When we apply neural networks to safety-critical systems such as self-driving cars, the reliability of their predictions must be considered. However, recent deep neural networks have tended to output biased confidence. Additionally, the extent of confidence bias estimated by object detectors varies depending on factors such as the detected object’s position and size. To address this problem, many researchers have proposed methods for calibrating confidences estimated by object detectors. In this study, we investigate the factors that may cause bias in the confidence of LiDAR-based 3D object detectors and show that our calibration method compensates for the effect of these factors to provide reliable confidence estimations, regardless of the neural network model used or the situations in which objects are detected.

This paper deals with an automatic recovery mechanism designed to prevent the occurrence of subtract errors at the computer architecture level. This mechanism always detects the occurence of subtract errors whenever addition or subtraction is performed. If a subtract error is detected, backtracking to a certain location is carried out, and from there recomputing is performed in order to realize the required precision. An adaptive algorithm required for this mechanism is described in detail. An experimental adaptive computer was implemented on a Burroughs B1700. The evaluation of the experiments proves that the inherent possibilities of dynamic microprogramming play a very important role in realizing this mechanism, and that the proposed mechanism will be useful for the development of future high-level computer systems.