Shared Cache Privacy With Background Model Loading

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing storage partitioning methods are inadequate for managing privacy in the cross-site sharing of large artificial intelligence (AI) models, leading to privacy risks and inefficiencies such as redundant downloads, computations, and latency issues.

Innovation Solution

Implementing a 'Background Model Load' mechanism that uses deduplication and emulates cache misses to ensure privacy by allowing model activation only when the requesting page is inactive, combined with user interface controls to manage model availability and hash-based keying to prevent information leakage.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If storage partitioning is used to manage privacy, then privacy protection is improved, but redundant downloads and computations occur leading to increased loss of time and energy

Engineering Contradiction:

Improveprivacy protectionVSAvoidredundant downloads and computations

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent segments the cache into per-origin caches while introducing a shared cache layer for AI models. This segmentation allows different origins to have separate caches for their resources while sharing the large AI model cache, thereby reducing redundant downloads and computations while maintaining privacy through origin-based separation.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent merges per-origin caches with a shared cache layer. The shared cache stores AI models that can be accessed by multiple origins, while per-origin caches store resources specific to each origin. This merging eliminates redundancy in AI model storage across origins while maintaining privacy through the per-origin cache layer.

Inventive Principle:

Principle #5Merging (Combining)

2Reliability

If storage partitioning is used to manage privacy, then privacy protection is improved, but latency increases due to inability to share resources

Engineering Contradiction:

Improveprivacy protectionVSAvoidlatency

Core Design Contradiction:

ReliabilityVSSpeed

Solution Approach 1:

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The system performs preliminary actions by pre-loading AI models into the shared cache before they are needed by multiple origins. This allows fast retrieval of AI models without requiring real-time downloads, reducing latency while maintaining privacy through the shared cache architecture.

Inventive Principle:

Principle #10Preliminary action

3Productivity

If AI models are shared across sites, then resource efficiency is improved, but privacy risks increase due to detectability of model presence

Engineering Contradiction:

Improveresource efficiencyVSAvoidprivacy risks

Core Design Contradiction:

ProductivityVSObject-affected harmful factors

Solution Approach 1:

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The shared cache acts as an intermediary between multiple origins and the AI model storage. It mediates the access to AI models by multiple origins while maintaining privacy through the per-origin cache layer that prevents direct detection of model presence by any single origin.

Inventive Principle:

Principle #24Intermediary (Mediator)

4Loss of energy

If cross-site cache sharing is implemented, then bandwidth savings are improved, but information leakage risks increase through timing attacks

Engineering Contradiction:

Improvebandwidth savingsVSAvoidinformation leakage through timing attacks

Core Design Contradiction:

Loss of energyVSLoss of information

Solution Approach 1:

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by making different parts of the caching system have different properties. Per-origin caches store resources specific to each origin with strict access control, while the shared cache layer stores AI models that can be accessed by multiple origins. This local differentiation ensures privacy for origin-specific resources while enabling efficient sharing of AI models.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS20250307334A1Methods, systems, articles of manufacture and apparatus to manage privacy with a shared cache

Publication Date: 2025.10.02 INTEL CORP

AI summary

This disclosure relates generally to shared caches and, more particularly, to methods, systems, articles of manufacture, and apparatus to manage privacy with a shared cache in the context of networked systems running untrusted code, such as web browsers. An example apparatus comprises machine-readable instructions, and at least one processor circuit to be programmed by the machine-readable instructions to determine if a machine learning model is located in a shared cache of a web browser, determine an activity state of a client application, and when the activity state of the client application is inactive and the machine learning model is located in the shared cache of the web browser, cause a simulated model download before the client application is notified of an availability of the machine learning model in the shared cache of the web browser.