The rapid advancement of pre-trained language models (PLMs) has demonstrated promising results for various code-related tasks. However, their effectiveness in detecting real-world vulnerabilities remains a critical challenge. While existing empirical studies evaluate PLMs for vulnerability detection (VD), they suffer from data leakage, limited scope, and superficial analysis, hindering the accuracy and comprehensiveness of evaluations. This paper begins by revisiting the common issues in existing research on PLMs for VD through the evaluation pipeline. It then proceeds with an accurate and extensive evaluation of 18 PLMs, spanning model parameters from millions to billions, on high-quality datasets that feature accurate labeling, diverse vulnerability types, and various projects. Specifically, we compare the performance of PLMs under both fine-tuning and prompt engineering, assess their effectiveness and generalizability across various training and testing settings, and analyze their robustness to perturbations such as code normalization, abstraction, and semantic-preserving transformations.

Federated learning (FL), an emerging distributed machine learning paradigm, has been applied to various privacy-preserving scenarios. However, due to its distributed nature, FL faces two key issues: the non-independent and identical distribution (non-IID) of user data and vulnerability to Byzantine threats. To address these challenges, in this paper, we propose FedCAP, a robust FL framework against both data heterogeneity and Byzantine attacks. The core of FedCAP is a model update calibration mechanism to help a server capture the differences in the direction and magnitude of model updates among clients. Furthermore, we design a customized model aggregation rule that facilitates collaborative training among similar clients while accelerating the model deterioration of malicious clients. With a Euclidean norm-based anomaly detection mechanism, the server can quickly identify and permanently remove malicious clients. Moreover, the impact of data heterogeneity and Byzantine attacks can be further mitigated through personalization on the client side. We conduct extensive experiments, comparing multiple state-of-the-art baselines, to demonstrate that FedCAP performs well in several non-IID settings and shows strong robustness under a series of poisoning attacks.

This article presents a novel neural network-based hybrid mode-switching control strategy, which successfully stabilizes the flapping wing aerial vehicle (FWAV) to the desired 3-D position. First, a novel description for the dynamics, resolved in the proposed vertical frame, is proposed to facilitate further position loop controller design. Then, a radial base function neural network (RBFNN)-based adaptive control strategy is proposed, which employs a switching strategy to keep the system away from dangerous flight conditions and achieve efficient flight. The learning process of the neural network pauses, resumes, or alternates its update strategy when switching between different modes. Moreover, saturation functions and barrier Lyapunov functions (BLFs) are introduced to constrain the lateral velocity within proper ranges. The closed-loop system is theoretically guaranteed to be semiglobally uniformly ultimately bounded with arbitrarily small bound, based on Lyapunov techniques and hybrid system analysis. Finally, experimental results demonstrate the excellent reliability and efficiency of the proposed controller. Compared to existing works, the innovations are the put forward of the vertical frame and the cooperative switching learning and control strategies.

Both airborne and mobile LiDAR are suitable for the fast information acquisition of large-scale urban buildings, which respectively express the rooftop and the facade information of buildings. The fusion of airborne and mobile LiDAR can provide a relatively complete building point cloud, which is beneficial to constructing precise building models. Due to the different data acquisition methods, airborne LiDAR can get the relatively complete point cloud data of the building rooftop, while the building facade point cloud from mobile LiDAR is often incomplete or even missing because of the tree occlusion or the limitation of operating conditions, which to some extent limits the application of the fused data. To solve the problems above, a method of building point cloud similarity matching based on airborne LiDAR and mobile LiDAR data fusion is proposed. Finally a test result is provided to illustrate the proposed algorithm.