Salesforce Com, Inc.

Salesforce, Inc.

Tejaswini Ganapathi

Satish Raghunath

Shauli Gal

Kartikeya Chandrayana

Xu Che

Andrey Karapetov

An data driven approach to generating synthetic data matrices is presented. By retrieving historical network traffic data, probabilistic models are generated. Optimal distribution families for a set of independent data segments are determined. Applications are tested and performance metrics are determined based on the generated synthetic data matrices.

On demand synthetic data matrix generation

Network request data is collected over a time window. The network request data is filtered to generate bypass network traffic records. Network performance categories are generated from the bypass network traffic records. Sufficient statistics of network optimization parameters are calculated for the network performance categories. The sufficient statistics of the network optimization parameters are used to generate network optimization parameters to determine data download performances of web applications.

Estimation of network quality metrics from network request data

Bypass network traffic records are generated for a web application. Sufficient statistics of network optimization parameters are calculated for network performance categories. The bypass network traffic records are partitioned for the network performance categories into network traffic buckets. Sufficient statistics and the network traffic buckets are used to generate network quality mappings. The network quality mappings are used as training instances to train a machine learner for generating network optimization policies to be implemented by user devices.

Machine learning based end to end system for TCP optimization

Network traffic data associated with data requests to computer applications based on static policies is collected. An optimization order is established among network parameters. A first network parameter of a higher rank in the optimization order is estimated based on the collected network traffic data before one or more other network parameters of lower ranks are estimated. Optimal values for the other network parameters are estimated based at least in part on the estimated first optimal value for the first network parameter. The estimated first optimal value of the first network parameter and the estimated optimal values for the other network parameters are propagated to be used by user devices to make new data requests to the computer applications.

Simultaneous optimization of multiple TCP parameters to improve download outcomes for network-based mobile applications

Network requests are made to download a data object for a display page with different time delays. Page load outcomes of the display page are determined. A criticality of downloading the data object with respect to the display page is determined using page load outcomes. Criticalities of data objects of the display page are used to generate a specific data object download order that prioritizes critical and/or blocking objects of the display page.

Critical path estimation for accelerated and optimal loading of web pages

Machine learning based end to end system for tcp optimization

Network traffic data associated with computer applications is collected based on static policies. First network parameter vectors are generated over a time period. Each network parameter vector of the first network parameter vectors comprises first optimal values, estimated by a Bayesian learning module using a generative model, for network parameters. Second network parameter vectors are generated over the same time period. Each network parameter vector of the second network parameter vectors comprises second optimal values, computed by a best parameter generator through optimizing an objective function, for the network parameters. It is determined whether the first network parameter vectors converge to the second network parameter vectors and whether network parameter optimization for the network parameters is performing normally.

Automatic performance monitoring and health check of learning based wireless optimization framework

Network requests are made to download a data object with different settings of network parameters. Download outcomes of the data object as requested by the network requests are determined. An elasticity of downloading the data object is determined with respect to a specific network parameter in the network parameters. The elasticity is used to generate a network optimization policy that identifies an optimal value for the specific network parameter to be implemented by user devices and/or other devices/elements for downloading the data object.

Network request and file transfer prioritization based on traffic elasticity

Based on expert input comprising recommended parameter values, a machine learning framework is constrained to perform machine learning to estimate optimal parameter values for TCP parameters in specific regions of an output parameter space. Network traffic data associated with a plurality of data requests to one or more computer applications are collected, over a time block, based on sampled parameter values for the TCP parameters. The sampled parameter values are distributed within the specific regions of the output parameter space. The machine learning is used to estimate the optimal parameter values for the TCP parameters. The optimal parameter values, for the TCP parameters, are propagated to and used by user devices to make new data requests to the computer applications.

Incorporation of expert knowledge into machine learning based wireless optimization framework

An adaptive multi-phase approach to estimating network parameters is presented. By gathering and aggregating raw network traffic data and comparing against default network parameters, a training data set may be generated. A black box optimization may be used in tandem with a supervised learning algorithm to bias towards better choices and eventually pick network parameters which optimize performance. Data delivery strategies are applied to deliver content using the optimized network policies based on the estimated parameters.

Adaptive multi-phase network policy optimization

A data driven approach to emulating application performance is presented. By retrieving historical network traffic data, probabilistic models are generated to simulate wireless networks. Optimal distribution families for network values are determined. Performance data is captured from applications operating on simulated user devices operating on a virtual machine with a network simulator running sampled tuple values.

Growing community of inventors

San Francisco, CA, United States of America

Tejaswini Ganapathi