In-Vehicle Data Selection for Low-Confidence Feature Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing volume of sensor data from advanced vehicle sensors for autonomous driving poses a challenge in efficiently processing and transmitting data for feature detection, leading to burdens on computing resources and high costs for data transmission and storage.
Innovation Solution
An in-vehicle feature detection device determines the confidence metric of detected features and transmits sensor data to an external server only when the confidence metric is below a threshold, allowing for selective data transmission and re-training of feature detection models externally, thereby reducing unnecessary data transmission and storage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all sensor data is transmitted to external servers for processing, then feature detection model accuracy can be improved through comprehensive training data, but data transmission costs and computing resource burdens increase significantly
Solution Approach 1:
The patent extracts only the necessary subset of sensor data (cases with low confidence predictions) for external transmission and model retraining, rather than transmitting all sensor data. This selective extraction reduces data transmission costs while still providing sufficient training data to improve model accuracy.
Solution Approach 2:
The patent applies different processing quality levels to different data subsets: high-confidence predictions are processed locally and discarded, while low-confidence predictions are extracted for external retraining. This local quality differentiation optimizes the balance between model improvement and transmission cost.
2Loss of energy
If sensor data is transmitted only when confidence metric is below threshold, then data transmission costs are reduced, but the quantity of training data available for model improvement decreases
Solution Approach 1:
The patent changes the parameter of data selection from random or uniform sampling to confidence-based filtering. By using the confidence metric as a selection parameter, the system identifies and transmits only the most informative data samples (those with low confidence predictions), maximizing training value while minimizing transmission volume.
3Measurement precision
If feature detection models are retrained externally using selected sensor data, then model accuracy improves over time, but processing time and computational resources are consumed
Solution Approach 1:
The patent performs preliminary filtering of sensor data at the vehicle端 before transmission, identifying only low-confidence cases that require retraining. This preliminary action reduces the volume of data requiring external processing, thereby reducing model retraining time while still achieving accuracy improvements.
Solution Approach 2:
The patent applies partial retraining using only the subset of data that is most beneficial (low-confidence cases) rather than retraining on all available data. This partial action approach achieves sufficient model improvement with reduced computational time and resources.
Data Source
AI summary
An approach is provided for selecting training observations for machine learning models. The approach involves determining a first distribution of a plurality of features observed in the training data set, and a second distribution of the plurality of features observed in the candidate pool of observations. The approach further involves selecting one or more observations in the candidate pool of observations for annotation based on the first distribution and the second distribution. The approach further involves adding the one or more observations to the training data set after annotation. The training data set is used for training the machine learning model.