Graph Convolution Network Based on Graphlets Structural Information

Elite Class Student Research Training Supervised by Prof. Lianli Gao

Graph is a kind of data structure that can show node information and node connection at the same time, which is very useful in organizing data. Graph Convolutional Network(GCN) introduces machine learning method into graph research and achieves good effect on the semi-supervised classification of nodes. However, at present, GCN only uses the attribute information of graph without considering the structure information of graphs (Kipf and Welling, 2016). We get inspiration from the research method of graphlets for complex network and aim to use both the structure information of graph and attribute information of node to classify nodes (Yaveroglu et al., 2014).

My Method
My Method

Our model calculates the index of structural information for each node in the graph and adds it to the attribute information of nodes for the later convolution operations. In a number of experiments, our approach has improved the accuracy about 0.5% of Graph Convolutional Networks.

Graphlets
Graphlets

You can find full pdf file here

A Comparison of Different Networks Based on the Graphlet Correlation Distance

Elite Class Student Research Training Supervised by Prof. Linyuan Lv

A large number of complex systems in nature can be described by a variety of networks so that complex networks have attracted more and more attention. Nowadays, research concentrate more on the interaction between a small number of roles and some graphlet indicators have been come up with. However, a macroscopic map about different networks based on the indicators remains unclear. In this paper, we used the Graphlet Correlation Distance (GCD) to compare and classify different networks, such as road network, Facebook network and so on. The result showed that some seemingly related networks do have smaller distance, which inspires us the internal formation mechanism of similar networks can be same. This work adds significance to the literature by extending the research on network classification and help us to have a deeper understanding of the nature of the network.

GCM of European Aircraft Route Network
GCM of European Aircraft Route Network

You can find full pdf file here.

Louvre Emergency Evacuation Optimization Plan

Meritorious Winner of 2019 Interdisciplinary Context in Modeling

Since the 21st century, terrorist attacks have intensified. As a famous attraction in France and the world, the research on Louvre’s emergency evacuation program has far-reaching guiding significance for the establishment of emergency evacuation programs for large public facilities in the world.

In order to establish an emergency evacuation model, we analyze the psychological and behavioral characteristics of tourists, the impact of different disasters, and the impact of different population characteristics and density. For the emergency evacuation problem of multi-floor complex buildings, our model is divided into intra-floor models and inter-floor models.

Our intra-floor model consists of the security response zone division model and real-time path planning model based on network.

In the Security Response Zone Division Model, the size of the security response zone(SRZ) is related to the number of visitors in the area and the properties of the passage. We use the goal planning method to calculate the total evacuation time as our objective function to determine the size of the SRZ corresponding to different exits or stairways. Meanwhile, calculating the objective function, we could get the minimum total time for evacuation in a floor.

In the Real-time Path Planning Model Based on Network, the doors are regarded as the nodes. We define the congestion factor, the risk factor, and the corridor length as three factors that influence the evacuation of tourists. The equivalent length, which is based on the three factors, can more accurately reflect the time and the safety of the evacuation of tourists. Therefore, the optimal path with the minimum equivalent length is chosen as the target.

Transforms in Escape Network
Transforms in Escape Network

For the establishment of the inter-floor model, the structure of the network is used to calculate the flow distribution of the human flow after passing through different nodes and edges. According to the real-time traffic time image, we can effectively avoid the occurrence of over-aggregated during the evacuation process.

Finally, we offer a range of solutions that can be combined with our emergency evacuation program to deal with difficult situations, such as how to effectively evacuate small language visitors, how to arrange rescue workers to enter the venue as soon as possible, and so on.

You can find full pdf file here.

Model Structure
Model Structure

More Projects

You can find most of my undergraduate projects here.