Emergency Medical Services (EMS) are critical for saving lives. In highly populated cities, where millions of emergencies are reported each year, delays in ambulance response and inefficient hospital routing can result in increased mortality and adverse patient outcomes. Traditional forecasting systems lack the spatial resolution and capacity to integrate essential factors like regional demand, hospital facility demand, and weather dynamics. To overcome these limitations, this paper presents a forecasting system that predicts EMS demand at ZIP-code granularity using spatio-temporal deep learning methods and a hospital routing algorithm. Utilizing a decade-long real-world EMS incident dataset, we demonstrate how the incident data patterns impact forecasting, hospital routing, and how these insights can be integrated into intelligent transportation and public health systems. Github: https://github.com/saideepthi112/EMS-Demand-Prediction-Project

Predicting traffic accident hotspots is crucial for ensuring public safety, improving transport planning, and reducing transportation costs. Traditional deep learning models, such as Transformers and LSTMs, have been successful in this field but fail to integrate critical attributes essential for accurate prediction. To address these limitations, we propose utilizing a Temporal Convolutional Network (TCN), which efficiently learns spatial, temporal, and other external factors integral to accident hotspot prediction. Our proposed TCN architecture 1 demonstrate superior performance over state-of-the-art methods, offering valuable insights for proactive accident mitigation.

Graph Neural Networks (GNNs) have shown promising performance in many applications, yet remain extremely difficult to train over large-scale graph datasets. Existing weight pruning techniques can prune out the layer weights; however, they cannot fully address the high computation complexity of GNN inference, caused by large graph size and complicated node connections. In this paper, we propose an Iterative Gradient Rank Pruning (IGRP) algorithm to find graph lottery tickets (GLT) of GNNs where each GLT includes a pruned adjacency matrix and a sub-network. Our IGRP can avoid layer collapse and the winning ticket achieves Maximal critical compression. We evaluate the proposed method on small-scale (Cora and Citeseer), medium-scale (PubMed and Wiki-CS), and large-scale (Ogbn-ArXiv and Ogbn-Products) graph datasets. We demonstrate that both Single-shot and Multi-shot of IGRP outperform the state-of-the-art unified GNN sparsification (UGS) framework on node classification. The source code can be found in https://github.com/poweiharn/IGRP_GNN.

The location-based alert system plays a primary factor on determining who is at risk during an emergency, such as a war zone in Ukraine. While users are willing to reveal their locations in exchange for timely alert in those situations, there is no guarantee that their private information does not fall into the wrong hands. For example, a soldier may be killed if his movement pattern is known by the enemy. One resolution to this issue is to encrypt the location information by trusted authority public key before it is transmitted. This approach provides location privacy and allows decryption only when the recipient’s location satisfies a certain predicate. However, the encryption itself may still be compromised if the location encoding is leaked. In this paper, we propose a Hilbert Curve Encoding which encrypts the user’s message along with her locations for private processing with the trusted authority. We further propose a hybrid HNGM−NEncoding which combines the Hilbert Curve Encoding and Gray Encoding. HNGM−Nhas the proprieties of a Hilbert Curve Encoding in its identifier and a Hamming distance of 1 between neighboring cells in a subgrid. As a consequence, the proposed HNGM−Nis less likely to leak neighboring cell identifier than Gray Encoding under random guessing attacks. Extensive experiment results show that our encoding methods are better than Hierarchical Encoding and comparable to Gray Encoding in terms of user response time, token remaining percentage, and execution time.