Continuous-time dynamic graphs (CTDG) composed of chronological events, arise in many real-world complex network applications with high dynamics. Memory-based temporal graph neural networks, due to capacity to update and retain historical node states, excel at capturing dynamic changes, handling long-term dependencies with better interpretability compared to memory-less methods. Despite the effectiveness, memory-based methods suffer from two fatal fundamental flaws (i.e. long-standing staleness and memory offline update time issues), which are usually overlooked by constraints of rigid assumptions in recent works. To address the above flaws, in this work, we propose a novel Dynamic Fresh Memory Vigorous Preserver named DyFMVP, a well-designed state-traceable dual memory architecture for efficient anti-staling representation learning. To eliminate staleness within memory, a continuous memory neural ordinary differential equation process (NDP) is proposed to model the underlying memory transition distribution, rejuvenating the memory to a fresh state. In addition, a history-augmented temporal graph attention is proposed to enhance neighbor aggregation in a history-aware manner. Furthermore, a time-aligning RNN-ODE (TARO) restarter is proposed to be offline updating-free and restart at any future timestamps flexibly. Extensive experiments on various downstream tasks demonstrate the primary model DyFMVP and restarter TARO significantly outperform other state-of-the-art CTDG methods by a large margin, while also exhibiting strong robustness against extreme sparsity and long-term interval. Our code has been available on GitHub11https://github.com/CrescentMoon3/icde2025DyFMVP.

Dense prediction in medical volume provides enriched guidance for clinical analysis. CNN backbones have met bottleneck due to lack of long-range dependencies and global context modeling power. Recent works proposed to combine vision transformer with CNN, due to its strong global capture ability and learning capability. However, most works are limited to simply applying pure transformer with several fatal flaws (i.e., lack of inductive bias, heavy computation and little consideration for 3D data). Therefore, designing an elegant and efficient vision transformer learner for dense prediction in medical volume is promising and challenging. In this paper, we propose a novel 3D Shuffle-Mixer network of a new Local Vision Transformer-MLP paradigm for medical dense prediction. In our network, a local vision transformer block is utilized to shuffle and learn spatial context from full-view slices of rearranged volume, a residual axial-MLP is designed to mix and capture remaining volume context in a slice-aware manner, and a MLP view aggregator is employed to project the learned full-view rich context to the volume feature in a view-aware manner. Moreover, an Adaptive Scaled Enhanced Shortcut is proposed for local vision transformer to enhance feature along spatial and channel dimensions adaptively, and a CrossMerge is proposed to skip-connect the multi-scale feature appropriately in the pyramid architecture. Extensive experiments demonstrate the proposed model outperforms other state-of-the-art medical dense prediction methods.