We introduce a multilingual sentiment analysis model using bidirectional encoders and fusion attention. It handles code-switched and low-resource languages effectively by aligning syntactic and semantic structures, enhancing robustness in culturally diverse online communication streams.

Emily Caroline Walters, Michael Andrew Scott, Olivia Louise Henderson, James Patrick Turner, Sarah Madeleine Clarke

This research proposes a meta-learning strategy tailored for time series tasks with non-stationary trends. The model enables rapid adaptation by learning optimal initialization states, ensuring stable forecasting performance under evolving temporal patterns and sudden structural shifts in data distributions.

Ravi Pradeep Menon, Carlos Daniel Ruiz, Nikhil Shankar Desai, Elena Francesca Martini, Taro Haruki Yamada

This study introduces a hierarchical memory model for task planning through language instructions in robotics. The model supports multi-step reasoning by retaining historical states and dynamically focusing on future objectives, enhancing the agent’s ability to complete compound tasks autonomously.

Laura Margaret Jensen, Robert Andrew Hayes, Thomas Bradley Morgan, Sarah Catherine Winters, Benjamin Lucas Armstrong

This study presents a federated learning framework for clinical collaboration. The model uses adaptive gradient sharing and privacy-preserving layers to enable secure, distributed model training across international healthcare providers without data exchange, improving prediction accuracy while adhering to privacy regulations.

Rohan Deepak Iyer, Jean Claude Dupont, Taro Masahiro Tanaka, Priya Meenakshi Verma, Fatima Noor Al-Rashid

We propose a contrastive knowledge distillation method to enhance multimodal image-language model generalization across domains. The technique aligns semantic embeddings while minimizing feature loss, enabling robust zero-shot predictions in visually grounded language tasks with sparse labeled data.

Sarah Olivia White, Benjamin James Carter, Thomas Andrew Wright, Emily Lauren Clarke, Robert Lucas Freeman

This paper introduces a self-supervised graph learning approach for fraud detection in financial networks. The model enhances scalability and label efficiency using structure-preserving augmentation and contrastive embedding alignment, improving performance in imbalanced, high-volume data environments.

Chen Rui Bo, Zhang Wei Jun, Gao Ming Hao, Xu Lin Feng, Huang Jie Xiang, Liu Qiang Zhao

This paper presents a temporal fusion network that combines LiDAR, RGB, and depth signals to enhance scene perception in autonomous driving. The approach synchronizes modality timelines and leverages cross-signal attention to improve segmentation, tracking, and navigation tasks in real-world settings.

Chen Rong Wei, Liu Kai Sheng, Xu Wen Tao, Zhang Hui Zhong, Gao Liang Rui