This paper presents a hierarchical memory network with time-aware gating to improve forecasting in irregular temporal sequences. The model dynamically adjusts memory span and feature weighting, supporting enhanced forecasting performance in financial, healthcare, and telemetry time series.

Liu Cheng Hao, Zhang Wei Jun, Xu Fang Yu, Huang Xiao Rong, Gao Jun Ming

This paper proposes a multi-objective deep learning framework that enhances spatiotemporal reasoning. By combining evolutionary strategies with neural adaptation, the system optimizes inference speed, energy efficiency, and interpretability in complex real-time environments, enabling high-stakes decision-making with minimal latency.

Siddharth Ramesh Nair, Elena Maria Gutierrez, Fatima Noor Al-Hassan, Rohan Arvind Mehta, Carlos Miguel Soto

We present a transformer-based predictive model optimized for industrial IoT environments. Incorporating multi-head temporal attention and hierarchical encoding, the model improves forecasting accuracy under fluctuating sensor signals while maintaining stability and generalizability across different operating conditions and deployments.

James Frederick Collins, Olivia Katherine Adams, Robert Michael Bennett, Laura Christine Wright, Benjamin Thomas Fields

This study introduces an attention-augmented residual framework to detect faults in multimodal sensor data. The model uses temporal-spatial attention and residual correction to improve anomaly localization, reduce false positives, and adapt dynamically to shifts in sensor input distribution.

Chen Zhao Ming, Xu Li Rong, Liu Bo Tian, Zhang Xiu Wen, Wang Qiang Lin

This work explores self-supervised graph learning methods to improve node classification in heterogeneous information networks. Using structure-preserving contrastive loss and adaptive feature encoding, the model transfers learned representations across domains with minimal labeled data, achieving superior generalization in real-world settings.

Amit Vinod Kapoor, Maria Gabriela Rossi, Taro Kenjiro Watanabe, Neha Shruthi Reddy, Jean Louis Dupont

We propose a temporal causal graph network to model event dependencies in clinical data. The system generates interpretable predictions, supports real-time risk stratification, and identifies causal triggers by mapping latent structures within patient histories using a causal temporal attention mechanism.

Sarah Jennifer Parker, David Anthony Grant, Emily Lauren Whitaker, Thomas Samuel Morris, Olivia Isabelle Clarke

This study introduces a meta-optimized representation model that adapts sequentially to changing task objectives using dual-head attention and feedback loops. It maintains long-term relevance and minimizes performance drop-off in streaming environments and adaptive intelligence platforms.

Emily Catherine Brooks, Ryan Daniel Summers, Nathan Joseph Adams, Laura Isabelle Clarke, Thomas Brandon Ellis, Olivia Rose Ward