This paper presents a federated meta-learning approach for cross-silo training in healthcare. It adapts learning rates per client using gradient modulation, allowing privacy-preserving model sharing and achieving personalized outcomes across hospitals with varied patient populations and local data distributions.

Amit Raghunath Mehta, Fatima Sofia Noor, Jean Michel Bourdain, Priya Kavita Desai, Carlos Eduardo Suarez

This study introduces a curriculum-based few-shot framework for rare disease classification. It employs episodic memory and structured clinical embeddings, enabling rapid generalization to new diseases using few examples, while maintaining performance and interpretability in medical diagnosis systems.

Neha Swathi Narayanan, Jean Paul Laurent, Elena Maria Gutierrez, Kunal Arvind Shetty, Taro Kenji Nakamoto

We propose a contrastive framework for aligning multimodal data in egocentric action recognition. By synchronizing audio-visual embeddings with temporal contrast, the model enhances interaction understanding, improving performance in wearable camera footage and assistive technology applications.

Thomas Edward Hamilton, Olivia Claire Thompson, Rachel Louise Adams, Benjamin Michael Carter, Emily Isabelle Green

This paper presents a legal document retrieval system using hierarchical embeddings and graph-based reasoning. The architecture enables context-aware analysis of complex legislative documents and improves semantic search precision in legal research tools and case law analytics platforms.

Chen Hao Ming, Zhang Rui Yong, Liu Fang Zhen, Gao Xiu Ming, Xu Lin Bo, Huang Jing Tao

This work introduces a self-supervised model for audio-visual scene understanding. Using layered attention and temporal encoding, it aligns modalities without labels, achieving strong performance in surveillance, media tagging, and human activity recognition tasks under minimal supervision.

Michael James Foster, Emily Anne Patterson, Laura Victoria Clark, Robert Anthony Wells, Olivia Grace Simmons

We propose a reinforcement learning model for smart grid optimization. It adapts to energy demand and market fluctuations, improving cost efficiency and reliability in demand response systems while meeting real-time operational and environmental compliance goals.

Ravi Shankar Kulkarni, Elena Sofia Navarro, Priya Deepa Iyer, Jean Claude Dupont, Carlos Rodrigo Herrera

We propose a graph neural model with spatiotemporal attention for anomaly detection. It processes high-dimensional sensor inputs across space and time, identifying deviations in dynamic systems, and enhancing early warning accuracy in complex industrial monitoring applications.

Chen Zhi Wei, Liu Min Fang, Zhang Tian Lei, Xu Bo Han, Gao Rui Sheng, Huang Lin Cheng