Lujia Yang, Weicai Yan, Yongbo He, Qifei Zhang, Tao Jin, Jinshan Zhang, Meng Xi, Jianwei Yin. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

The application of Radio Frequency (RF) sensors in human-centric perception has attracted considerable attention. Human pose estimation (HPE) using millimeter-wave (mmWave) radar holds promise for use in private and sensitive environments, such as medical monitoring and domestic rehabilitation assessment. Recent methods that utilize mmWave point clouds for pose estimation rely on raw 3D coordinate input data, failing to effectively utilize key velocity information. Additionally, existing research primarily focuses on reflection point clouds from a single person in a fixed position, making it difficult to apply in practical multi-person environments with interference. In this paper, we introduce the first multi-person mmWave radar point cloud dataset (mmVPose) focused on rehabilitation training. To effectively leverage velocity information for enriching point cloud features, we propose a velocity-dependent heatmap generation method, which represents input point clouds and predicted joint distributions as heatmaps in 3D voxel space. Furthermore, to preserve the complete latent human body structure while reducing computational complexity, we design a sparse attention mechanism for voxelized point clouds. Extensive experiments demonstrate that our method mmVPE achieves state-of-the-art performance in multi-person scenarios.

ABSTRACT Objective Collaborative edge computing (CEC) addresses the service quality issues that arise from the limited resources of a single node in traditional edge computing architectures by integrating resources from multiple edge nodes. However, ensuring reliable task offloading in this collaborative environment remains a significant challenge. Existing solutions often struggle to balance the intelligence and trustworthiness of offloading decisions effectively. This imbalance can lead to poor performance and reduced task success rates, especially if tasks are offloaded to malicious nodes. Methods To tackle these challenges, this paper proposes a trust‐enabled decentralized task offloading scheme that combines blockchain technology and deep reinforcement learning (DRL). First, we introduce a blockchain‐based reputation mechanism within the CEC architecture to facilitate trusted collaboration among nodes, utilizing smart contracts for reputation management. Next, we propose a beta distribution‐based three‐factor reputation update (BTRU) algorithm to enhance the accuracy of reputation evaluation. Finally, we present a decentralized and trust‐enabled task offloading (DTTO) algorithm based on DRL, which uses on‐chain reputation data to guide agents in learning trustworthy task offloading policies, thereby maximizing offloading trustworthiness and task success rates. Result To thoroughly assess the effectiveness and practicality of our proposed scheme, we develop a testbed for CEC task offloading based on Kubernetes and Ethereum. Experimental results demonstrate that the BTRU algorithm effectively distinguishes malicious nodes, reducing their average reputation by 97.54%, with an improvement of 9.94% compared to competitive algorithms. Meanwhile, the DTTO algorithm significantly enhances the efficiency and reliability of task offloading, raising the task success rate by at least 3.04%, especially when the proportion of malicious nodes reaches 40%, its task success rate is at least 5.41% higher than that of competitive algorithms. Conclusion The proposed trust‐enabled decentralized task offloading scheme successfully combines blockchain‐based reputation management with DRL to achieve both intelligent and trustworthy task offloading in the CEC environments. The experimental validation confirms the scheme's effectiveness in identifying malicious nodes and improving task success rates under various system conditions.

Current methods for scene text detection face four primary challenges: deficiencies of the anchor box approach in capturing text shapes, limited modeling capabilities for arbitrary-shaped text, high computational complexity in method representation, and significant post-processing overhead in detection algorithms. Addressing these challenges, we integrate computational geometry to introduce an efficient, interpretable, and shape-capable arbitrary-shaped text instance representation method, AFRG. This method adapts curve fitting through a selection of adaptive point sets based on geometric medians, leading to the development of a novel anchor-free module suitable for arbitrary-shaped text detection, APG-ATD, applied in scene text detection contexts. Furthermore, we propose an algorithm named MSEAP, which employs multidimensional supervision and evaluation of the adaptive point set across the dimensions of positioning, shape, spatial distribution, and classification to learn higher quality adaptive point sets. Building on the APG-ATD module, we further design an arbitrary-shaped text detection algorithm, MRAPG-ATD, based on multidimensional supervision and evaluation of adaptive points to enhance detection performance.

Image restoration under adverse weather conditions has been extensively explored, leading to numerous high-performance methods. In particular, recent advances in All-in-One approaches have shown impressive results by training on multi-task image restoration datasets. However, most of these methods rely on dedicated network modules or parameters for each specific degradation type, resulting in a significant parameter overhead. Moreover, the relatedness across different restoration tasks is often overlooked. In light of these issues, we propose a parameter-efficient All-in-One image restoration framework that leverages task-aware enhanced prompts to tackle various adverse weather degradations. Specifically, we adopt a two-stage training paradigm consisting of a pretraining phase and a prompt-tuning phase to mitigate parameter conflicts across tasks. We first employ supervised learning to acquire general restoration knowledge, and then adapt the model to handle specific degradation via trainable soft prompts. Crucially, we enhance these task-specific prompts in a task-aware manner. We apply low-rank decomposition to these prompts to capture both task-general and task-specific characteristics, and impose contrastive constraints to better align them with the actual inter-task relatedness. These enhanced prompts not only improve the parameter efficiency of the restoration model but also enable more accurate task modeling, as evidenced by t-SNE analysis. Experimental results on different restoration tasks demonstrate that the proposed method achieves superior performance with only 2.75M parameters.

With frequent software-induced activations on DRAM rows, bit flips can occur on their physically adjacent rows (i.e., RowHammer). Existing studies leverage FPGA platforms to characterize RowHammer, which have identified key factors that contribute to RowHammer bit flips, e.g., data pattern. As the FPGA-based studies have removed the interference of the OS and the memory controller, their findings on the identified contributing factors do not always work as reported in a real-world computing system, resulting in negative effects on system-level RowHammer attacks and defenses. In this paper, we carry out a system-level empirical study on factors from both the software side and the DRAM side that contribute to RowHammer. We conduct the study on 33 DRAM modules including both DDR4 and DDR3, with 292 DRAM chips from various vendors. Our experimental results from the software side show that some prior findings about existing factors are inconsistent with our observations, thus not applicable to a real-world system. Also, we contribute to identifying one new factor that effectively affects RowHammer bit flips. Our DRAM-side results identify three types of new contributing factors and indicate that DRAM modules are more vulnerable if they achieve better performance and lower power consumption. Particularly, Intel XMP, intended for improving DRAM performance, might be abused for RowHammer attacks.

The goal of few-shot learning is to classify new classes of samples with a few labeled training samples. State-of-the-art few-shot learners train a backbone on sufficient datasets and use its extracted features for classification. However, biased data distributions can lead to severe overfitting in few-shot learning. In this paper, we propose a novel feature transformation method that utilizes the statistical characteristics of sufficient data to perform feature transformation on few-shot data to alleviate overfitting caused by biased data distributions. We show an interesting phenomenon that removing the component along the mean feature of the base classes in meta-testing improves the performance for few-shot learning. Our method can be used on off-the-shelf pretrained feature extractors without extra parameters. We show that our method achieves the new state-of-the-art accuracy in the prototype-based method and comparable accuracy with state-of-the-art accuracy in the optimization-based method.

This paper presents a novel Convolutional Based Temporal Attention (CBTA) module that improves the performance of temporal convolutional networks (TCN) in lipreading tasks without requiring any additional data.Our CBTA method enhances recognition accuracy by focusing attention on relevant frames in the time sequence of a video.The study demonstrates how our strategy achieves groundbreaking sucacess on the Lip Reading in the Wild (LRW) dataset, achieving an accuracy of 92.61%-surpassing contemporary methods by approximately 0.5%.The experiment also indicates the broad utility and effectiveness of the adaptable CBTA.The proposed module substantially boosts Top-1 Accuracy for challenging words, offering a promising direction for overall performance improvement in lip-reading tasks.

Security and trust have become the key issues in the Internet of Things (IoT) environment. Characterized by the centralized control and high-energy consumption, the traditional trust management schemes are not suitable for the IoT systems, in which most of the interactions are short-duration, random and maybe one-time, and the terminal devices always have resource constraints. Therefore, this article proposes a distributed and two-layered trust management framework based on blockchain architecture for IoT. The hierarchical architecture of the cloud, the edge, the IoT subgroups, and devices solves the resource limitation problem and improves the privacy protection of the IoT applications. And a novel lightweight$Q$-learning improved DPoS consensus algorithm named QV-DPoS is proposed to solve the problems of large energy consumption and high complexity of consensus mechanism of blockchain. Ethereum is used to build a blockchain-based IoT prototype system, and some experiments were designed to verify whether the proposed platform can successfully conduct trust management and achieve identity and behavior authentication between the IoT entities. Moreover, the simulation experiments based on NetLogo is also designed to test the performance of the trust and consensus mechanisms. The results of the experiments show that the proposed mechanisms can effectively enhance the credibility of the interactions in the IoT environments, improve the transaction success rate, and reduce energy consumption at least 10% compared with the traditional algorithms.

Source-free domain adaptation (SFDA) aims at target adaptation without access to source data, but with only pre-trained source model. Some recent works proposed to automatically combine source models with learnable weights when there are multiple pre-trained source models. In this paper, we propose a simple yet effective framework for multi-source-free domain adaptation (MSFDA). Specifically, based on discriminability towards target samples, we determine transferability of source models before adaptation and generate pseudo-labels during training. To quantify target discriminability, we introduce net confidence which refers to the probability difference between the largest and the second largest probabilities. We empirically show, on several benchmark datasets, our proposed method is competitive to the state-of-the-art methods.

Unsupervised domain adaptation (UDA) methods usually assume data from multiple domains can be put together for centralized adaptation. Unfortunately, this assumption impairs data privacy, which leads to the failure of traditional methods in practical scenarios. To cope with the above issue, we present a novel decentralized domain adaptation approach which conducts target adaptation in an iterative training process during which only models can be delivered across domains. More specifically, to train a promising target model, we leverage Adversarial Examples (AEs) to filter out error prone predictions of source models towards each target sample based on both robustness and confidence, and then treat the most frequent prediction as the pseudo-label. Besides, to improve central model aggregation, we introduce Knowledge Contribution (KC) to compute reasonable aggregation weights. Extensive experiments conducted on several standard datasets verify the superiority of the proposed method.

Source-Free domain adaptive Semantic Segmentation (SFSS) aims to transfer knowledge from source domain to the target domain with only pre-trained source segmentation model and the unlabeled target dataset. Only a few works have been researched for SFSS, relying on entropy minimization, pseudo-labeling. Nevertheless, due to the domain bias, these methods tend to suffering from the confusion of classes with a similar visual appearance in different domains. To address the above issue, we propose to enhance discriminability towards target samples with masked image modeling to model spatial context relations as additional recognition clues. Specifically, we design a novel Dual-Level Masked Consistency method, which explicitly encourages the model to learn comprehensive context relations, i.e. patch-wise context and channel-wise context, on the target domain. By randomly masking target images and forcing the model to reconstruct predictions of the entire image with left unmasked part, the model has to make full use of spatially contextual information. To take a step further, we propose a novel masking strategy considering both local context and global context information by applying patch-wise masking on image patches and channel-wise masking on latent features. Notably, patch-wise context learning and channel-wise context learning can complement each other. Extensive experiments demonstrate the effectiveness of our proposed method and our method achieves state-of-the-art performance on two synthetic-to-real benchmarks: GTA5→Cityscapes and SYNTHIA→Cityscapes.

Summary With the problems of data sharing and information diddling in the field of education, we construct a highly available education consortium blockchain platform to ensure trusted sharing and privacy protection of education data. We employ erasure codes to process blockchain ledger files and optimize the data storage model according to the characteristics of education data, which can reduce the storage volume effectively. A HotStuff consensus algorithm is designed to access the ordering service of Hyperledger Fabric. A suitable educational blockchain network architecture based on the node complexity of education scenarios is proposed to achieve the high availability of the platform. To manage the education blockchain network, we implement the Fabric deployment based on Kubernetes and achieve the goal of including chaincode into Kubernetes environmental management. To improve the resource utilization of chaincode, we explore the new way of chaincode management by the functional computing service. Finally, on the premise of ensuring a 1/2 fault tolerance rate, the total ledger has decreased by 53.56%. Our platform enhanced the Byzantine fault tolerance while ensuring higher efficiency. Experimental results show that our platform is quite suitable for education scenario with many nodes.

With the maturity of Internet and rapid development of the Mobile Internet, the network protocols drafted by IETF, 3GPP, 3GPP2 and other standard organizations grow massively, and at the same time attacks against the network protocols are also increasing rapidly. The Robustness test for the network protocols is becoming increasingly important, since the protocols and applications will be absolutely safe if the protocols have been tested according to the requirements of the Robustness Test. But Robustness Test is different from Conformance Test, Interoperability Test and Performance Test in its requirement of a huge number of test cases, and the number of test cases is more than 2^320 for a packet with 40-Byte protocol header, thus, it's too difficult to generate all the test cases for the serial algorithm. In this paper a parallel algorithm based on MapReduce is proposed, where an original input packet is divided into different fields and the Map function processes on each single field and the Reduce function reassembles the processed fields to a new completed packet. Finally, experiment results show the parallel algorithm is far better than the serial algorithm when generating large data set. In addition, the scalability of the cluster is also verified.

In the existing large-scale performance test of IPsec tunnel, it often needs special software and hardware. To solve the problem, this article proposed a new method, in which packet was encapsulated in user space, and a multi-tunnel controller was designed and implemented with the method of FSM(finite state machine), which controlled the negotiation and establishment of multiple tunnel, including L2tp, IKEv1, IKEv2, IKEv2+EAP and L2tp Over IPsec. Libpcap was used as the bottom layer driver of package, and the application of zero copy technique had reduced system cost immensely. At last, the result of the experiment verified the performance of the IKEv1 tunnel on Tunnel-mode and Transport-mode.

Expectation and requirements for future wireless communication systems continue to grow and evolve. Thus, 3GPP has considered LTE/SAE to ensure its competitiveness in the future. In LTE/SAE, one of the recurring problems is dimensioning and testing. Modeling is an effective way to solve the problem because model is easy to generate test scenarios and inexpensive in changing test configurations and running test cases. This paper is to introduce how to build an accurate enough LTE/SAE model in NS2 so that other optimization features can be tested. This open source model includes traffic model and network model. The network model concentrates on the air interface and S1 interface. In the end of the paper, one example is given to demonstrate how to use the model.