The integration of Large Language Models (LLMs) into clinical decision support is critically obstructed by their opaque and often unreliable reasoning.In the high-stakes domain of healthcare, correct answers alone are insufficient; clinical practice demands full transparency to ensure patient safety and enable professional accountability.A pervasive and dangerous weakness of current LLMs is their tendency to produce "correct answers through flawed reasoning."This issue is far more than a minor academic flaw; such process errors signal a fundamental lack of robust understanding, making the model prone to broader hallucinations and unpredictable failures when faced with real-world clinical complexity.In this paper, we establish a framework for trustworthy clinical argumentation by adapting the Toulmin model to the diagnostic process.We propose a novel training pipeline: Curriculum Goal-Conditioned Learning (CGCL), designed to progressively train LLM to generate diagnostic arguments that explicitly follow this Toulmin structure.CGCL's progressive three-stage curriculum systematically builds a solid clinical argument: (1) extracting facts and generating differential diagnoses; (2) justifying a core hypothesis while rebutting alternatives; and (3) synthesizing the analysis into a final, qualified conclusion.We validate CGCL using T-Eval, a quantitative framework measuring the integrity of the diagnosis reasoning.Experiments show that our method achieves diagnostic accuracy and reasoning quality comparable to resourceintensive Reinforcement Learning (RL) methods, while offering a more stable and efficient training pipeline.1

In-band Network Telemetry (INT) has emerged as a promising solution for fine-grained, real-time monitoring in programmable data planes. However, existing INT approaches often incur excessive overhead due to per-hop metadata accumulation or lack fine-grained control over telemetry frequency. This paper presents TT-INT, a lightweight INT framework designed for P4-enabled networks, which introduces a time-threshold-based mechanism to regulate telemetry insertion dynamically. Each switch enforces local constraints based on per-flow time intervals and metadata capacity, enabling reduced overhead while preserving path visibility. Additionally, TT-INT supports a two-window byte-level anomaly detector and a controller-driven adjustment mechanism for further extensibility. Experiments on a real-worldderived backbone topology demonstrate that TT-INT reduces the average per-packet telemetry overhead to as low as 3.4 bytes under the 100 ms/5v configuration at 300 pps, achieving a 97.1% reduction compared to P4-INT under the same traffic rate. Compared to DLINT-5v and PLINT-5v (fixed at 20 and 26 bytes per packet, respectively), TT-INT-5v-100ms achieves up to 83.0% and 86.9% lower overhead. It also reaches a maximum path update detection rate of 97.9% (under the 50 ms configuration) and a minimum detection delay of 0.2 seconds, confirming TT-INT's effectiveness in balancing overhead, responsiveness, and monitoring fidelity under high-throughput conditions.

This paper proposes a novel parameter-efficient fine-tuning method that combines the knowledge completion capability of deconvolution with the subspace learning ability, reducing the number of parameters required for fine-tuning by 8 times . Experimental results demonstrate that our method achieves superior training efficiency and performance compared to existing models.

Existing visual token pruning methods target prompt alignment and visual preservation with static strategies, overlooking the varying relative importance of these objectives across tasks, which leads to inconsistent performance. To address this, we derive the first closed-form error bound for visual token pruning based on the Hausdorff distance, uniformly characterizing the contributions of both objectives. Moreover, leveraging $\epsilon$-covering theory, we reveal an intrinsic trade-off between these objectives and quantify their optimal attainment levels under a fixed budget. To practically handle this trade-off, we propose Multi-Objective Balanced Covering (MoB), which reformulates visual token pruning as a bi-objective covering problem. In this framework, the attainment trade-off reduces to budget allocation via greedy radius trading. MoB offers a provable performance bound and linear scalability with respect to the number of input visual tokens, enabling adaptation to challenging pruning scenarios. Extensive experiments show that MoB preserves 96.4\% of performance for LLaVA-1.5-7B using only 11.1\% of the original visual tokens and accelerates LLaVA-Next-7B by 1.3-1.5$\times$ with negligible performance loss. Additionally, evaluations on Qwen2-VL and Video-LLaVA confirm that MoB integrates seamlessly into advanced MLLMs and diverse vision-language tasks. The code will be made available soon.

To address the challenge of cyberattacks, intrusion detection systems (IDSs) are introduced to recognize intrusions and protect computer networks. Among all these IDSs, conventional machine learning methods rely on shallow learning and have unsatisfactory performance. Unlike machine learning methods, deep learning methods are the mainstream methods because of their capability to handle mass data without prior knowledge of specific domain expertise. Concerning deep learning, long short-term memory (LSTM) and temporal convolutional networks (TCNs) can be used to extract temporal features from different angles, while convolutional neural networks (CNNs) are valuable for learning spatial properties. Based on the above, this paper proposes a novel interlaced and spatiotemporal deep learning model called CRGT-SA, which combines CNN with gated TCN and recurrent neural network (RNN) modules to learn spatiotemporal properties, and imports the self-attention mechanism to select significant features. More specifically, our proposed model splits the feature extraction into multiple steps with a gradually increasing granularity, and executes each step with a combined CNN, LSTM, and gated TCN module. Our proposed CRGT-SA model is validated using the UNSW-NB15 dataset and is compared with other compelling techniques, including traditional machine learning and deep learning models as well as state-of-the-art deep learning models. According to the simulation results, our proposed model exhibits the highest accuracy and F1-score among all the compared methods. More specifically, our proposed model achieves 91.5% and 90.5% accuracy for binary and multi-class classifications respectively, and demonstrates its ability to protect the Internet from complicated cyberattacks. Moreover, we conduct another series of simulations on the NSL-KDD dataset; the simulation results of comparison with other models further prove the generalization ability of our proposed model.

Aspect-based sentiment analysis is the key to natural language processing, and it focuses on the polarity of emotions associated with specific text aspects. Traditional models that combine text and visual data tend to ignore the deeper interconnections between patterns. To solve this problem, the authors propose a multimodal sentiment-oriented analysis (BiCCM-ABSA) model based on bidirectional complementary correlation. The model utilizes text-image synergy through a novel cross-modal attention mechanism to align text with image features. With the transformer architecture, it is not only a simple fusion, but also ensures the complex alignment of multi-modal features and gating mechanisms. Experiments were conducted on the Twitter-15 and Twitter-17 datasets, achieving 69.28 accuracy and 67.54% F1 score, respectively. The experimental results demonstrate the advantages of BiCCM-ABSA, the bidirectional approach of the model and the effective cross-modal correlation set a new benchmark in the field of multimodal emotion recognition, providing insights beyond traditional single-modal analysis.

Recently, large language models (LLMs) with conversational-style interaction, such as ChatGPT and Claude, have gained significant importance in the advancement of artificial general intelligence (AGI). However, the extensive resource requirements during pre-training, instruction fine-tuning (IF), and reinforcement learning through human feedback (RLHF) pose challenges, particularly for individuals and studios with limited resources. Moreover, sensitive data that cannot be deployed on remote training platforms or queried through APIs further exacerbates this issue. To address these limitations, researchers have introduced a parameter-efficient framework called low-rank adaptation (LoRA) for IF on LLMs. However, training individual LoRA networks faces capacity constraints and struggles to adapt to large domains with significant distributional shifts across different tasks. In this paper, we propose a novel framework called chain-of-LoRA to enhance the IF performance of LoRA. Our approach involves training a LoRA network to classify the instruction type and then utilizing task-specific LoRA networks to accomplish the respective tasks. By training multiple task-specific LoRA networks, we exploit a trade-off between performance and disk storage, leveraging the easily expandable and cost-effective nature of disk storage compared to precious graphical resources. Our experimental results demonstrate that our proposed framework achieves comparable performance to typical direct IF on LLMs.

In the field of time series forecasting, time series are often considered as linear time-varying systems, which facilitates the analysis and modeling of time series from a structural state perspective. Due to the non-stationary nature and noise interference in real-world data, existing models struggle to predict long-term time series effectively. To address this issue, we propose a novel model that integrates the Kalman filter with a state space model (SSM) approach to enhance the accuracy of long-term time series forecasting. The Kalman filter requires recursive computation, whereas the SSM approach reformulates the Kalman filtering process into a convolutional form, simplifying training and enhancing model efficiency. Our Kalman-SSM model estimates the future state of dynamic systems for forecasting by utilizing a series of time series data containing noise. In real-world datasets, the Kalman-SSM has demonstrated competitive performance and satisfactory efficiency in comparison to state-of-the-art (SOTA) models.

Intelligent Cyber-Physical Systems relies heavily on data for real-time monitoring, analysis, and control of physical systems. By converting offline handwriting into online handwriting, it provides ICPS with more diverse and abundant input data, enriching the variety and quantity of available information. Based on the acquisition approach, there are two kinds of handwriting data: online and offline data. Generally, online data which contains pen trajectory of static character, has more advantages than offline data in terms of character recognition and analysis. Due to limited means of acquiring online data, inferring from offline data is an attractive approach. In this paper we introduce a novel framework to recover handwriting trajectory from single static character image. The trajectory can be represented by two types of sequence: points sequence and frame sequence. Therefore, we design two streams: points sequence prediction stream and frame sequence prediction stream, based on the encoder-decoder structure. We combine the two streams by a novel sequence consistent module to synchronize the training process. With the two streams and their complementary advantages, our methods can predict trajectory with both high spatial and temporal accuracy. Extensive experiments demonstrate the effectiveness of our network through qualitative and quantitative comparison.

Text-independent Chinese writer identification does not depend on the text content of the query and reference handwritings. In order to deal with the uncertainty of the text content, text-independent approaches usually give special attention to the global writing style of handwriting, rather than the properties of each individual character or word. Thanks to the existence of high-frequency characters, some characters probably appear in both the query and reference handwritings in most cases. If character images in the query handwriting are similar to those in the reference handwriting, this query handwriting and the corresponding reference handwriting are very likely to be written by the identical writer. In this paper, we exploit the above characteristic to improve the performance of Chinese writer identification. We first present an identification scheme using edge co-occurrence feature (ECF). Then, we detect the character pairs in the query and reference handwritings using a two-step framework and propose the displacement field-based similarity (DFS) to determine whether a character pair is written by the identical writer. The character pairs help to re-rank the candidate list obtained by text-independent ECF-based similarity and finally decide the writer of the query handwriting. The proposed method is evaluated on the HIT-MW and CASIA-2.1 datasets. Experimental results demonstrate that our proposed method outperforms the existing ones, and its Top-1 accuracy on the two datasets reaches 97.1% and 98.3%, respectively.

The key issue of Chinese writer identification is the uncertainty of the text content in the query and reference handwriting images. We propose a method for Chinese writer identification using Contour-directional Feature (CDF) and Character Pair Similarity Measurement (CPSM). CDFs are extracted from the query and reference handwriting images and are used to calculate the text-independent similarity between the query and reference handwriting images. Meanwhile, characters appearing in both the query and reference handwriting images are also utilized to measure the similarity of character pairs. The text-independent similarity and the similarity of character pairs are fused to the final similarity between the query and reference handwriting images. The proposed method is evaluated on two public datasets. The best Top-1 identification accuracy on the HIT-MW and CASIA-2.1 dataset reaches 96.7% and 97.9% respectively, which outperforms other previous approaches.

Writer recognition is to identify a person on the basis of handwriting, and great progress has been achieved in the past decades. In this paper, we concentrate ourselves on the issue of off-line text-independent writer recognition by summarizing the state of the art methods from the perspectives of feature extraction and classification. We also exhibit some public datasets and compare the performance of the existing prominent methods. The comparison demonstrates that the performance of the methods based on frequency domain features decreases seriously when the number of writers becomes larger, and that spatial distribution features are superior to both frequency domain features and shape features in capturing the individual traits.

This paper presents a method for text-independent writer identification using SIFT descriptor and contour-directional feature (CDF). The proposed method contains two stages. In the first stage, a codebook of local texture patterns is constructed by clustering a set of SIFT descriptors extracted from images. Using this codebook, the occurrence histograms are calculated to determine the similarities between different images. For each image, we obtain a candidate list of reference images. The next stage is to refine the candidate list using the contour-directional feature and SIFT descriptor. The proposed method is evaluated with two datasets: the ICFHR2012-Latin dataset and the ICDAR2013 dataset. Experimental results show that the proposed method outperforms the state-of-the-art algorithms and archives the best performance.