Tag
TRACE is a graph neural network simulator for granular dynamics that uses contact-edge memory to improve accuracy and efficiency, reducing long-rollout errors by 31–62% and achieving speedups over traditional solvers.
PULSE is a new executable contract language for spatiotemporal knowledge graph engineering, providing a typed runtime with role-based write effects, safety properties verified in Lean 4, and trace parity across large datasets.
DSETA is a dual-stage continual learning framework for travel time prediction that separates intra-day real-time adaptation from inter-day long-term trend learning, with online A/B tests showing MAE reductions across three cities and successful deployment in DiDi's production environment.
VideoChat3 is a fully open, efficient, and generalist video-centric multimodal large language model that introduces Inflated 3D Vision Transformer (I3D-ViT) and Adaptive Frame Resolution for streaming video perception, along with scalable video data synthesis pipelines, achieving superior performance with only 4B parameters.
Introduces MobiDiff, an end-to-end discrete diffusion framework for generating human mobility data by denoising multi-channel semantic skeletons, achieving faster inference and competitive fidelity on real-world datasets.
This paper introduces a physics-guided machine learning framework that integrates physical constraints into deep learning models (ConvLSTM, AFNONet, ViViT) to predict fuel density for wildfire management, outperforming purely data-driven approaches.
This paper proposes a confusion matrix-based graph construction method and a hybrid loss function for Graph Neural Networks to improve multi-site pollution prediction accuracy and interpretability, evaluated on real-world air pollution data.
MIT researchers developed a long-term memory framework called DAAAM that enables robots to form and recall spatiotemporal memories of environments using language, allowing them to answer queries like 'Where did I leave my wallet?' The method combines advanced map representations with rich descriptions and outperforms state-of-the-art approaches.
This paper introduces SpatioTemporal Causal Network Diagnostics (ST-CND), a framework that uses data-driven causal networks and dynamic mode decomposition to provide localized early warning of geographic tipping points, outperforming classical spatial indicators on synthetic and observational benchmarks.
TrajGenAgent proposes a hierarchical LLM agent framework that decouples macro-level activity planning from micro-level spatiotemporal instantiation for realistic human mobility trajectory generation without fine-tuning. It also introduces an anomaly-detection-based evaluation for behavioral fidelity.
PatchSTG introduces a patch-based spatiotemporal graph Transformer for traffic forecasting on irregular sensor networks, achieving near-linear complexity while maintaining competitive performance.
EnergyMamba proposes a novel spatiotemporal framework combining a graph-enhanced selective state space model and adaptive conformalized quantile regression for accurate and reliable energy consumption prediction with uncertainty estimates, achieving improvements on real-world datasets from Florida, New York, and California.
CHAM-net introduces a contrastive hierarchical adaptive meta-network that captures site-specific and cross-year dynamics for robust global methane flux prediction, outperforming baseline methods on simulation and observational datasets.
Investigates neural integral-operator-based models for fMRI encoding and decoding tasks, focusing on the role of nonlocal spatiotemporal context and showing that larger temporal windows improve performance across datasets.
The Well is a large-scale collection of 15TB of diverse physics simulation datasets across 16 domains, designed to benchmark machine learning surrogate models for spatiotemporal physical systems. It provides a unified PyTorch interface and example baselines to accelerate simulation-based workflows.