Edge Proxy Head for Rare Data Mining in Autonomous Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data mining techniques for autonomous systems face challenges such as high computational demands, cost-prohibitive solutions, and labor-intensive manual interactions, making it difficult to efficiently identify and collect rare data cases for training machine learning models.
Innovation Solution
Implementing a data source proxy head on an edge platform within autonomous vehicles, which utilizes existing machine learning model infrastructure to selectively collect data for mining purposes, reducing computational demands and leveraging existing computations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing data mining techniques are used for autonomous systems, then data can be collected for training, but computational demands become excessively high and costs become prohibitive
Solution Approach 1:
The patent creates a simplified copy of the training pipeline that runs on the edge device. Instead of performing complex data mining computations centrally, a lightweight version of the model and data mining logic is deployed to the autonomous vehicle, enabling local data collection and filtering without high computational demands on central servers
Solution Approach 2:
The patent extracts the data mining functionality from the central training system and places it directly on the edge device. By taking out the data collection and filtering operations from the centralized architecture, the system eliminates the need for high computational resources at the central server while maintaining data mining effectiveness
2Reliability
If comprehensive data collection is performed to improve training data quality, then rare data cases can be identified, but computational loads increase significantly
Solution Approach 1:
The patent implements partial action by collecting only the specific portion of data that is most valuable for training - rare and edge cases. Rather than collecting all possible data comprehensively, the system selectively collects only what is necessary, reducing computational load while maintaining training data quality
Solution Approach 2:
A lightweight copy of the data mining system is deployed on the edge device, enabling local identification of rare data cases without requiring comprehensive centralized analysis. This copying approach allows the system to perform selective data collection with reduced computational complexity
3Measurement precision
If manual interactions are used for data mining, then data can be curated, but the process becomes labor-intensive and less efficient
Solution Approach 1:
The system enables self-service data mining by automatically collecting, filtering, and identifying rare data cases using the deployed model on the edge device. The autonomous vehicle performs data curation itself without requiring manual human intervention, maintaining high data quality while dramatically improving mining efficiency
Solution Approach 2:
The automated data mining system is copied to the edge device, replacing manual data curation processes with autonomous automated operations. This allows the system to maintain the precision of manual curation while achieving the efficiency of automated processing
Data Source
AI summary
Disclosed are embodiments for facilitating data mining on an edge platform using repurposed neural network models in autonomous systems. In some aspects, an embodiment includes receiving, by a processing device hosting a data source proxy head of a machine learning (ML) model deployed on an autonomous vehicle (AV), a set of features selected from raw data by a backbone network of the ML model; utilizing, by the data source proxy head, the set of features selected from the raw data as input data to a trained data source mining model of the data source proxy head; identifying, by the trained data source mining model based on the input data, a portion of the raw data to classify as mining data; and providing, by the data source proxy head, identification of the portion of the raw data as a data mining output.