Fan Gao, Cheng Huang, Yutong Liu, Nyima Tashi, Xiangxiang Wang, Thupten Tsering, Ban Ma-bao, Renzeng Duojie, Gadeng Luosang, Rinchen Dongrub, Dorje Tashi, Xiao Feng Cd, Yongbin Yu, Hao Wang. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Language models based on deep neural networks are vulnerable to textual adversarial attacks. While rich-resource languages like English are receiving focused attention, Tibetan, a cross-border language, is gradually being studied due to its abundant ancient literature and critical language strategy. Currently, there are several Tibetan adversarial text generation methods, but they do not fully consider the textual features of Tibetan script and overestimate the quality of generated adversarial texts. To address this issue, we propose a novel Tibetan adversarial text generation method called TSCheater, which considers the characteristic of Tibetan encoding and the feature that visually similar syllables have similar semantics. This method can also be transferred to other abugidas, such as Devanagari script. We utilize a self-constructed Tibetan syllable visual similarity database called TSVSDB to generate substitution candidates and adopt a greedy algorithm-based scoring mechanism to determine substitution order. After that, we conduct the method on eight victim language models. Experimentally, TSCheater outperforms existing methods in attack effectiveness, perturbation magnitude, semantic similarity, visual similarity, and human acceptance. Finally, we construct the first Tibetan adversarial robustness evaluation benchmark called AdvTS, which is generated by existing methods and proofread by humans.

Road extraction for the high-resolution remote sensing images has important applications in urban planning, traffic management, geographic information systems(GIS) and other fields. However, The impact of the environment such as pedestrians and farms causes discontinuity and incompleteness in extraction results. In this paper, We present D-Linknet with transformer and Concurrent Spatial and Channel Squeeze and Channel Excitation(scSE) module(TSDlinknet) which is built with D-Linknet architecture. Firstly, we use transformer encoders in the central part instead of D-Block to handle long-range dependencies efficiently. Secondly, scSE module is added after each convolutional layer in the decoding stage to refine spatial information and channel information of the feature map. Experimental results on CHN6-CUG road dataset demonstrate that our method outperforms D-Linknet101 by 5.6% IOU higher, in addition, our extraction results are more continuous and complete.

Semantic segmentation of high-resolution remote sensing (HRRS) images is an important task in the field of remote sensing image analysis. However, the presence of a large number of complex ground objects in HRRS images poses challenges for its semantic segmentation. In this paper, we propose a high-order transformer semantic segmentation network (HOT-Net) for HRRS images. The network uses ResNet-50 as backbone to capture local features at the encoder stage. In order to expand the receptive field of the network and enhance its ability to perceive global contextual information, several proposed high-order transformer blocks are uesd to establish global dependencies in local features. Global enhancement attention modules (GEAM) are used to enhance the representation of global features during the decoder stage. We conducte ablation and comparative experiments on the LoveDA dataset, and the experimental results show that our method has excellent performance compared to other popular methods.

Online action detection aims to identify ongoing actions within untrimmed video streams, with extensive applications in real-life scenarios. However, in practical applications, video frames are received sequentially over time and new action categories continually emerge, giving rise to the challenge of catastrophic forgetting - a problem that remains inadequately explored. Generally, in the field of video understanding, researchers address catastrophic forgetting through class-incremental learning. Nevertheless, online action detection is based solely on historical observations, thus demanding higher temporal modeling capabilities for class-incremental learning methods. In this paper, we conceptualize this task as Class-Incremental Online Action Detection (CIOAD) and propose a novel framework, TS-ILM, to address it. Specifically, TS-ILM consists of two components: task-level temporal pattern extractor and temporal-sensitive exemplar selector. The former extracts the temporal patterns of actions in different tasks and saves them, allowing the data to be comprehensively observed on a temporal level before it is input into the backbone. The latter selects a set of frames with the highest causal relevance and minimum information redundancy for subsequent replay, enabling the model to learn the temporal information of previous tasks more effectively. We benchmark our approach against SoTA class-incremental learning methods applied in the image and video domains on THUMOS'14 and TVSeries datasets. Our method outperforms the previous approaches.

Tibetan-Chinese machine translation has become a focal point of interest within the Tibetan community due to its importance for effective communication and cultural preservation. Although technological advancements have been made, current Tibetan-Chinese translation systems still fall short of satisfactory performance. Moreover, evaluating these systems using a publicly available and reliable benchmark dataset remains problematic. To address these challenges, we propose an enhanced TibetanChinese machine translation framework. This framework includes data filtering techniques and utilizes the semantic knowledge of the state-of-the-art Chinese cross-lingual model, CINO. After preprocessing, we trained our model using the Transformer architecture and found that our approach significantly improves translation performance. Additionally, we have developed a highquality, expert-reviewed test dataset to support community evaluations of translation systems. The details of our experimental model and the test dataset can be accessed at11https://github.com/UTibetNLP/Ti-ChTrans.

In the domain of remote sensing (RS) image analysis, capturing global context is the key for precise semantic segmentation. Current vision transformer (ViT) advance this field by addressing convolutional neural network’s (CNN) local receptive field limitations. However, ViT predominantly rely on the first-order information in image to establish global relationships, often overlooking the potential of second-order information, which is crucial for enhancing the discrimination of ground objects that exhibit high similarity and constant changes. To address this issue, we propose a global adaptive second-order transformer network (GASOT-Net). Specifically, the proposed global adaptive second-order transformer (GASOT) enhances the existing ViT structure by mining second-order information and adaptively fusing it with the first-order information during the process of establishing global dependency relationships. This approach enables the extraction of more discriminative features, thereby enriching the representation of global features. In addition, the local feature aggregation module (LFAM) is proposed to effectively aggregate features from different stages of CNN as input to the GASOT blocks. Moreover, to refine boundaries of complex ground objects, the global feature enhancement module (GFEM) is used in the decoder stage. In particular, GFEM includes two sub modules—feature shift module (FSM) and hierarchical feature fusion module (HFFM). FSM is used to enhance the local feature representation at first, and then, HFFM hierarchically aggregates local and global features from different stages. We conduct extensive experiments on four benchmark RS datasets, and the results show that our GASOT-Net outperforms other state-of-the-art methods. The code will be available at:https://github.com/j136812832/GASOT-Net.

This article is concerned with the exponential synchronization of coupled memristive neural networks (CMNNs) with multiple mismatched parameters and topology-based probability impulsive mechanism (TPIM) on time scales. To begin with, a novel model is designed by taking into account three types of mismatched parameters, including: 1) mismatched dimensions; 2) mismatched connection weights; and 3) mismatched time-varying delays. Then, the method of auxiliary-state variables is adopted to deal with the novel model, which implies that the presented novel model can not only use any isolated system (regard as a node) in the coupled system to synchronize the states of CMNNs but also can use an external node, that is, not affiliated to the coupled system to synchronize the states of CMNNs. Moreover, the TPIM is first proposed to efficiently schedule information transmission over the network, possibly subject to a series of nonideal factors. The novel control protocol is more robust against these nonideal factors than the traditional impulsive control mechanism. By means of the Lyapunov-Krasovskii functional, robust analysis approach, and some inequality processing techniques, exponential synchronization conditions unifying the continuous-time and discrete-time systems are derived on the framework of time scales. Finally, a numerical example is provided to illustrate the effectiveness of the main results.

This article investigates the problem of relaxed exponential stabilization for coupled memristive neural networks (CMNNs) with connection fault and multiple delays via an optimized elastic event-triggered mechanism (OEEM). The connection fault of the two or some nodes can result in the connection fault of other nodes and cause iterative faults in the CMNNs. Therefore, the method of backup resources is considered to improve the fault-tolerant capability and survivability of the CMNNs. In order to improve the robustness of the event-triggered mechanism and enhance the ability of the event-triggered mechanism to process noise signals, the time-varying bounded noise threshold matrices, time-varying decreased exponential threshold functions, and adaptive functions are simultaneously introduced to design the OEEM. In addition, the appropriate Lyapunov-Krasovskii functionals (LKFs) with some improved delay-product-type terms are constructed, and the relaxed exponential stabilization and globally uniformly ultimately bounded (GUUB) conditions are derived for the CMNNs with connection fault and multiple delays by means of some inequality processing techniques. Finally, two numerical examples are provided to illustrate the effectiveness of the results.

This article addresses the problem of dynamic pinning synchronization of fuzzy-dependent-switched (Fds) coupled memristive neural networks (CMNNs) with mismatched dimensions on time scales. To begin with, the probabilistic coupling delays, time scales, mismatched dimensions, and function projective synchronization rules are considered to design the novel CMNNs to improve the reliability and generalization ability of the model. Then Fds rules and dynamic pinning control (DPC) method are adopted to design the CMNNs, which can effectively promote the information exchange between the switching signals and the fuzzy processes and can improve the utilization of the communication bandwidth between the nodes of CMNNs. Meanwhile, the method of constructing auxiliary state variables is adopted here to deal with the presented model, so that the coupled and isolated systems with different dimensions can realize information exchange and data sharing. This method also provides a solution for researchers by using low-dimensional systems to estimate or synchronize high-dimensional systems. Moreover, by means of Lyapunov–Krasovskii functional, auxiliary orthogonal matrix, and some inequality processing techniques, the conditions of modified function projective synchronization for Fds CMNNs are derived via the DPC on time scales. Finally, two numerical examples are provided to illustrate the effectiveness of the main results.

This article investigates the heterogeneous impulsive synchronization for T-S fuzzy probabilistic coupled delayed neural networks (CDNNs) with mode-dependent parameters and piecewise membership functions. To begin with, a novel CDNNs model with adjustable coupling strength and probabilistic coupling delays is designed to ensure the accuracy of the CDNNs model. Meanwhile, the generalized isolated node with four types of mismatched parameters, named heterogeneous isolated delayed neural network, is first considered to extend the synchronization problem. Then, the mode-dependent fuzzy rules are introduced to design the novel model, which implies that switching signals and fuzzy processes are interdependent and can share information to communicate. To improve the hybrid controller’s reliability, the mode-dependent impulses are also developed here, in which the impulsive effects with different properties can occur at any moment in the switching interval. The exponential synchronization conditions are derived by means of the method of auxiliary state variables, Lyapunov–Krasovskii functional, average switching dwell period, and mode-dependent average impulsive dwell period. Moreover, the improved mode-dependent piecewise approximated membership functions are proposed to reduce the main results’ conservatism. Finally, a numerical example is provided to illustrate the effectiveness of the main results.

This article investigates the problem of robust exponential stability of fuzzy switched memristive inertial neural networks (FSMINNs) with time-varying delays on mode-dependent destabilizing impulsive control protocol. The memristive model presented here is treated as a switched system rather than employing the theory of differential inclusion and set-value map. To optimize the robust exponentially stable process and reduce the cost of time, hybrid mode-dependent destabilizing impulsive and adaptive feedback controllers are simultaneously applied to stabilize FSMINNs. In the new model, the multiple impulsive effects exist between two switched modes, and the multiple switched effects may also occur between two impulsive instants. Based on switched analysis techniques, the Takagi-Sugeno (T-S) fuzzy method, and the average dwell time, extended robust exponential stability conditions are derived. Finally, simulation is provided to illustrate the effectiveness of the results.