Learning with softmax cross-entropy on one-hot labels often leads to overconfidence on the correct class. While label smoothing regulates this overconfidence by redistributing some confidence from the correct class to other incorrect classes, it compromises the representation in the logits about the similarity between samples of different classes and may hurt calibration if higher confidence is required for high accuracy. To overcome these limitations, we propose a Virtual Smoothing (VS) label that redistributes certain confidence from the correct class to additional VS classes to regularize overconfidence. In VS labels, the VS class nodes act as adversaries to the original class nodes, enforcing regularization by clustering samples across all classes. The zero confidence assigned to each incorrect class also allows the incorrect logits to be different from each other without erasing information about sample similarities. The prediction probability can still approach 1 when applying softmax to the logits of the original real classes, which avoids harming but consistently improves calibration. Experiments show that VS labels consistently improve accuracy and calibration while providing better logits for improved knowledge distillation. Additionally, VS labels exhibit effectiveness in improving adversarial training, robust distillation, and out-of-distribution detection.

The application of machine learning methods to solve combinatorial problems has garnered considerable research interest. In this paper, we propose MAgg (**M**etamorphic **Agg**regation), a method to augment machine learning models for combinatorial problems at inference time using metamorphic relations. MAgg models metamorphic relations using directed graphs, which are then fed to a Graph Neural Network (GNN) model to improve the aggregation of predictions across transformed input instances. By incorporating metamorphic relations, MAgg essentially extends standard Test-Time Augmentation (TTA), eliminating the necessity of label-preserving transformations and expanding its applicability to a broader range of supervised learning tasks for combinatorial problems. We evaluate the proposed MAgg method on three mainstream machine learning tasks for combinatorial problems, namely Boolean Satisfiability Prediction (SAT), Decision Traveling Salesman Problem Satisfiability Prediction (Decision TSP), and Graph Edit Distance Estimation (GED). The evaluation result shows significant improvements over base models in all three tasks, corroborating the effectiveness and versatility of the proposed method.

One of the aims of the development and spread of autonomous driving technology is to reduce traffic accidents caused by human factors. But recently reported data on fatal accidents involving autonomous driving system (ADS) shows that this important goal has not been achieved. So there is an emerge requirement on more comprehensive and targeted testing especially on safe driving. In this paper, we propose an approach to automatically building critical testing scenarios from real-world accident data. Firstly, we propose a new model called M-CPS (Multi-channel Panoptic Segmentation) to extract the effective information from the accident record (such as images or videos), and separate the independent individuals of different traffic participants for further scene recovery. Compared with the traditional panoramic segmentation models, M-CPS model is able to effectively handle segmentation challenges due to the shooting angle, image quality, pixel overlap and other problems existing in the accident record. Next, the extracted core information is then connected with the virtual testing platform to generate the original scene set. Besides, we also design a mutation testing solution on the basis of the original scene set, thus greatly enriching the scene library for testing. In our experiments, the M-CPS model reaches a result of 66.1% PQ on CityScapes test set, shows that our model has only slight fluctuations on performance compared with the best benchmark model on pure panoptic segmentation task. It also reaches a result of 84.5% IoU for semantic segmentation branch and 40.3% mAP for instance segmentation branch on SHIFT dataset. Then we use UCF-Crime, CADP and US-Accidents datasets to generate the original and mutated scene set. Those generated scene sets are connected to Apollo and Carla simulation platforms to test ADS prototypes. We find three types of scenarios that can lead to accidents of ADS prototypes, which indicates that the existing ADS prototype has defects. Our solution provides a new possible direction for the recovery of key scenarios in ADS testing, and can improve the efficiency in related fields.