Neural PDE solvers offer a powerful tool for modeling complex dynamical systems, but often struggle with error accumulation over long time horizons and maintaining stability and physical consistency. We introduce a multiscale implicit neural emulator that enhances long-term prediction accuracy by conditioning on a hierarchy of lower-dimensional future state representations. Drawing inspiration from the stability properties of numerical implicit time-stepping methods, our approach leverages predictions several steps ahead in time at increasing compression rates for next-timestep refinements. By actively adjusting the temporal downsampling ratios, our design enables the model to capture dynamics across multiple granularities and enforce long-range temporal coherence. Experiments on turbulent fluid dynamics show that our method achieves high short-term accuracy and produces long-term stable forecasts, significantly outperforming autoregressive baselines while adding minimal computational overhead.

Increased vehicular flow on roads along with proposed deployment of autonomous vehicles has necessitated the need for accurate traffic forecasting so as to achieve effective route guidance, traffic management, public safety and congestion avoidance. Although a number of traffic forecasting algorithms have been proposed but most of these algorithms perform short term traffic predictions. However future vehicular systems also defined as intelligent VANETs will require a hybrid traffic forecasting model that predicts the vehicular traffic for varying values of time. This paper proposes a time varying forecasting model that predicts vehicular flow by utilizing Long Short-Term Memory (LSTM) and Convolutional Neural Network(CNN). The model is based on large-scale, network-wide traffic with spatio-temporal features. The temporal features learned by LSTM and spatial features learned by CNNs from the matrices are further fused with external factors to derive the final forecast. Model has been implemented on the traffic data set of Chandigarh city in India, mapped onto three two-dimensional matrices of time and space. The predicted information is then forwarded by the vehicle to all the other vehicles in their vicinity using vehicular adhoc networks. Experimental results indicate that the proposed model performs significantly better than other state-of-the-art models in terms of accuracy and efficiency.