider and a quantity of SIM profiles for enabling the plurality of computing devices to access the mobile communications network. SIM data corresponding to the quantity of SIM profiles is received. The SIM data is encrypted and received over an encrypted channel.

Abstract: A computing system receives an instruction to initiate audio presentation of electronic communications for a recipient, and outputs an audio presentation responsive to the instruction. The audio presentation includes an initial portion that includes a presentation road map, and a subsequent portion that includes audible output of text content of a plurality of unreviewed electronic communications for the recipient. The presentation road map identifies an estimated duration of time to present the subsequent portion of the audio presentation.

Abstract: Examples are disclosed that relate to the restoration of degraded images acquired via a behind-display camera. One example provides a method of training a machine learning model, the method comprising inputting training image pairs into the machine learning model, each training image pair comprising an undegraded image and a degraded image that represents an appearance of the undegraded image to a behind-display camera, and training the machine learning model using the training image pairs to generate frequency information that is missing from the degraded images.

Abstract: The description relates to deep learning cluster scheduler modular toolkits. One example can include generating a deep learning cluster scheduler modular toolkit that includes multiple DL scheduler abstraction modules and interactions between the multiple DL scheduler abstraction modules and allows user composition of the multiple DL scheduler abstraction modules to realize a deep learning scheduler.

Abstract: A data processing system implements receiving textual content from a first application on a first client device associated with a first user for analysis by a context adaptive writing assistant configured to provide suggestions for improving the textual content, obtaining contextual information indicative of a level of formality of the textual content; and categorizing the textual content as being associated with a first level of formality selected from a plurality of levels of formality. The system is further implements analyzing the textual content to identify one or more suggested improvements to the textual content; selecting a subset of suggested improvements from the one or more suggested improvements to the textual content to the first user based on the first level of formality associated with the textual content; sending the subset of suggested improvements to the first client device; and causing the first client device to display the subset of suggested improvements.

Abstract: A user can select a capacity setting for a transitional partition that determines the allocation between a low-density partition and a high-density partition in the transitional partition. The transitional partition can dynamically change among multiple settings having different capacities for the low-density partition. If the current setting of the transitional partition does not efficiently utilize the available storage space based on the user's preferences for storing different types of data in the low-density partition and the high-density partition, then the user can choose to change the transitional partition to a different setting that better suits the individual user's storage allocation preferences. Therefore, valuable storage space will not be under-utilized but instead will be repurposed for more efficient use by converting a low-density partition to a high-density partition, and vice versa.

Abstract: A wearable computing device, including a device body configured to be affixed to a body of a user. The wearable computing device may further include an inertial measurement unit (IMU) and a processor. The processor may receive kinematic data from the IMU while the device body is affixed to the body of the user. The processor may perform a first coordinate transformation on the kinematic data into a training coordinate frame of a training wearable computing device. At a first machine learning model trained using training data including training kinematic data collected at the training wearable computing device, the processor may compute a training-frame velocity estimate for the wearable computing device based on the transformed kinematic data. The processor may perform a second coordinate transformation on the training-frame velocity estimate to obtain a runtime-frame velocity estimate and may output the runtime-frame velocity estimate to a target program.

Abstract: A computing system including a plurality of processing devices configured to execute a Mixture-of-Experts (MoE) layer included in an MoE model. The processing devices are configured to execute the MoE layer at least in part by, during a first collective communication phase between the processing devices, splitting each of a plurality of first input tensors along a first dimension to obtain first output tensors. Executing the MoE layer further includes processing the first output tensors at a respective a plurality of expert sub-models to obtain a plurality of second input tensors. Executing the MoE layer further includes, during a second collective communication phase between the processing devices, receiving the second input tensors from the expert sub-models and concatenating the second input tensors along the first dimension to obtain second output tensors. Executing the MoE layer further includes outputting the second output tensors as output of the MoE layer.

Abstract: Systems and methods for efficiently querying distributed data are disclosed. An example method includes receiving a query from a device; transmitting the query to multiple distributed data stores; receiving, from the data stores, first data items satisfying the query; generating first-page results from the first data items; based on the first-page results, generating a first-page watermark; transmitting the first-page results and the first-page watermark to the device; receiving a second-page request from the device, the second-page request including the query and the first-page watermark; transmitting the query and at least a portion of the first-page watermark to the data stores; receiving, from the data stores, second data items satisfying the query and the portion of the first-page watermark; generating second-page results from the received second data items; and transmitting the second-page results to the device.

Abstract: A computing system is provided, including a plurality of processing devices configured to execute a Mixture-of-Experts (MoE) layer included in an MoE model. The MoE layer includes a plurality of expert sub-models that each have a respective plurality of parameter values. The MoE layer is configured to be switchable between a data parallel mode and an expert-data-model parallel mode without conveying the respective parameter values of the expert sub-models among the plurality of processing devices.

Abstract: Computing devices and methods for performing touch detection on a touch screen display and trackpad are disclosed. In one example, a trackpad input signal from a trackpad is received at a processor of the device. Using at least the trackpad input signal, a touch screen touch detection algorithm is modified. A touch screen input signal is then received at the processor from the touch screen display. The touch screen input signal is processed with the modified touch screen touch detection algorithm.

Abstract: A method of recommending an unplayed game application to a user includes obtaining a game history associated with a user profile, wherein the game history includes at least one game record of a game application the user profile has been used to play, and evaluating the game history to obtain a plurality of game events associated with a game application of the game history. The method further includes accessing a table of tags associated with the plurality of game events associated with the game application, aggregating a plurality of tags in a tag list in the user profile based at least partially on the table of tags and the plurality of game events, and presenting to the user a recommendation of an unplayed game application based at least partially on the plurality of tags.

Abstract: Techniques and tools for encoding enhancement layer video with quantization that varies spatially and/or between color channels are presented, along with corresponding decoding techniques and tools. For example, an encoding tool determines whether quantization varies spatially over a picture, and the tool also determines whether quantization varies between color channels in the picture. The tool signals quantization parameters for macroblocks in the picture in an encoded bit stream. In some implementations, to signal the quantization parameters, the tool predicts the quantization parameters, and the quantization parameters are signaled with reference to the predicted quantization parameters. A decoding tool receives the encoded bit stream, predicts the quantization parameters, and uses the signaled information to determine the quantization parameters for the macroblocks of the enhancement layer video. The decoding tool performs inverse quantization that can vary spatially and/or between color channels.

Abstract: System and methods for performing analytical operations are described. A hardware-based regular expression (RegEx) engine performs a regular expression operation on a stream of data units based on a finite automata (FA) graph. Performing includes configuring a regular expression engine of a hardware-based regular expression accelerator to, beginning at a root node in the plurality of nodes of the FA graph, step the regular expression engine through one or more nodes of the FA graph until the regular expression engine arrives at a skip node and to consume, at the skip node, two or more data units from the stream of data units before traversing one of the directional arcs to another node.

Abstract: A computing system including a processor configured to receive a stabilizer circuit specification of a stabilizer circuit that includes one or more elementary operations. The elementary operations are each selected from the group consisting of an allocation of one or more qubits in a stabilizer state, an allocation of one or more random classical bits, a Clifford unitary, a Pauli unitary conditional on respective parities of measurement outcomes and/or respective parities of the random classical bits, a joint multi-qubit Pauli measurement, and a destructive one-qubit Pauli measurement. The processor is further configured to compute a standardized stabilizer instrument specification of a stabilizer instrument based at least in part on the stabilizer circuit specification. The standardized stabilizer instrument specification includes an input Clifford unitary, an output Clifford unitary, and a plurality of bit matrices.

Abstract: A copy-on-write union filesystem. A computer system identifies a first filesystem namespace as a mutable layer, and a second filesystem namespace as an immutable layer. The computer system exposes a merged filesystem namespace as a union filesystem that merges the mutable layer and the immutable layer, and that overlays the first filesystem namespace over the second filesystem namespace. The computer system identifies an operation on the merged filesystem namespace that requests to read from a file within the merged filesystem namespace, the file being stored in the second filesystem namespace and not being stored in the first filesystem namespace. The computer system returns a handle referencing a proxy object stored in a system memory. The proxy object includes a mapping to a file object corresponding to the file as stored in the second filesystem namespace.

Abstract: The disclosure herein describes using a transcript generation model for generating a transcript from a multi-speaker audio stream. Audio data including overlapping speech of a plurality of speakers is obtained and a set of frame embeddings are generated from audio data frames of the obtained audio data using an audio data encoder. A set of words and channel change (CC) symbols are generated from the set of frame embeddings using a transcript generation model. The CC symbols are included between pairs of adjacent words that are spoken by different people at the same time. The set of words and CC symbols are transformed into a plurality of transcript lines, wherein words of the set of words are sorted into transcript lines based on the CC symbols, and a multi-speaker transcript is generated based on the plurality of transcript lines. The inclusion of CC symbols by the model enables efficient, accurate multi-speaker transcription.