The widespread application of Distributed Acoustic Sensing (DAS) in security monitoring has increased the demand for advanced intrusion recognition algorithms. Conventional approaches often divide the process into two stages: signal processing and recognition modeling. However, this separation introduces two key challenges: (1) signal processing quality is typically assessed only through the final recognition outcome, lacking a unified optimization mechanism, which limits efficiency and generalization. (2) The two-stage framework is unsuitable for tasks requiring high real-time performance. Therefore, we propose the Frequency-Time Integrated Transformer (FTIformer) to address these challenges. FTIformer integrates the Multi-head Time-Frequency Perception (MTFP) module and the Self-Attention Module (SAM), achieving unified optimization of time-frequency signal extraction and pattern recognition through loss function backpropagation. Furthermore, in complex environments, multiple events of interest may occur within the same detection cycle, causing feature overlap and potential confusion of model outputs. To mitigate this, we decompose the multi-event recognition task into multiple single-event recognition tasks. Building on this approach, class-specific loss is proposed for the first time, which guides the model in learning unique representations for each event. We conducted experiments using a private dataset and a public dataset. Experimental results indicate that FTIformer achieves an accuracy, precision, and recall of 0.946, 0.976, and 0.979, respectively, on the private dataset, and 0.897, 0.956, and 0.966 on the public dataset. The detailed implementation of the proposed model is available at https://github.com/linjie1888/FTIformer-notebook.

Uncertainty estimation is essential for practical applications such as decision-making, risk assessment, and human-AI collaboration. However, Uncertainty estimation in open-ended question-answering (QA) tasks presents unique challenges. The output space for open-ended QA is vast and discrete, and the autoregressive nature of LLMs, combined with the rapid increase in model parameters, makes inference sampling significantly costly. An ideal uncertainty estimation for LLMs should meet two criteria: 1) incur no additional inference cost and 2) capture the semantic dependencies of token-level uncertainty within sequences. We propose a promising solution that converts redundancy into randomness in the extensive parameters of LLMs to quantify knowledge uncertainty. We can obtain token-level Monte Carlo samples without multiple inferences by introducing randomness during a single forward pass. We theoretically analyze the FLUE sampling method and employ a post-processing method to learn the state transitions from token uncertainty to sequence uncertainty. In open-ended question-answering tasks, we demonstrate that FLUE can achieve competitive performance in estimating the uncertainty of generated sentences without adding extra inference overhead.

Large Language Models (LLMs) have shown strong potential for recommendation by framing item prediction as a token-by-token language generation task. However, existing methods treat all item tokens equally, simply pursuing likelihood maximization during both optimization and decoding. This overlooks crucial token-level differences in decisiveness—many tokens contribute little to item discrimination yet can dominate optimization or decoding. To quantify token decisiveness, we propose a novel perspective that models item generation as a decision process, measuring token decisiveness by the Information Gain (IG) each token provides in reducing uncertainty about the generated item. Our empirical analysis reveals that most tokens have low IG but often correspond to high logits, disproportionately influencing training loss and decoding, which may impair model performance. Building on these insights, we introduce an Information Gain-based Decisiveness-aware Token handling (IGD) strategy that integrates token decisiveness into both tuning and decoding. Specifically, IGD downweights low-IG tokens during tuning and rebalances decoding to emphasize tokens with high IG. In this way, IGD moves beyond pure likelihood maximization, effectively prioritizing high-decisiveness tokens. Extensive experiments on four benchmark datasets with two LLM backbones demonstrate that IGD consistently improves recommendation accuracy, achieving significant gains on widely used ranking metrics compared to strong baselines. Our codes are available at \url{https://github.com/ZJLin2oo1/IGD}.

The existing research efforts in Multimodal Sentiment Analysis (MSA) have focused on developing the expressive ability of neural networks to fuse information from different modalities. However, these approaches lack a mechanism to understand the complex relations within and across different modalities, since some sentiments may be scattered in different modalities. To this end, in this paper, we propose a novel hierarchical graph contrastive learning (HGraph-CL) framework for MSA, aiming to explore the intricate relations of intra- and inter-modal representations for sentiment extraction. Specifically, regarding the intra-modal level, we build a unimodal graph for each modality representation to account for the modality-specific sentiment implications. Based on it, a graph contrastive learning strategy is adopted to explore the potential relations based on unimodal graph augmentations. Furthermore, we construct a multimodal graph of each instance based on the unimodal graphs to grasp the sentiment relations between different modalities. Then, in light of the multimodal augmentation graphs, a graph contrastive learning strategy over the inter-modal level is proposed to ulteriorly seek the possible graph structures for precisely learning sentiment relations. This essentially allows the framework to understand the appropriate graph structures for learning intricate relations among different modalities. Experimental results on two benchmark datasets show that the proposed framework outperforms the state-of-the-art baselines in MSA.

Stroke has become a leading cause of death and long-term disability in the world with no effective treatment. Deep learning-based approaches have the potential to outperform existing stroke risk prediction models, but they rely on large well-labeled data. Due to the strict privacy protection policy in health-care systems, stroke data is usually distributed among different hospitals in small pieces. In addition, the positive and negative instances of such data are extremely imbalanced. Transfer learning can solve small data issue by exploiting the knowledge of a correlated domain, especially when multiple source of data are available. In this work, we propose a novel Hybrid Deep Transfer Learning-based Stroke Risk Prediction (HDTL-SRP) scheme to exploit the knowledge structure from multiple correlated sources (i.e., external stroke data, chronic diseases data, such as hypertension and diabetes). The proposed framework has been extensively tested in synthetic and real-world scenarios, and it outperforms the state-of-the-art stroke risk prediction models. It also shows the potential of real-world deployment among multiple hospitals aided with 5 G/B5G infrastructures.