In official promotional videos and media press, the malicious cut-out editing of content to misrepresent the speaker’s intended meaning has long been recognized as a problem of misinformation/disinformation. This issue of selective cut-out editing has gained increasing attention, particularly in the context of the growing need for fact-checking mechanisms. Consequently, there is a strong demand for technical approaches to counteract propaganda strategies that exploit partial multimedia content extraction. However, cut-out editing is also an essential technique for compressing lengthy content into highlights or summaries, making outright regulation of such actions impractical. This study proposes a framework that applies cut-and-paste editing to official video releases, permitting clipping some segments and concatenating them while preventing malicious edits without undermining editorial discretion. By employing semantic analysis to identify potentially exploitable segments and their contextual or meaning-based relationships, the approach restricts intentional misrepresentative cut-and-paste editing. The resulting restricted version is then published as the original content, offering a defensive strategy against such manipulations.

Steganography, the art of information hiding, has continually evolved across visual, auditory and linguistic domains, adapting to the ceaseless interplay between steganographic concealment and steganalytic revelation. This study seeks to extend the horizons of what constitutes a viable steganographic medium by introducing a steganographic paradigm in robotic motion control. Based on the observation of the robot’s inherent sensitivity to changes in its environment, we propose a methodology to encode messages as environmental stimuli influencing the motions of the robotic agent and to decode messages from the resulting motion trajectory. The constraints of maximal robot integrity and minimal motion deviation are established as fundamental principles underlying secrecy. As a proof of concept, we conduct experiments in simulated environments across various manipulation tasks, incorporating robotic embodiments equipped with generalist multimodal policies.

Deep neural networks (DNNs) remain vulnerable to adversarial attacks that cause misclassification when specific perturbations are added to input images. This vulnerability also threatens the reliability of DNN-based monocular depth estimation (MDE) models, making robustness enhancement a critical need in practical applications. To validate the vulnerability of DNN-based MDE models, this study proposes a projection-based adversarial attack method that projects perturbation light onto a target object. The proposed method employs physics-in-the-loop (PITL) optimization—evaluating candidate solutions in actual environments to account for device specifications and disturbances—and utilizes a separable covariance matrix adaptation evolution strategy. Experiments confirmed that the proposed method successfully created adversarial examples that lead to depth misestimations, resulting in parts of objects disappearing from the target scene.

An interconnection network is an inevitable component in a parallel computer. It offers communication capabilities in the parallel machine, affecting performance issues of parallel computation. Thus, various discussions are being made from a variety of aspects to reduce the communication cost and to improve the performance. This paper addresses the packet scheduling problem, which is a promising method for improving the performance, in the collective communication. Our preceding work has proposed a unique optimization method Lopit (lazy optimization of packet injection timing). This paper extends the method by introducing a group nature in collective communication situations and proposes a new method G-Lopit (grouped Lopit). Evaluation results in our interconnection network simulator reveal the significant effectiveness of the proposed method. The G-Lopit method outperforms the traditional GA and the preceding Lopit methods. It improves the performance of collective communication at most 1.18 times from the Lopit method in 32×32 2D-torus network with bcmp traffic. In comparison with unoptimized situations, it achieves at most 1.73 times improvement in the shfl traffic.

As one of the most widely used programming languages in modern software development, Python hosts a vast open-source codebase on GitHub, where code reuse is widespread. This study leverages open-source Python projects from GitHub and applies automated testing to discover pairs of functionally equivalent methods. We collected and processed methods from 5,100 Python repositories, but Python’s lack of static type checking presented unique challenges for grouping these methods. To address this, we conducted detailed type inference and organized methods based on their inferred types, providing a structured foundation for subsequent analysis. We then employed automated test generation to produce unit tests for each method, running them against one another within their respective groups to identify candidate pairs that yielded identical outputs from the same inputs. Through manual verification, we ultimately identified 68 functionally equivalent method pairs and 683 functionally non-equivalent pairs. These pairs were compiled into a comprehensive dataset, serving as the basis for further examination. With this dataset, we not only evaluated the ability of large language models (LLMs) to recognize functional equivalence, evaluating both their accuracy and the challenges posed by diverse implementations, but also conducted a systematic performance evaluation of equivalent methods, measuring execution times and analyzing the underlying causes of efficiency differences. The findings demonstrate the potential of LLMs to identify functionally equivalent methods and highlight areas requiring further advancement.

Human activity recognition (HAR) is necessary for detection of unsafe activity in industrial production, but there are still some issues that need to be solved, such as limited data in different scenarios and the lack of a unified model for different situations. Therefore, a novel meta-federated learning framework with distillation of activation boundaries (AB) is proposed, in which a federation is viewed as a meta-distribution and all federations work together without a central server. Specifically, the personalized model from the previous federation serves as the teacher model for the next federation, where general knowledge is extracted by AB knowledge distillation, the personalized knowledge is acquired through local training, and a high-quality model is obtained for the current federation by dynamically fusing general knowledge and personalized knowledge. To evaluate the effectiveness and superiority of the proposed framework, experiments were conducted on one popular HAR datasets (PAMAP2) and a chemical scenario dataset (WACID) constructed by our laboratory. The experimental results show that our proposed framework outperforms the state-of-the-art methods with fewer communication costs, achieving the recognition accuracies of 91.23% and 95.66% on the PAMAP2 dataset and WACID dataset, respectively.

Recently, websites that enable users to share and search for cooking recipes have gained popularity. Each recipe typically includes various pieces of information, including a title, a list of ingredients, and detailed steps described in text and illustrated with photos. The estimated cooking time for each recipe is another valuable information when selecting a recipe. However, it can be difficult to accurately determine cooking time because it depends on various factors, such as heat level, ingredient quantity, and cooking skill level. Therefore, some recipes do not include information on cooking time. In this study, we consider the prediction of cooking time in general scenarios based on a list of ingredients and a textual description of each recipe’s cooking process using BERT, a natural language processing model. To this end, we propose an additional pre-training method that assigns greater weight to words related to cooking time using a cooking ontology. Our experimental results show that our method outperforms a fine-tuned BERT model with additional pre-training using a commonly employed approach. Notably, words representing “Kitchen Tools” are particularly associated with cooking time.

Multi-objective reinforcement learning (MORL) addresses sequential decision-making problems involving conflicting objectives. While most existing methods assume access to known or explicitly defined utility functions, many real-world tasks feature implicit, nonlinear utilities that are only available as delayed black-box feedback. To tackle this challenge, we propose Adaptive Multi-Objective Actor-Critic (AMOAC), a scalable framework that dynamically aligns policy optimization with implicit utility signals, without requiring prior knowledge of the utility function’s form. AMOAC employs a multi-critic architecture to maintain computational efficiency as the number of objectives grows, and introduces a dynamic direction-aligned weighting mechanism to guide policy updates toward utility maximization. Experiments on benchmark environments—including Deep Sea Treasure, Minecart, and Four Room—demonstrate that AMOAC consistently matches or exceeds the performance of baselines with explicit utility access, achieving robust adaptation and convergence under both linear and nonlinear utility scenarios. These results highlight the potential of dynamic weight adjustment in MORL for handling implicit preference structures and limited feedback settings.

The composition model in ROS 2 enables multiple nodes to run within a single process, reducing the overhead of inter-process communication (IPC). However, this architecture introduces memory safety and concurrency challenges due to a shared address space and a common Executor. Existing tracing tools lack the granularity to detect node-level runtime anomalies in such settings. We present CompShield, a system that extends ROS2Trace to enable intra-process node-level tracing and misbehavior detection. CompShield combines static analysis with enhanced runtime tracing to identify temporal anomalies and concurrency-related performance issues, such as Executor monopolization and prolonged callback execution. Evaluation on a ROS 2 composition application shows that CompShield effectively detects such issues with low overhead.

The proliferation of fake accounts in social networks has prompted growing attention to the development of effective detection techniques for ensuring cyberspace security. These fake accounts frequently employ sophisticated camouflage strategies to evade detection, which compromises the reliability of local neighborhood information. We propose GRFA, a novel approach for fake account detection that incorporates similarity-based adaptive graph reconstruction. The framework introduces a reinforcement learning-based adaptive mechanism to construct similarity edges, which dynamically refines the graph structure to better capture global dependencies. These refined structures are then incorporated into a heterogeneous graph neural network with dual aggregation, significantly improving the detection of camouflaged accounts. Experimental results demonstrate that GRFA outperforms state-of-the-art methods across multiple real-world datasets.

Unlike conventional speech-based depression detection (SDD), cross-elicitation SDD presents a more challenging task due to the differing speech elicitation conditions between the labeled source (training) and unlabeled target (testing) speech data. In such scenarios, a significant feature distribution gap may exist between the source and target speech samples, potentially reducing the detection performance of most existing SDD methods. To address this issue, we propose a novel deep transfer learning method called the Deep Elicitation-Adapted Neural Network (DEANN) in this letter. DEANN aims to directly learn both depression-discriminative and elicitation-invariant features from speech spectrograms corresponding to different elicitation conditions using two weight-shared Convolutional Neural Networks (CNNs). To achieve this, the CNNs are first endowed with depression-discriminative capability by establishing a relationship between the source speech samples and the provided depression labels. Subsequently, a well-designed constraint mechanism, termed Bidirectional Sparse Reconstruction, is introduced. This mechanism ensures that source and target speech samples can be sparsely reconstructed by each other at the same feature layer of both CNNs, allowing the learned features to maintain adaptability to changes in speech elicitation conditions while preserving their original depression-discriminative capability. To evaluate DEANN, we conduct extensive cross-elicitation SDD experiments on the MODMA dataset. The experimental results demonstrate the effectiveness and superiority of the proposed DEANN in addressing the challenge of cross-elicitation SDD compared to many existing state-of-the-art transfer learning methods.

Ellipse detection plays a critical role in fields such as medical diagnosis, environmental monitoring, and industrial automation. However, traditional methods (e.g., Hough transform, least-squares fitting, and edge-following techniques) suffer from high computational complexity and poor noise robustness. To address these limitations, we propose a hybrid framework that integrates deep learning with geometric constraints. First, Faster RCNN is employed to localize axis-aligned bounding boxes (AABBs) of ellipses. Then, a point-pair filtering strategy extracts edge points satisfying predefined geometric constraints, followed by weighted least-squares fitting to estimate ellipse parameters. Compared with traditional approaches, our method directly identifies AABBs, significantly enhancing both the efficiency and accuracy of multi-target ellipse detection in practice. Experiments are conducted on two synthetic datasets. The results show that our proposed method achieves superior precision and F-measure compared to conventional ellipse detection algorithms.

As a downstream task of visual entity and relationship extraction, human-object interaction detection focuses on complex relationships centered around humans as the primary subject. This has significant potential for application in some labour-intensive industries such as construction engineering. However, the data in these contexts often display a long-tailed distribution, featuring numerous unknown entities and relationships that are not present in standard datasets. This phenomenon places considerable demands on the model’s zero-shot learning capabilities. To tackle this challenge, this letter proposed an end-to-end human-object interaction detection method that utilized domain knowledge graph embeddings as part of prior queries for the decoders. In the case study, this method achieved a mean Average Precision (mAP) of 48.57% for the Full types across various scenarios. Specifically, the Rare types achieved a mAP of 52.45%, while the Non-Rare types achieved a mAP of 41.67%.