This research presents a reinforcement learning-based approach to feature engineering for dynamic risk assessment in autonomous systems. The model adaptively constructs predictive representations, improving decision-making reliability in high-risk, unstructured environments encountered during robotic navigation and exploration.

Ravi Ganesh Patil, Elena Gabrielle Rossi, Aditya Vikas Mehra, Jean Pierre Duval, Maria Lourdes Serrano

A multitask attention network is proposed for analyzing legal and policy texts. It learns shared representations for multi-label classification, improving relevance detection and interpretability in long, domain-specific documents across legal systems with complex terminologies and overlapping jurisdictions.

Liu Chang Hao, Chen Jing Yu, Xu Zhi Feng, Gao Rui Xin, Zhang Lei Ming, Huang Hong Jie

This paper presents a federated transfer learning framework designed for risk prediction in healthcare. By enabling knowledge sharing across institutions without data exchange, the model enhances accuracy and generalization while preserving patient privacy in decentralized, resource-constrained clinical environments.

Amit Raghavendra Mehta, Jean Louis Dupont, Fatima Noor Al-Qasim, Priya Nandita Reddy, Carlos Eduardo Martinez

We propose a dual-attention emotion recognition model using sequential context alignment to fuse visual and audio data. The system enhances temporal sensitivity and emotion label consistency across modalities, improving classification in real-world conditions where expressions vary across time and context.

Emily Jane Cooper, Robert Andrew Simmons, Olivia Katherine Moore, Benjamin Charles Wright, Laura Michelle Thompson

This paper introduces a temporal graph network for predictive maintenance using irregular IoT data. The model captures relational and time-dependent signals across machines, improving failure prediction by modeling interactions and adapting to asynchronous sampling intervals in high-frequency environments.

Liu Fang Zhou, Zhang Rui Cheng, Xu Wei Liang, Gao Tian Ming, Huang Xiu Lin, Chen Bo Wen

We introduce a meta-learning memory network tailored for personalized tutoring systems. The architecture adapts to individual learning trajectories using sparse feedback and historical patterns, enabling fine-grained educational personalization without the need for frequent retraining or extensive interaction data.

Neha Swathi Kapoor, Taro Masashi Nakamura, Elena Francesca Bianchi, Kunal Arvind Desai, Jean Pierre Rousseau

This paper presents an ensemble of deep networks enhanced with Bayesian inference to detect rare anomalies in financial time series. It captures uncertainty through posterior sampling, offering interpretable predictions and strong generalization across highly volatile and imbalanced financial datasets.

James Thomas Walker, Sarah Louise Bennett, Robert Daniel Greene, Olivia Rachel Morgan, Emily Grace Holmes