Adobe, Inc.

Ankur Garg

Sunav Choudhary

Manoj Ghuhan A

Saurabh Kumar Mishra

Iftikhar Ahamath Burhanuddin

Payal Bajaj

Sweta Agrawal

Kritin Kesav Sai Sathi

Kirnesh Nandan

Aditya Prakash

Shubham Agrawal

Abhishek Kedia

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.

Asynchronously training machine learning models across client devices for adaptive intelligence

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.

Updating machine learning models on edge servers

In some embodiments, a computing system determines, based on stored user information retrieved from a mobile user device and associated with a particular user, a location-specific affinity of the particular user for a product at a particular geographical location. The location-specific affinity indicates an interest of the particular user in the product that increases when the particular user is positioned at the particular geographical location. The computing system designs a geo-fence targeted to the particular user based on the location-specific affinity, where messages are transmitted to the mobile user device if the particular user is within a boundary of the geo-fence. The geo-fence defines a geographical area that includes the particular geographical location and that is associated with a provider of the product. The computing system causes a telecommunication server to transmit the message to the user device when the user device is positioned within the designed geo-fence.

Smart geo-fencing using location sensitive product affinity

In implementations of a recommendation system based on individualized privacy settings, a computing device maintains user profiles of information and recommendations associated with users of the recommendation system. The computing device includes a recommendation module that is implemented to receive a privacy level selection for a type of items corresponding to a user profile in the system. The recommendation module can determine a privacy setting for a user associated with the user profile, where the privacy setting is individualized for the user in context of the type of items with an algorithmic noise function utilized to obfuscate a proportional level of the information associated with the user and the type of items based on the received privacy level selection. The recommendation module can also generate recommendations of relevant items for the user based on the determined privacy setting as individualized for the user in context of the type of items.

Recommendation system based on individualized privacy settings

Growing community of inventors

Chandigarh, India

Ankur Garg