Accurate traffic flow prediction plays a critical role in intelligent transportation systems, providing essential support for urban planning, traffic control, and congestion mitigation. To address the challenges of spatial heterogeneity and temporal dynamics inherent in traffic data, this paper proposes AST-GNNFormer, an adaptive spatio-temporal graph neural network that integrates graph attention mechanisms with temporal convolution. The model introduces three key components to enhance predictive accuracy and generalization: (1) a Layer-aware Information Preservation mechanism that mitigates over-smoothing in deep GNNs by retaining original node features across layers; (2) an Inter-Layer At... More >

Poetry, as a profound and creative form of human expression, presents unique challenges in interpretation and summarization due to its reliance on figurative language, symbolism, and deeper meanings. Building upon the PoemSum dataset, which introduced the task of poem summarization, we extend its scope by exploring multimodal applications. Specifically, we implement and fine-tune two state-of-the-art abstractive summarization models—BART and T5—to generate concise and meaningful interpretations of poems, focusing on figurative summarization that captures metaphorical and symbolic elements inherent in poetic language. These summaries are then transformed into visual representations using... More >

Accurate segmentation of colorectal cancer (CRC) from endoscopic images is crucial for computer-aided diagnosis. Visual intelligence enhances detection precision, supporting clinical decision-making. However, current segmentation methods often struggle with accurately delineating fine-grained lesion boundaries due to limited context comprehension and inadequate attention to optimal features. Additionally, the poor fusion of multi-scale semantic cues hinders performance, especially in complex endoscopic scenarios. To address these issues, we introduce ColoSegNet, a Visual Intelligence-Driven Triple Attention Feature Fusion Network designed for high-precision CRC segmentation. Our approach beg... More >

Accurate and real-time polyp segmentation plays a vital role in the early detection of colorectal cancer. However, existing methods often rely on computationally expensive backbones, single attention mechanisms, and suboptimal feature fusion strategies, limiting their practicality in real-world scenarios. In this work, we propose a lightweight yet effective deep learning framework that strikes a balance between precision and efficiency through a carefully designed architecture. Specifically, we adopt a MobileNetV4-based hybrid backbone to extract rich multi-scale features with significantly fewer parameters than conventional backbones, making the model well-suited for resource-constrained cl... More >

To address evolving security challenges in cloud computing, this study proposes a hybrid deep learning architecture integrating Bidirectional Long Short-Term Memory (BiLSTM) and Bidirectional Gated Recurrent Units (BiGRU) for cloud intrusion detection. The BiLSTM-BiGRU model synergizes BiLSTM's long-term dependency modeling with BiGRU's efficient gating mechanisms, achieving a detection accuracy of 96.7% on the CIC-IDS 2018 dataset. It outperforms CNN-LSTM baselines by 2.2% accuracy, 3.3% precision, 3.6% recall, and 3.6% F1-score while maintaining 0.03% false positive rate. The architecture demonstrates operational efficiency through 20% reduced computational latency and 15% lower memory foo... More >