17928372. PRIVACY-ENHANCED TRAINING AND DEPLOYMENT OF MACHINE LEARNING MODELS USING CLIENT-SIDE AND SERVER-SIDE DATA simplified abstract (GOOGLE LLC)

PRIVACY-ENHANCED TRAINING AND DEPLOYMENT OF MACHINE LEARNING MODELS USING CLIENT-SIDE AND SERVER-SIDE DATA

Organization Name

GOOGLE LLC

Inventor(s)

Abhradeep Guha Thakurta of Los Gatos CA (US)

Li Zhang of Saratoga CA (US)

Prateek Jain of Bangalore (IN)

Shuang Song of Cupertino CA (US)

Steffen Rendle of Mountain View CA (US)

Steve Shaw-Tang Chien of San Carlos CA (US)

Walid Krichene of Fremont CA (US)

Yarong Mu of Kirkland WA (US)

PRIVACY-ENHANCED TRAINING AND DEPLOYMENT OF MACHINE LEARNING MODELS USING CLIENT-SIDE AND SERVER-SIDE DATA - A simplified explanation of the abstract

This abstract first appeared for US patent application 17928372 titled 'PRIVACY-ENHANCED TRAINING AND DEPLOYMENT OF MACHINE LEARNING MODELS USING CLIENT-SIDE AND SERVER-SIDE DATA

Simplified Explanation

The abstract describes a computer-implemented method for training a decentralized model for making personalized recommendations using user activity data.

• Obtaining client-side training data with features and labels from user activity data.
• Training a decentralized model on the client device in training rounds.
• Receiving current server-side embedding from the server in each training round.
• Generating client-side embedding based on client-side training data.
• Updating the client-side machine learning model using client-side and server-side embeddings and training labels.
• Transmitting updated client-side embedding for updating the server-side machine learning model.

---

1. 1. 1. Potential Applications

• Personalized recommendations in e-commerce platforms.
• Content recommendations in streaming services.
• Personalized advertising in digital marketing.

1. 1. 1. Problems Solved

• Providing personalized recommendations without compromising user data privacy.
• Efficient training of machine learning models on client devices.
• Improving recommendation accuracy by incorporating client-side data.

1. 1. 1. Benefits

• Enhanced user experience with personalized recommendations.
• Increased privacy protection for user data.
• Reduced server load and improved scalability for recommendation systems.

Original Abstract Submitted

Computer-implemented systems and methods for training a decentralized model for making a personalized recommendation. In one aspect, the method comprising: obtaining, using user activity data, client-side training data that includes features and training labels; and training, by the client device, a decentralized model in training rounds, wherein training, in each training round comprises: receiving, first data including a current server-side embedding generated by the server-side machine learning model, wherein the first data received from the server does not include any server-side data used in generating the current server-side embedding; generating, using the client-side machine learning model, a client-side embedding based on the client-side training data; updating, using the client-side embedding and the current server-side embedding and based on the training labels, the client-side machine learning model; generating, an updated client-side embedding; and transmitting second data including the updated client-side embedding for subsequent updating of the server-side machine learning model.