The emergence of quantum computing presents significant challenges for intelligent transportation systems (ITS), which rely on traditional public-key cryptography, demanding both strong security and low-latency operation. This highlights an urgent requirement for post-quantum cryptographic (PQC) algorithms to guarantee long-term protection. However, a direct integration of PQC into ITS may introduce latency and inefficiency, as static implementations cannot adapt to dynamic traffic conditions and may apply excessive security in low-risk contexts, resulting in unnecessary computational overhead, particularly within resource-constrained edge computing environments. To mitigate these issues, the PDRL-PQCS (Predictive Deep Reinforcement Learning for Post-Quantum Cryptography Selection) framework is proposed. This approach employs a Transformer-based module to generate multi-step predictions of traffic flow and computational load, thereby providing proactive contextual awareness. These predictions subsequently inform a deep reinforcement learning (DRL) agent, which dynamically selects optimal parameter configurations for the CRYSTALS-Kyber PQC algorithm in real time, enabling adaptive security-performance balancing across the heterogeneous nodes of an ITS edge computing network. Through this two-stage approach, PDRL-PQCS adapts PQC parameters within defined security bounds to optimally balance security assurance with operational efficiency. An extensive experimental evaluation assesses the framework under three representative traffic scenarios with diverse resource availability and congestion levels. Results indicate that PDRL-PQCS achieves 10%–20% higher overall performance compared to baseline methods across various challenging conditions.

Compute eXpress Link (CXL) memory disaggregation technology has emerged as a prominent solution within the realm of memory decoupling, characterized by its low latency, supporting for memory semantics and hardware-level cache coherency. It aims to break through the limitations of traditional CPU-centric architectures, driving data centers towards a more flexible and composable architecture to accommodate the rapid development of advanced heterogeneous computing applications, such as high-performance computing and large-scale Artificial Intelligence (AI) models. This survey reviews the development of CXL memory disaggregation technology across three generations of CXL protocols, associated with the evolution of representative functionalities and key concepts in CXL specifications. Meanwhile, the fundamental performance and heterogeneous interleaving capabilities of CXL memory on the latest Intel Xeon platform are disclosed and analyzed in this paper based on practical tests. Furthermore, the paper conducts a comprehensive assessment of the current software and hardware ecosystem surrounding CXL memory disaggregation, delineating its advantageous application scenarios and the technological trends in adjacent fields, as informed by collective academic and industrial experiences. Last, this paper concludes the opportunities and challenges for CXL memory disaggregation technology catering to emerging applications like AI from a commercialization standpoint, denoting its potential in the design of next-generation intelligent computing centers.

Exploring potential disease risk factors is critical for precision medicine, enabling the identification of biomark-ers and supporting accurate diagnostic tools and interventions. However, traditional statistical and machine learning methods often fail to reduce disease risk effectively, as they focus on correlations while ignoring essential causal relationships. Causality, on the other hand, could identify causal rela-tionships and predict effects from large-scale observational data, discovering underlying data generation mechanisms. Therefore, we propose a novel Causality Driven Analysis Framework (CDAF) to identify causal risk factors from obser-vational data. CDAF employs a converging causality growth algorithm to discover essential causal relationships through Bayesian scoring criterion, modeled as a causal graph. It estimates the average treatment effects (ATE) of direct causal edges using causal inference, ranking the importance of risk factors. CDAF demonstrated excellent performance on both benchmark datasets and other real clinical datasets. We em-pirically demonstrate that our framework offers a powerful tool for uncovering complex disease mechanisms, advanc-ing precision medicine, and enabling personalized treatment strategies.

As Artificial Intelligence (AI) demonstrates an impressive performance across various tasks, its ability to process diverse data types has been widely adopted. However, AI models often lack transparency and interpretability, making it challenging to identify the most effective data categories and models for specific contexts. Current strategies typically rely on common sense or exhaustive searches, both of which are inefficient and costly. To fill this gap, we propose the INdicator-assisted FramewOrk (INFO). INFO enables researchers and practitioners to directly identify effective data categories and guide model selection for tasks involving multiple data categories. By assessing the inherent signal strength of the data, INFO provides a proxy for evaluating task performance, thereby reducing the need for time-consuming model trials and reproducibility efforts. We validate the effectiveness of INFO using real-world datasets across five different domains, focusing on the critical task of user churn detection as the case study. Our experiments show that INFO can reveal the efficacy of data categories before large-scale trials. INFO shows promise as an automated data engineering methodology that clarifies the relationship between model performance and dataset characteristics.

Large language models (LLMs) have demon-strated excellent performance in complex tasks through Chain-of-Thought (CoT) reasoning. However, LLMs usually gener-ate long thinking content during reasoning, leading to high inference latency. Speculative decoding (SD) is a general acceleration method but has two limitations that weaken its acceleration effect in CoT reasoning scenarios. Therefore, in this study, we proposed the aggressive speculative decoding (ASD) method for efficient CoT reasoning, alleviating the two shortcomings. First, SD adopts a strict acceptance strategy to ensure the generated results are consistent with the target model. However, multiple feasible reasoning trajectories may lead to correct answers. Therefore, ASD adopts an aggressive acceptance strategy, which accepts more draft tokens based on the deviation between the probabilities. Second, SD only generates a small fixed number of draft tokens in each drafting stage. In contrast, ASD adopts an aggressive drafting strat-egy, which adaptively generates a suitable number of draft tokens by estimating the expected transformation value of the inference latency through confidence. Extensive experiments on different reasoning datasets and LLMs demonstrate the effectiveness of our method. In particular, ASD achieved a significantly better acceleration effect than SD on all datasets while ensuring accuracy.

In recent years, drone-based wireless data collection has emerged as a significant research hotspot in the Internet of Things (IoT) field. However, most existing studies primarily focus on dedicated drones, which incur high deployment costs. In contrast, reusing existing delivery drones provides a cost-effective alternative. This paper concerns the problem of scheduling delivery drones during their delivery to enable cost-efficient wireless data collection. Specifically, the problem aims to maximize the data collection amount while subject to the constraints of the drone’s battery capacity and package delivery deadlines. Nevertheless, solving this problem is challenging due to the involvement of three coupled variables: collection task allocation, collection time distribution, and flying path planning. To tackle this challenge, we propose MdSche, a multi-delivery drone scheduling scheme for wireless data collection. Specifically, we first design a time slicing-based data collection time discretization algorithm, which transforms the original problem into a Multi-Drone Scheduling Problem (MDSP) involving two variables. Then, an auxiliary graph construction method is employed to decompose MDSP into multiple independent single-drone scheduling subproblems. Finally, we construct a surrogate function to simplify each subproblem into a collection task allocation problem with only one variable. Theoretical analysis verifies that MdSche achieves an approximation ratio of 1/2(1 − (1/e)^1/4 ) in polynomial time. Extensive trace-based simulations show that, compared to five baseline methods, our approach improves the data collection amount by an average of 69.68%.

Single-source Domain Generalization (SDG) is a promising yet challenging technology that aims to transfer knowledge from a singular source domain to multiple and unfamiliar target domains. Existing SDG methods typically rely on domain expansion to implement data variation and broaden the coverage of the training domain. However, due to the lack of proper semantic consistency and sample diversity constraints, these methods have limited improvement in generalization performance for most practical applications. In this paper, we propose a Causality-Aware Single-source Domain Generalization (CASDG) method to utilize both semantic consistency and diversity during the data transformation process. First, a causality-aware module is designed to accurately measure the causal effect between latent features and labels. Then, we introduce a causal domain expansion module, which utilizes the causal effect matrix as a semantic consistency constraint and mutual information as a sample diversity constraint. These two constraints are jointly used to encourage the style transformer to generate new auxiliary samples that are undeviated from the original samples. The image classification model using our method can produce the best classification performance for unknown domain data compared to the state-of-the-art methods.