Ford global technologies, llc (20240202503). DATA DRIFT IDENTIFICATION FOR SENSOR SYSTEMS simplified abstract
Contents
DATA DRIFT IDENTIFICATION FOR SENSOR SYSTEMS
Organization Name
Inventor(s)
Sandhya Bhaskar of Sunnyvale CA (US)
Jinesh Jain of South San Francisco CA (US)
Nikita Jaipuria of Pittsburgh PA (US)
Shreyasha Paudel of Sunnyvale CA (US)
DATA DRIFT IDENTIFICATION FOR SENSOR SYSTEMS - A simplified explanation of the abstract
This abstract first appeared for US patent application 20240202503 titled 'DATA DRIFT IDENTIFICATION FOR SENSOR SYSTEMS
The patent application describes a system and method for identifying data drift in a trained object detection deep neural network.
- The system receives a dataset based on real-world use, containing scores for each class in an image, including a background class.
- It measures an intersection-over-union (IOU) conditioned expected calibration error (ECE) IOU-ECE by calculating ECE under a white-box setting with pre-nms detections conditioned on a specific IOU threshold.
- If the IOU-ECE is greater than a preset threshold, it performs white-box temperature scaling (WB-TS) calibration on the pre-nms detections to extract a temperature T.
- Data drift is identified if the temperature T exceeds a preset threshold.
Potential Applications: - Enhancing the accuracy and reliability of object detection systems. - Improving the performance of deep neural networks in real-world scenarios.
Problems Solved: - Detecting data drift in trained object detection DNNs. - Ensuring the consistency and reliability of object detection systems over time.
Benefits: - Early detection of data drift to prevent inaccuracies in object detection. - Improved performance and reliability of deep neural networks in real-world applications.
Commercial Applications: - This technology can be used in industries such as autonomous vehicles, surveillance systems, and medical imaging for accurate object detection and classification.
Questions about the technology: 1. How does the system measure data drift in a trained object detection DNN? 2. What are the potential implications of data drift in real-world applications?
Original Abstract Submitted
a system and method to identify a data drift in a trained object detection deep neural network (dnn) includes receiving a dataset based on real world use, wherein the dataset includes scores associated with each class in an image, including a background (bg) class, measuring an intersection-over-union (iou) conditioned expected calibration error (ece) iou-ece by calculating an ece under a white-box setting with detections from the dataset prior to non-maximum suppression (pre-nms detections) that are conditioned on a specific iou threshold, upon a determination of the iou-ece being greater than a preset first threshold, performing a white-box temperature scaling (wb-ts) calibration on the pre-nms detections of the dataset to extract a temperature t, and identifying that the data drift has occurred upon a determination that temperature t exceeds a preset second threshold.
