AI/ML Smart Gateway With Local Video Model Tuning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network systems lack the ability to provide privacy and security for local video analysis by relying on centralized AI/ML models that may not be specific to individual user needs, leading to potential false positives and negatives in object detection and requiring off-premises data transmission.
Innovation Solution
Deploying Local Modeling Gateways (LMGs) with AI/ML processing units that generate locally tuned models based on shared AI/ML models and local video data, enabling accurate and secure detection of specific objects or individuals without transmitting video data off-premises.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If centralized AI/ML models are used for video analysis, then model updates and improvements can be managed centrally, but privacy and security are compromised due to off-premises data transmission
Solution Approach 1:
The system segments the AI/ML model management into two parts: centralized model generation and distribution (performed by the network gateway) and local model execution (performed at the edge device). This segmentation allows centralized control over model updates while keeping video data processing local, thus maintaining privacy and security.
Solution Approach 2:
The network gateway acts as an intermediary that generates AI/ML models centrally based on video data characteristics, then distributes these models to edge devices. The gateway processes the video data locally to understand its characteristics without transmitting the actual video content, serving as a mediator between centralized model management and local privacy-preserving execution.
2Ease of manufacture
If generic AI/ML models are deployed, then deployment is simplified, but detection accuracy decreases due to false positives and negatives
Solution Approach 1:
The system applies local quality by generating AI/ML models that are specifically tuned to local video data characteristics. The network gateway analyzes the characteristics of local video data (such as lighting conditions, object types, and environmental factors) and generates customized models that are optimized for those specific conditions, thereby improving detection accuracy while maintaining ease of deployment through automated generation.
3Power
If video data is transmitted off-premises for analysis, then centralized processing capabilities can be utilized, but network bandwidth is consumed and privacy is compromised
Solution Approach 1:
The system extracts only the essential characteristics and patterns from video data locally at the network gateway, then uses these extracted features to generate AI/ML models. Instead of transmitting the entire video data stream off-premises, only the generated models (which are much smaller in size) need to be transmitted to edge devices, significantly reducing network bandwidth consumption while retaining centralized processing capabilities.
Data Source
AI summary
A system described herein may receive a first set of artificial intelligence/machine learning (“AI/ML”) models; receive first locally captured video information; generate a second set of AI/ML models based on the first set of AI/ML models and the locally captured video information; receive second locally captured video information; identify, based on the second locally captured video information and the second set of AI/ML models, one or more classifications for the second locally captured video information; and output, via a network and to an action system, the one or more classifications, without outputting the second locally captured video information via the network, wherein the action system identifies a particular action associated with the one or more classifications, and performs the identified particular action.