Search

Graph Neural Networks (GNNs) learn low dimensional representations of nodes by aggregating information from their neighborhood in graphs. However, traditional GNNs suffer from two fundamental shortcomings due to their local ($l$-hop neighborhood) aggregation scheme. First, not all nodes in the neighborhood carry relevant information for the target node. Since GNNs do not exclude noisy nodes in their neighborhood, irrelevant information gets aggregated, which reduces the quality of the representation. Second, traditional GNNs also fail to capture long-range non-local dependencies between nodes. To address these limitations, we exploit mutual information (MI) to define two types of neighborhood, 1) \textit{Local Neighborhood} where nodes are densely connected within a community and each node would share higher MI with its neighbors, and 2) \textit{Non-Local Neighborhood} where MI-based node clustering is introduced to assemble informative but graphically distant nodes in the same cluster. To generate node presentations, we combine the embeddings generated by bi-level aggregation - local aggregation to aggregate features from local neighborhoods to avoid noisy information and non-local aggregation to aggregate features from non-local neighborhoods. Furthermore, we leverage self-supervision learning to estimate MI with few labeled data. Finally, we show that our model significantly outperforms the state-of-the-art methods in a wide range of assortative and disassortative graphs., Accepted at IJCNN 2021

Traffic congestion research is on the rise, thanks to urbanization, economic growth, and industrialization. Developed countries invest a lot of research money in collecting traffic data using Radio Frequency Identification (RFID), loop detectors, speed sensors, high-end traffic light, and GPS. However, these processes are expensive, infeasible, and non-scalable for developing countries with numerous non-motorized vehicles, proliferated ride-sharing services, and frequent pedestrians. This paper proposes a novel approach to collect traffic data from Google Map’s traffic layer with minimal cost. We have implemented widely used models such as Historical Averages (HA), Support Vector Regression (SVR), Support Vector Regression with Graph (SVR-Graph), Auto-Regressive Integrated Moving Average (ARIMA) to show the efficacy of the collected traffic data in forecasting future congestion. We show that even with these simple models, we could predict the traffic congestion ahead of time. We also demonstrate that the traffic patterns are significantly different between weekdays and weekends.

Research in deep learning models to forecast traffic intensities has gained great attention in recent years due to their capability to capture the complex spatio-temporal relationships within the traffic data. However, most state-of-the-art approaches have designed spatial-only (e.g. Graph Neural Networks) and temporal-only (e.g. Recurrent Neural Networks) modules to separately extract spatial and temporal features. However, we argue that it is less effective to extract the complex spatio-temporal relationship with such factorized modules. Besides, most existing works predict the traffic intensity of a particular time interval only based on the traffic data of the previous one hour of that day. And thereby ignores the repetitive daily/weekly pattern that may exist in the last hour of data. Therefore, we propose a Unified Spatio-Temporal Graph Convolution Network (USTGCN) for traffic forecasting that performs both spatial and temporal aggregation through direct information propagation across different timestamp nodes with the help of spectral graph convolution on a spatio-temporal graph. Furthermore, it captures historical daily patterns in previous days and current-day patterns in current-day traffic data. Finally, we validate our work's effectiveness through experimental analysis, which shows that our model USTGCN can outperform state-of-the-art performances in three popular benchmark datasets from the Performance Measurement System (PeMS). Moreover, the training time is reduced significantly with our proposed USTGCN model., Comment: Accepted for publication in International Joint Conference on Neural Networks (IJCNN-2021)

This paper analyses how well a Fast Fully Convolutional Network (FastFCN) semantically segments satellite images and thus classifies Land Use/Land Cover(LULC) classes. Fast-FCN was used on Gaofen-2 Image Dataset (GID-2) to segment them in five different classes: BuiltUp, Meadow, Farmland, Water and Forest. The results showed better accuracy (0.93), precision (0.99), recall (0.98) and mean Intersection over Union (mIoU)(0.97) than other approaches like using FCN-8 or eCognition, a readily available software. We presented a comparison between the results. We propose FastFCN to be both faster and more accurate automated method than other existing methods for LULC classification.

This paper demonstrates an application which provides vehicle monitoring and tracking system to prevent sexual/physical harassment in public transport via IoT-based smart technology. In the current scenario of public transport in Bangladesh and all over the world, especially females who usually travel in public transport for various purposes are mostly victims of sexual harassment by some perverse people during travel. Day by day, these kinds of offenses are increasing, and sometimes, it leads to rape and murder. Lack of proper monitoring systems is also responsible for increasing this kind of crime. This paper introduces an Internet of things (IoT)-based application to prevent sexual harassments in public transports by employing tracking system which can provide extremely accurate position, velocity, and real-time information of vehicles. Most of the time, complaining about harassment in the bus by calling police does not prevent harassment because the victim is unable to provide proper information regarding bus’s current location and bus number. Moreover, the authority is also unable to track the bus instantly because they do not have any real-time location information. Due to the lack of real-time information, most of the time sexual harasser gets away. In our system using a smartphone, the passenger will be able to provide bus information and will be able to ask for help from the authorities if the passenger is in danger. As our system provides real-time bus location and shares this information to both the passenger and the authority, it takes less time to take action and helps preventing the sexual harassment.

The availability of images of events almost in real time on social media has a prospect in many application developments. A humanitarian technology for disaster type and level assessment can be developed using the images and video available on social media. In this paper, we investigate the potential use of various available deep learning techniques to develop such an application. For our research, based on the use of publicly available image data, we have started collecting disaster images from various sources from South Asia. We created the South Asia Disaster (SAD) image dataset containing 493 images from various online news portals. Using the Keras as our framework to run our models: Visual Geometry Group (VGG-16 and VGG-19), Inception-V3, and Inception-ResNet-V2 (ResNet: Residual Network). However, to boost up the training speed, we dropped the fully connected layer and added a small, fully connected model. To identify the five different disasters: fire disaster, flood disaster, human disaster, infrastructure disaster, and natural disaster; our proposed method with VGG-16 model’s recognition accuracy was 83.37%, which is the highest accuracy on the SAD dataset.