In-Browser Neural Network for Predictive Pre-Fetching

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing browser systems face delays and connectivity issues when retrieving web page content due to poor network quality and server overload, and they lack effective methods for predicting user behavior and optimizing content retrieval.

Innovation Solution

Implementing an in-browser neural network model that predicts user behavior and pre-fetches content before selection, reducing server workload and enhancing data security by training the model on client devices using user behavior data without transmitting it externally.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If content is retrieved from server in real-time, then data freshness is maintained, but network dependency increases and retrieval delays occur

Engineering Contradiction:

Improvecontent retrieval reliabilityVSAvoidcontent retrieval time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary actions by training the neural network model on client devices to predict user behavior and pre-fetching content before it is actually requested. This advance preparation reduces retrieval time when content is needed while maintaining reliability through local processing capabilities.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If server processes all neural network training, then model accuracy improves, but server workload increases and scalability decreases

Engineering Contradiction:

Improveprediction accuracyVSAvoidserver processing capacity

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The system segments the neural network training workload from the server and distributes it to individual client devices. Each device trains its own model locally using its own data, which maintains prediction accuracy through personalized learning while dramatically reducing server workload and improving scalability.

Inventive Principle:

Principle #1Segmentation

3Adaptability or versatility

If user behavior data is transmitted to server for training, then model personalization improves, but data security and privacy decrease

Engineering Contradiction:

Improvemodel personalizationVSAvoiddata privacy risk

Core Design Contradiction:

Adaptability or versatilityVSObject-affected harmful factors

Solution Approach 1:

Instead of transmitting user data to the server for training, the system inverts the approach by bringing the training process to the client device. The neural network model is trained locally on each user's device using their own behavior data, which maintains high personalization while eliminating data transmission risks and enhancing privacy security.

Inventive Principle:

Principle #13The other way round (Inversion)

Data Source

PatentUS10795965B2Distributed in-browser deep learning for predictive pre-fetching

Publication Date: 2020.10.06 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

Techniques for distributed processing and pre-fetching content using an in-browser neural network model are disclosed herein. In some embodiments, a server transmits a neural network model to a client device, where the neural network model is stored a persistent store of a browser on the client device, and, during a networking session in which the browser on the client device is accessing a page of an online service, the client device predicts at least one link from a plurality of links on the page using the stored neural network model. The client device then fetches content associated with the predicted link(s) from a server of the online service prior to any selection of the predicted link(s) during the networking session.