Deep Neural Networks (DNNs) are increasingly deployed in safety-critical systems, yet their numerical reliability remains a pressing concern due to defects like NaN/INF occurrences during training/inference. Existing static analyzers detect operator-level risks but lack the ability to establish contextual mappings between operator-level anomalies and their originating code implementations, thereby limiting their practical applicability in real-world debugging scenarios. To address this limitation, this paper introduces a novel hybrid method that synergizes risk sensitive paths derived from static analysis with Large Language Model(LLM)-guided semantic reasoning. By anchoring LLMs to identified risk-sensitive computation paths, the proposed method precisely correlates operator-level numerical defects with code-level vulnerabilities, while mitigating LLM hallucinations. Evaluated on a widely-used real-world benchmark, the proposed method achieves $\mathbf{1 0 0} \boldsymbol{\%}$ defect detection accuracy, matching state-of-the-art static analyzers, but significantly outperforms them in code-level root-cause localization. This work pioneers a new paradigm for DNN reliability assurance by combining the static numerical analysis with the semantic reasoning power of LLMs, establishing a practical numerical defect analysis pipeline.

Graph data and graph neural networks (GNNs) have demonstrated their significance in a wide range of high-impact applications. However, GNNs are vulnerable to adversarial attacks and can be easily deceived by data perturbations, posing challenges for their application in safety- and security-critical tasks. Striving to improve the robustness of GNNs, we propose applying hardware-supported approximate computing (AxC) operators in GNN computation. Notably, the effectiveness of adversarial attacks on exact GNN models is difficult to transfer to their AxC counterparts, since approximation-induced noise can increase classification margins, thereby reducing misclassification rates across evasion attacks. As defensive AxC heightens computational robustness, it can be integrated with other defensive approaches focusing on graph topology recovery. Evaluation indicates that: (1) defensive AxC enhances the robustness of GNNs, leading to an average 2.5× increase in the fraction of correctly classified target nodes; (2) it can additionally strengthen robustness beyond a graph topology recovery approach, with significant increases in classification margins; (3) approximate GNN models achieve comparable classification accuracy without retraining and reduce hardware resource utilization and energy consumption. Deoxys is available at https://github.com/lydiawunan/Deoxys.

Emergency vehicles (EMVs) play a crucial role in responding to time-critical events such as medical emergencies and fire outbreaks in an urban area. The less time EMVs spend traveling through the traffic, the more likely it would help save people's lives and reduce property loss. To reduce the travel time of EMVs, prior work has used route optimization based on historical traffic-flow data and traffic signal pre-emption based on the optimal route. However, traffic signal pre-emption dynamically changes the traffic flow which, in turn, modifies the optimal route of an EMV. In addition, traffic signal pre-emption practices usually lead to significant disturbances in traffic flow and subsequently increase the travel time for non-EMVs. In this paper, we propose EMVLight, a decentralized reinforcement learning (RL) framework for simultaneous dynamic routing and traffic signal control. EMVLight extends Dijkstra's algorithm to efficiently update the optimal route for the EMVs in real-time as it travels through the traffic network. The decentralized RL agents learn network-level cooperative traffic signal phase strategies that not only reduce EMV travel time but also reduce the average travel time of non-EMVs in the network. This benefit has been demonstrated through comprehensive experiments with synthetic and real-world maps. These experiments show that EMVLight outperforms benchmark transportation engineering techniques and existing RL-based signal control methods.

The development of social media has made the dissemination of information more convenient, which means part of people can gather more information and then change their behaviors. In other words, information spreading would effect disease-control. To investigate the effect of information on vaccination game, an epidemic SEIR/VA (susceptible-exposed-infected-removed-vaccinated-avoided) model and a combined strategy-updating rule are proposed. Besides, temporary protective measures triggered by information are used to reflect the role of information dissemination. Based on the mean field approximation theory, differential equations are established to study the effect of information and vaccination coverage as well as vaccination efficacy under mandatory policy. Then, considering payoff and peer pressure, differential equations are developed to characterize the change of vaccination rate over time and the results of the final epidemic size as well as mean social payoff corresponding to different vaccine cost and efficiency. It can be observed that individuals with information obtained become more sensitive to the price and utility of vaccines. Scilicet, only when the vaccine is effective and cheap enough, information spreading can promote individual to take vaccination, which in turn eliminates the disease. Otherwise, as information spreads, people become more reluctant to get vaccinated. In other words, information dissemination is a double-edged sword.