Abstract: Techniques for joint optimization of user segments and delivery channels are described. In one aspect, a method, includes obtaining activity data from a user device associated with a user, selecting, using a selector of a machine learning model, a user segment for the user based on the activity data, mapping, using a mapping function of the machine learning model, activity data for the user segment to features defined by multiple media channels, each media channel assigned a resource component, generating, using an objective predictor of the machine learning model, an objective prediction for the user segment based on the features and resource components of the media channels, the objective prediction identifying a media channel from the multiple media channels with a composite scalar metric above a defined threshold, and providing content to the user device via the media channel. Other embodiments are described and claimed.

Abstract: Systems and methods for identifying key moments, such as key moments within a livestream, are described. Embodiments of the present disclosure obtain video data and text data. In some cases, the text data is aligned with a timeline of the video data. The system then computes a moment importance score for a time of the video data using a machine learning model based on the video data and the text data, and presents content to a user at the time of the video data based on the moment importance score.

Abstract: Systems and methods for delivery aware audience segmentation and subsequent delivery of content are described. Embodiments are configured to obtain activity data for a user, assign the user to a user segment based on the activity data using a machine learning model, generate a reach prediction for the user segment, select a media channel for communicating with the user based on the user segment and the reach prediction, and provide targeted content to the user via the selected media channel. According to some aspects, the machine learning model is trained based on content reach data.

Abstract: A system includes a representation generator subsystem configured to execute a user representation model and a task prediction model to generate a user representation for a user. The user representation model receives user event sequence data comprises a sequence of user interactions with the system. The task prediction model is configured to train the user representation model. The user representation includes a vector of a predetermined size that represents the user event sequence data and is generated by applying the trained user representation model to the user event sequence data. A storage requirement of the user representation is less than a storage space requirement of the user event sequence data. The system includes a data store configured for storing the user representation in a user profile associated with the user.

Abstract: The present disclosure relates to systems, methods, and non-transitory computer-readable media that implements a secure distributed data collaboration architecture for generating synthetic datasets. For example, the disclosed system sends a request to perform a data collaboration with a first dataset of a first local node and a second dataset of a second local node. The disclosed system receives intermediate feature maps from the local nodes that correspond with the datasets and generates a combined feature map. Further, the disclosed system generates a synthetic dataset from the combined feature map by utilizing a central generative model. Moreover, the synthetic dataset generated by the disclosed system is statistically representative of the first dataset and the second dataset.

Abstract: Systems and methods for identifying key moments, such as key moments within a livestream, are described. Embodiments of the present disclosure obtain video data and text data. In some cases, the text data is aligned with a timeline of the video data. The system then computes a moment importance score for a time of the video data using a machine learning model based on the video data and the text data, and presents content to a user at the time of the video data based on the moment importance score.

Abstract: A resource control system is described that is configured to control scheduling of executable jobs by compute instances of a service provider system. In one example, the resource control system outputs a deployment user interface to obtain job information. Upon receipt of the job information, the resource control system communicates with a service provider system to obtain logs from compute instances implemented by the service provider system for the respective executable jobs. The resource control system uses data obtained from the logs to estimate utility indicating status of respective executable jobs and an amount of time to complete the executable jobs by respective compute instances. The resource control system then employs a machine-learning module to generate an action to be performed by compute instances for respective executable jobs.

Abstract: Techniques are provided for scheduling multiple jobs on one or more cloud computing instances, which provide the ability to select a job for execution from among a plurality of jobs, and to further select a designated instance from among a plurality of cloud computing instances for executing the selected job. The job and the designated instance are each selected based on a probability distribution that a cost of executing the job on the designated instance does not exceed the budget. The probability distribution is based on several factors including a cost of prior executions of other jobs on the designated instance and a utility function that represents a value associated with a progress of each job. By scheduling select jobs on discounted cloud computing instances, the aggregate utility of the jobs can be maximized or otherwise improved for a given budget.

Abstract: A system includes a representation generator subsystem configured to execute a user representation model and a task prediction model to generate a user representation for a user. The user representation model receives user event sequence data comprises a sequence of user interactions with the system. The task prediction model is configured to train the user representation model. The user representation includes a vector of a predetermined size that represents the user event sequence data and is generated by applying the trained user representation model to the user event sequence data. A storage requirement of the user representation is less than a storage space requirement of the user event sequence data. The system includes a data store configured for storing the user representation in a user profile associated with the user.

Abstract: A method performed by one or more processors that preserves a machine learning model comprises accessing model parameters associated with a machine learning model. The model parameters are determined responsive to training the machine learning model. The method comprises generating a plurality of model parameter sets, where each of the plurality of model parameter sets comprises a separate portion of the set of model parameters. The method comprises determining one or more parity sets comprising values calculated from the plurality of model parameter sets. The method comprises distributing the plurality of model parameter sets and the one or more parity sets among a plurality of computing devices, where each of the plurality of computing devices stores a model parameter set of the plurality of model parameter sets or a parity set of the one or more parity sets. The method comprises accessing, from the plurality of computing devices, a number of sets comprising model parameter sets and at least one parity set.

Abstract: Systems and methods for product retrieval are described. One or more aspects of the systems and methods include receiving a query that includes a text description of a product associated with a brand; identifying the product based on the query by comparing the text description to a product embedding of the product, wherein the product embedding is based on a brand embedding of the brand; and displaying product information for the product in response to the query, wherein the product information includes the brand.

Abstract: Certain embodiments involve extracting seasonal, level, and spike components from a time series of metrics data, which describe interactions with an online service over a time period. For example, an analytical system decomposes the time series into latent components that include a seasonal component series, a level component series, a spike component series, and an error component series. The decomposition involves configuring an optimization algorithm with a constraint indicating that the time series is a sum of these latent components. The decomposition also involves executing the optimization algorithm to minimize an objective function subject to the constraint and identifying, from the executed optimization algorithm, the seasonal component series, the level component series, the spike component series, and the error component series that minimize the objective function. The analytical system outputs at least some latent components for anomaly-detection or data-forecasting.

Abstract: Determination of high value customer journey sequences is performed by determining customer interactions that are most frequent as length N=1 sub-sequences, recursively determining most frequent length N+1 sub-sequences that start with the length N sub-sequences, determining a first count indicating how often one of the sub-sequences appears in the sequences, determining a second count indicating how often the one sub-sequence resulted in the goal, and using the counts to determine the most or least effective sub-sequences for achieving the goal.

Abstract: A resource control system is described that is configured to control scheduling of executable jobs by compute instances of a service provider system. In one example, the resource control system outputs a deployment user interface to obtain job information. Upon receipt of the job information, the resource control system communicates with a service provider system to obtain logs from compute instances implemented by the service provider system for the respective executable jobs. The resource control system uses data obtained from the logs to estimate utility indicating status of respective executable jobs and an amount of time to complete the executable jobs by respective compute instances. The resource control system then employs a machine-learning module to generate an action to be performed by compute instances for respective executable jobs.

Abstract: This disclosure relates to methods, non-transitory computer readable media, and systems that asynchronously train a machine learning model across client devices that implement local versions of the model while preserving client data privacy. To train the model across devices, in some embodiments, the disclosed systems send global parameters for a global machine learning model from a server device to client devices. A subset of the client devices uses local machine learning models corresponding to the global model and client training data to modify the global parameters. Based on those modifications, the subset of client devices sends modified parameter indicators to the server device for the server device to use in adjusting the global parameters. By utilizing the modified parameter indicators (and not client training data), in certain implementations, the disclosed systems accurately train a machine learning model without exposing training data from the client device.

Abstract: Techniques are provided for scheduling multiple jobs on one or more cloud computing instances, which provide the ability to select a job for execution from among a plurality of jobs, and to further select a designated instance from among a plurality of cloud computing instances for executing the selected job. The job and the designated instance are each selected based on a probability distribution that a cost of executing the job on the designated instance does not exceed the budget. The probability distribution is based on several factors including a cost of prior executions of other jobs on the designated instance and a utility function that represents a value associated with a progress of each job. By scheduling select jobs on discounted cloud computing instances, the aggregate utility of the jobs can be maximized or otherwise improved for a given budget.

Abstract: Determination of high value customer journey sequences is performed by determining customer interactions that are most frequent as length N=1 sub-sequences, recursively determining most frequent length N+1 sub-sequences that start with the length N sub-sequences, determining a first count indicating how often one of the sub-sequences appears in the sequences, determining a second count indicating how often the one sub-sequence resulted in the goal, and using the counts to determine the most or least effective sub-sequences for achieving the goal.

Abstract: Systems and techniques are described herein for updating a machine learning model on edge servers. Local parameters of the machine learning model are updated at a plurality of edge servers using fresh data on the edge servers, rather than waiting for the data to reach a global server to update the machine learning model. Hence, latency is significantly reduced, making the systems and techniques described herein suitable for real-time services that support streaming data. Moreover, by updating global parameters of the machine learning model at a global server in a deterministic manner based on parameter updates from the edge servers, rather than by including randomization steps, global parameters of the converge quickly to their optimal values. The global parameters are sent from the global server to the plurality of edge servers at each iteration, thereby synchronizing the machine learning model on the edge servers.

Abstract: Graphing services are provided to a device cooperative that includes data contributors, e.g., website hosts. Anonymized user data, provided by the data contributors, is accessed, via a blockchain, decrypted, and aggregated. A device graph is generated based on the aggregated user data. Contribution metrics are provided to the data contributors. A first contribution metric for a first data contributor indicates a contribution to the device graph of a first portion of the user data that was provided by the first data contributor. In response to receiving a request for a verification of the first contribution metric, a zero knowledge proof of the first contribution metric is generated and provided to the first data contributor. The first data contributor is enabled to evaluate the zero knowledge proof independent of access to a second portion of the user data that was provided by a second data contributor of the device cooperative.

Abstract: The present disclosure describes systems, non-transitory computer-readable media, and methods for determining latent components of a metrics time series and identifying anomalous data within the metrics time series based on one or both of spikes/dips and level changes from the latent components satisfying significance thresholds. To identify such latent components, in some cases, the disclosed systems account for a range of value types by intelligently subjecting real values to a latent-component constraint for decomposing the time series and intelligently excluding non-real values from the latent-component constraint. The disclosed systems can further identify significant anomalous data values from latent components of the metrics time series by jointly determining whether one or both of a subseries of a spike-component series and a level change from a level-component series satisfy significance thresholds.