Yixuan Wang, Yue Huang, Hong Qian, Yunzhao Wei, Yifei Ding, Wenkai Wang, Zhi Liu, Zhongjing Huang, Aimin Zhou, Jiajun Guo. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Learner-item cognitive modeling plays a central role in the web-based online intelligent education system by enabling cognitive diagnosis (CD) across diverse online educational scenarios. Although ID embedding remains the mainstream approach in cognitive modeling due to its effectiveness and flexibility, recent advances in language models (LMs) have introduced new possibilities for incorporating rich semantic representations to enhance CD performance. This highlights the need for a comprehensive analysis of how LMs enhance embeddings through semantic integration across mainstream CD tasks. This paper identifies two key challenges in fully leveraging LMs in existing work: Misalignment between the training objectives of LMs and CD models creates a distribution gap in feature spaces; A unified framework is essential for integrating textual embeddings across varied CD tasks while preserving the strengths of existing cognitive modeling paradigms to ensure the robustness of embedding enhancement. To address these challenges, this paper introduces EduEmbed, a unified embedding enhancement framework that leverages fine-tuned LMs to enrich learner-item cognitive modeling across diverse CD tasks. EduEmbed operates in two stages. In the first stage, we fine-tune LMs based on role-specific representations and an interaction diagnoser to bridge the semantic gap of CD models. In the second stage, we employ a textual adapter to extract task-relevant semantics and integrate them with existing modeling paradigms to improve generalization. We evaluate the proposed framework on four CD tasks and computerized adaptive testing (CAT) task, achieving robust performance. Further analysis reveals the impact of semantic information across diverse tasks, offering key insights for future research on the application of LMs in CD for online intelligent education systems.

Localization of robots using subsurface features observed by ground-penetrating radar (GPR) enhances and adds robustness to common sensor modalities, as subsurface features are less affected by weather, seasons, and surface changes. We introduce an innovative multimodal odometry approach using inputs from GPR, an inertial measurement unit (IMU), and a wheel encoder. To efficiently address GPR signal noise, we introduce an advanced feature representation called the subsurface feature matrix (SFM). The SFM leverages frequency domain data and identifies peaks within radar scans. Additionally, we propose a novel feature matching method that estimates GPR displacement by aligning SFMs. The integrations from these three input sources are consolidated using a factor graph approach to achieve multimodal robot odometry. Our method has been developed and evaluated with the CMU-GPR public dataset, demonstrating improvements in accuracy and robustness with real-time performance in robotic odometry tasks.

For quantized deep neural networks widely deployed on hardware-accelerated platforms, bit-flip attacks (BFAs) have become a serious security threat because they can cripple or hijack model inference by modifying only a few bits. To this end, we propose KHAD (K-Hop and Activation-aware Defense), a unified, minimally intrusive defense framework that fuses k-hop propagation with activation statistics to precisely assess neuronal criticality and adaptively partitions neurons into three categories, to which it applies lightweight strategies: elastic boundary rectification, soft limiting with least significant bit masking, and range tightening with orthogonal diffusion, respectively. We conduct performance experiments across datasets of different scales and multiple models. The results show that, while exerting only a very small impact on clean accuracy (e.g., a decrease of 1.73% for ResNet-32 on CIFAR-100), KHAD exhibits strong defensive capability. Under untargeted attacks, it increases the minimum number of bit flips required to break the model by 5.6× to 14.3×; under targeted attacks, it effectively reduces the attack success rate (ASR) to 1.40%–15.60%. Moreover, the method can be fully deployed at model export time, yielding extremely low runtime overhead across CNN and Transformer architectures. These results indicate that, compared with global redundancy-based defenses, KHAD combines structural propagation properties with data-driven activity to precisely identify and block a small number of high-leverage propagation paths, trading a slight cost for substantial robustness gains.

Stealthy Bit-Flip Attacks, which manipulate DNN predictions by altering a few hardware-level weights, pose a severe threat to safety-critical applications. To address this threat, we propose the Dual-Mask Dynamic Defense (DMDD), whose core idea is to proactively and dynamically disrupt the static paths that attackers construct by exploiting model redundancy at inference time. DMDD’s defense is accomplished by two synergistic online mechanisms: 1) At the intermediate layers, Dual-mask Dynamic Neuron Pruning (DDNP) first employs an "Efficiency Mask" to prune task-irrelevant redundant neurons for efficiency gains, and then uses a "Security Mask" to sever potential attack paths by monitoring statistical anomalies in neuron behavior. 2) At the final layer, Dynamic Connection Purification (DCP) is activated to perform semantic-level filtering. To ensure the model can tolerate this online restructuring and the effects of residual attack paths, we also designed a complementary offline Pruning-aware Robustness Training (PART) strategy to enhance the model’s pruning tolerance and enforce larger inter-class distances. Unlike traditional defenses, DMDD transforms defense overhead into performance gains, achieving security enhancement with "negative computational overhead." Experiments show that DMDD can suppress the Attack Success Rate (ASR) of mainstream attacks like TBFA and TA-LBF from nearly 100% to below 15%, while simultaneously reducing the model’s computational load (FLOPs) by 26%-42%, providing an efficient and practical solution for secure DNN deployment in resource-constrained scenarios.

Robot localization using subsurface features captured by Ground-Penetrating Radar (GPR) complements and improves robustness over existing common sensor modalities, as subsurface features are less sensitive to weather, season and surface scene changes. Here, we propose a novel subsurface feature-based localization method that uses only GPR measurements with a known subsurface map. An efficient feature descriptor, the dominant energy curve (DEC), is designed to identify different locations in cluttered conditions. Specifically, image processing techniques that involve background segmentation, energy point detection, and energy curve refinement are designed to extract DEC features from a 2D radargram. With DECs features obtained, a metric subsurface feature map is constructed. Finally, we perform robot localization by feature matching under a particle swarm optimization framework. We have implemented our method and tested it with the public CMU-GPR dataset. The results show that our algorithm improves accuracy and robustness with real-time performance for robot localization tasks. Specifically, the mean localization error is 0.50 m for all cases.

Abstract Current skeleton‐based action recognition methods usually assume the input skeleton is complete and noise‐free. However, it is inevitable that the captured skeletons are incomplete due to occlusions or noisy due to changes in the environment. When dealing with these data, even State Of The Art (SOTA) recognition backbones experience significant degradation in recognition accuracy. Though a few methods have been proposed to address this issue, they still lack flexibility, efficiency and interpretability. In this work, an overcomplete Graph Convolutional Denoising Autoencoder (GCDAE) is proposed which can act as a flexible preprocessing module for pretrained recognition backbones and improve their robustness. Taking advantages of the overcomplete and fully graph convolutional structure, GCDAE is able to rectify noisy joints while keeping information of unspoiled details efficiently. On two large scale skeleton datasets NTU RGB+D 60 and 120, the introducing of GCDAE brings significant robustness improvements to SOTA backbones towards different types of noises.