Autonomous Data Collection for Under-Represented ML Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning model training methods often rely on biased data, neglecting under-represented data types, which can lead to error-prone models. Additionally, data collection is typically time-consuming and expensive, requiring human intervention.
Innovation Solution
A data collection device and method that automatically identifies and collects under-represented data for machine learning model training. This is achieved by executing a first machine learning model for lane detection, determining vehicle location using localization sensors, selecting lane geometry data from maps, generating offset values, and selecting interesting images based on these values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If biased training data is used for machine learning model training, then training efficiency is improved, but model accuracy and reliability deteriorate
Solution Approach 1:
The system implements feedback by using the trained machine learning model to analyze new data and automatically identify under-represented data types. The model's performance metrics and detection results feed back into the data collection process, guiding the automatic selection and collection of additional data samples that were previously neglected, thereby continuously improving model accuracy without manual intervention
Solution Approach 2:
The system enables self-service by automating the entire data collection process. The machine learning model independently identifies which data types are under-represented, automatically collects additional samples of those data types, and integrates them into the training set. This self-serving mechanism eliminates the need for human experts to manually curate balanced training data
2Measurement precision
If manual data collection methods are used, then data quality is improved, but time consumption and cost increase
Solution Approach 1:
The system replaces the mechanical manual data collection process with an automated computational system. Instead of human experts manually selecting and collecting data samples, the machine learning model automatically analyzes data distributions, identifies under-represented types, and collects additional samples programmatically, maintaining data quality while eliminating time consumption and human labor costs
Solution Approach 2:
The system introduces an intermediary automated data selection mechanism that acts as a bridge between raw data sources and the training process. This intermediary component uses the machine learning model to intelligently filter and select appropriate data samples, replacing direct manual intervention while preserving data quality through algorithmic selection criteria
Data Source
AI summary
In some embodiments, a method can include executing a first machine learning model to detect at least one lane in each image from a first set of images. The method can further include determining an estimate location of a vehicle for each image, based on localization data captured using at least one localization sensor disposed at the vehicle. The method can further include selecting lane geometry data for each image, from a map and based on the estimate location of the vehicle. The method can further include executing a localization model to generate a set of offset values for the first set of images based on the lane geometry data and the at least one lane in each image. The method can further include selecting a second set of images from the first set of images based on the set of offset values and a previously-determined offset threshold.