Skip to main content

Questions tagged [graph-neural-network]

Graph neural networks are a class of neural networks which are designed to operate on graph structured data. They typically make use of graph convolution layers, a generalization of the usual "lattice" convolutions used in CNNs.

Filter by
Sorted by
Tagged with
0 votes
1 answer
23 views

Does a model learn the same attention scores when retrained?

As in the title, should I expect a model to learn almost the same attention scores in its attention layers when I train it? Perhaps only in the first one if there are multiple such layers? It feels ...
rick's user avatar
  • 11
0 votes
0 answers
39 views

Systematic bias of neural network regression

I was trying to do graph-level regression task using graph convolutional networks, basically I concatenated 3 linear layers after several GCN modules, I used ReLU activation function before each ...
Tianjian Qin's user avatar
0 votes
1 answer
80 views

Background reason for the terms ‘Isotropic’ and ‘Anisotropic’ in the context of GNN Message Passing

I’m reading a paper on Graph Neural Networks (GNNs) that uses the terms ‘isotropic’ and ‘anisotropic’ in the context of message passing. I understand that these terms originate from physics, chemistry,...
MohammadJavad Vaez's user avatar
0 votes
0 answers
15 views

What does it mean for spectral graph representation methods to be limited to a single graph structure?

I'm trying to get deeper into the topic of graph neural networks. While reading this paper recently, I came across a statement that confused me. The authors state that spectral methods are limited to ...
Ferdinand Mütsch's user avatar
0 votes
1 answer
55 views

Missing features for nodes in a Graph Neural Network (GNN)

I have built an undirected graph that I want to use to train a binary classifier. The graph represents a network of clients connected through the addresses they used to register. For example, if my ...
Arturo Sbr's user avatar
0 votes
0 answers
34 views

Exploit temporal information in GCN for the Elliptic dataset

I am trying to reproduce the results obtained on the Elliptic dataset from [1] (the EvolveGCN) and [2] using PyTorch geometric and PyTorch Geometric temporal libraries. The problem I am facing is that ...
Luigi D'Amico's user avatar
1 vote
1 answer
38 views

Difference between GNN and sparsely connected feedforward NN

What characteristics differentiate a classic GNN and a sparsely connected feedforward NN (basically a modified fully-connected NN), where the sparse connectivity is given by a user-defined sparsity ...
12213119's user avatar
0 votes
0 answers
9 views

How can GCN model be applied to find next link based on similarity of values between its edges?

I want to apply GCN model to predict next link based on similarity of values between its neighbors nodes. For example I have total of 13 nodes and edges = [ (1, 2), (1, 3), (2, 3), (3, 4), (3, 5), (4, ...
Iqra's user avatar
  • 1
1 vote
1 answer
73 views

What is the “web” drawing for a graph neural network?

It is common to draw a neural network as a "web" of neurons and connections, such as the "web" below of a multilayer perception that has input neurons in white, hidden neurons in ...
Dave's user avatar
  • 63.7k
0 votes
0 answers
74 views

Alternatives to classical softmax for better aggregation

I am doing machine learning on phylogenetic trees. Species within the trees have several relations, such as living at the same time or not, and I would like to use a global attention mechanism which ...
25laps's user avatar
  • 43
0 votes
0 answers
15 views

Do all nodes have to be stored / indexed in a GNN?

Apologies in advance for what maybe a really silly question :( I am trying to construct a Graph Neural Network, in which I would like to learn representations for certain nodes, but not others. To be ...
dendog's user avatar
  • 212
0 votes
0 answers
136 views

Visualizing multisource Breadth-First Search (BFS) on large dataset

I have a very large dataset (2.4 million vertices and 123 million edges) and I'd like to visualize running multisource Breadth-First Search (BFS) on it. In particular, each starting node is given a ...
Yaseen's user avatar
  • 1
0 votes
0 answers
28 views

When using early stopping for image classification with graph convolutional networks, which value for minimal improvement and patience is recommended?

I'm using early stopping for avoiding overfitting in a task of image classification with a graph convolutional network. I'm trying to adopt a patience of 5 and a minimal improvement of 0.001. Is this ...
Zaratruta's user avatar
  • 1,008
1 vote
2 answers
429 views

Analogy between MPNNs and GCNs?

I'm currently trying to understand graph neural networks and got stuck with comparing MPNN and GCN. In my understanding, MPNN is more of a "general framework", while GCN is a specific "...
Ferdinand Mütsch's user avatar
0 votes
0 answers
17 views

How to deal with masked/fuzzy node in graph classification?

I’m new to graph embeddings, and I have some trouble formulating a solution Suppose that I have a generic node classification dataset that tell the directed edge links between each node along with its ...
Wakeme UpNow's user avatar
1 vote
0 answers
24 views

Node classification with random labels for GNNs

I decided to train GCN on the Cora dataset for the node classification task, however, with the random labels, i.e., applying np.random.shuffle(labels). For the ...
RobJan's user avatar
  • 198
1 vote
0 answers
113 views

How can I put GNN and CNN together?

I am trying to combine GNN and CNN in my model. Every node of my graph has 2d space coordinates, so as a whole it's like an irregular 2d mesh. I think we can't use deep GNNs due to oversmoothing. So I ...
landings's user avatar
  • 111
3 votes
1 answer
66 views

Advice on multi-output regression task: Forecasting Refugee Flows

I have a question on possible approaches in modeling and predicting refugee flows. Task My task is to predict the number of refugees for a country origin-destination pair for a given year. For example,...
Slash's user avatar
  • 93
1 vote
0 answers
190 views

How to be explicit about negative sampling for Link prediction using GNN?

Regarding recommendation systems of bi-partite graphs with PyTorch geometric, most of the tutorials I found about Link prediction using GNNs suggest using negative sampling in an (I guess) random way. ...
James's user avatar
  • 135
1 vote
0 answers
27 views

Link prediction on completely edgeless graphs

All existing link prediction methods (and indeed, all GNN methods in general) rely on edges being present in the input graph, and the method would be able to predict any missing edges. I've seen one ...
Haritha Jayasinghe's user avatar
2 votes
1 answer
168 views

How testing (new data points) works in graph neural network

In machine learning, data is divided into train and test splits. The machine learns weights using training data and we can utilize weights to predict test data. Similarly, we are also learning weights ...
Pragnesh Rana's user avatar
1 vote
0 answers
17 views

How to tell feature vectors are sufficient for a graph neural network?

Not sure the right terminology, but I am working on a node classification problem, and I am only able to obtain a training accuracy of 50% for my GNN (test accuracy is 40%). It seems I should be able ...
Ralff's user avatar
  • 203
2 votes
1 answer
2k views

How to interpret the global max pooling operation in graph neural networks?

I'm trying to use pytorch geometric for building graph convolutional networks. And I'm trying to interpret the result of the max pooling operation, which is described in this link: https://pytorch-...
Zaratruta's user avatar
  • 1,008
1 vote
0 answers
17 views

How to mathematically describe a graph convolutional network architecture?

I have developed a GCN architecture with 3 different inputs, related to 3 different blocks of layers. After that, the outputs of those layers are concatenated. And after that, the architecture has a ...
Zaratruta's user avatar
  • 1,008
1 vote
0 answers
33 views

What parameters of a neural network should be changed when increasing the number of training samples?

I was testing a DL model (GNN with two GCN layers and one linear layer) on a small dataset for a regression purpose, the resulted MAE and the scatter plot showed some really good results. However, ...
Batoul Diab's user avatar
4 votes
2 answers
47 views

Why Deep Learning needs to be performed in Graphical representations?

Why do we bother about devising new Deep Learning methods, which is to be applied on graphical representation of data? Why not use the traditional Deep Learning methods in the adjacency matrix ...
Hari Krishnan U's user avatar
1 vote
0 answers
54 views

Can I apply graph attention networks in datasets with graphs with different sizes?

I would like to test graph attention networks in a problem. However, my dataset has graphs with very different numbers of nodes (ranging from 1 to hundreds). Can I use this kind of network in this ...
Zaratruta's user avatar
  • 1,008
2 votes
1 answer
2k views

Why do graph convolutional neural networks use normalized adjacency matrices?

It seems that it is common to perform something like the following operation (like in Kipf & Welling, "Semi-Supervised Classification with Graph Convolutional Networks" 2017) to ...
pete's user avatar
  • 185
2 votes
1 answer
392 views

Predict unseen samples with Graph Neural Networks

I am recently studying graph neural networks (GNNs). I have read some papers, e.g., Semi-Supervised Classification with Graph Convolutional Networks, and blogs, e.g., Training Graph Convolutional ...
CLRW97's user avatar
  • 121
1 vote
1 answer
3k views

Why do we normalize a vector with its with l2 norm is there some intuition behind?

Hello I am trying to undestand the message propagation algorithm of GraphSAGE(https://arxiv.org/pdf/1706.02216.pdf) In step 7 there is a division with l2 norm (if I understand the notation correctly )....
partizanos's user avatar
2 votes
0 answers
291 views

Is normalizing input graph's node attributes before training GNNs required? How to do it

I am new in GNN, and I have a dataset of weighted graphs that each node in each graph has the following attributes: ...
meee's user avatar
  • 21
1 vote
0 answers
137 views

How can I apply graph neural networks with different number of nodes in each instance for image classification?

I have a dataset with images of different sizes and I would like to maintain all the information of the images. I would like to test graph neural networks for doing that. Is it possible to apply GNN ...
Zaratruta's user avatar
  • 1,008
3 votes
1 answer
469 views

Is it possible to deal with datasets of graphs with different number of nodes in graph nural networks?

I'm dealing with a graph classification problem. In my dataset, each graph has som specific number of nodes. The number of nodes has a range of 1-1000 nodes. At inference time (after training), the ...
Zaratruta's user avatar
  • 1,008
2 votes
0 answers
351 views

Encoding Geolocation data

I am working on routing bus from one stop to another, for which Geolocation data inform of latitude and longitude is required. In addition to xy coordinates, distance matrix of locations is also ...
Hina Ali's user avatar
1 vote
0 answers
47 views

Need help to interpret Node2Vec grammar

I am studying Link Prediction using Node2Vec refering to a tutorial on the web. The following equation is what I don't understand. Why is it this way? ...
xabzakabecd's user avatar
  • 3,495
0 votes
0 answers
48 views

How can I make Link Features for Link Prediction in Graph Convolutional Network or Logisitic Regression?

Currently studying Graph Convolutional Network or Logistic Regression and have come to understand how it works and how I can do node classifications with given node features but I have a difficulty in ...
xabzakabecd's user avatar
  • 3,495
1 vote
1 answer
285 views

Why do I even need DeepWalk and Node2Vec when I can build a visual graph structure?

While studying DeepWalk, I started wondering why I need "DeepWalk" when I can build a graph from data and visualize the structure of a graph. With a visualized graph, I can see which nodes ...
xabzakabecd's user avatar
  • 3,495
4 votes
1 answer
1k views

Weighted adjacency matrix normalization for GCN, how to normalize? what about self-loop values?

I am implementing a GCN that will work on a weighted graph. The edges' weights are in the range [1, 250]. When it comes to normalizing the adjacency matrix for GCNs, the standard formula of a ...
mik1904's user avatar
  • 378
4 votes
2 answers
2k views

Nodes' attribute scaling/normalization before graph embedding learning - GNN?

In a node classification setting, is it require to normalize/scale graph node attributes before learning node embeddings using graph neural networks? Why?
mik1904's user avatar
  • 378
3 votes
1 answer
132 views

Can you predict a totally unconnected datapoint with a Graph Neural Net model

Let's say I am building and training a model based on Graph Neural Net to detect bot accounts in the Social Network Graph. I have a graph dataset that I will be using to train, validate and test the ...
Kartikeya Sharma's user avatar
-1 votes
1 answer
138 views

Why do we need transpose graph (reverse edges) in GNN? [closed]

Especially in heterogeneous graphs, when we go on GNN tasks, reverse edges (transpose graph) are added and I do not know the reason exactly. Can someone explain this in a detailed way?
S. Jay's user avatar
  • 35
2 votes
1 answer
200 views

Spectral Graph Convolutions: What are the spectral filters functions

I am trying to understand the mathematical meaning of one of the steps that appear in the Convolution Theorem (Step 4 here). To give some context, this is related to applying the convolution theorem ...
Gonzalo Polo's user avatar
2 votes
0 answers
95 views

Find key nodes in Graph Neural Netwroks

Given a graph dataset, in which links of graphs are the same while features of each node may be varied, how can we locate those critical nodes in this graph structure that contribute the most to ...
yzongy's user avatar
  • 21
1 vote
0 answers
77 views

Are graphical neural networks the right approach for isomorphic graphs?

I have a set of $N$ observations ($N>100,000$): each observation takes the form of a homogeneous, undirected graph $G_n=(V,E)$ all graphs $G_n$ have the same nodes and edges - around 5,000 nodes ...
bayesian_brain's user avatar
0 votes
0 answers
38 views

Equivariance definition

Equivariance in deep learning, it is actually defined as a property of some function $f$ to permute the outputs according to permutation of inputs, i.e. $$f(PAP^T) = Pf(A)P^T$$ for some permutation ...
James Arten's user avatar
2 votes
1 answer
227 views

Trace of quadratic form with Laplacian matrix notation

Reading some papers about spectral graph analysis and graph neural networks I have found the following notation which I'm not too sure how to expand: Given matrices $F, L \in \mathbb{R}^{n \times n}$, ...
James Arten's user avatar
5 votes
1 answer
6k views

Difference between graph neural network and graph convolutional network

Which characteristics my neural network (NN) model should have to be considered as a graph convolutional network (GCN) instead of a graph neural network (GNN)? I know that GCN is a variant of GNN, but ...
Henrique Toste's user avatar
1 vote
1 answer
54 views

Can we add dilated layers to Graph Neural Nets?

GNNs are not as deep as CNNs (due to over-smoothing and other factors). Is this possible to have dilated in gcn?
Jia's user avatar
  • 31
2 votes
1 answer
109 views

What are the meanings of Node Classification, Link Prediction, Graph Classification in Graph Neural Network?

I am currently studying Graph Neural Network but I have some difficulty in understanding what I can do after having studied Graph Neural Network. From having gained a bit of understand in Graph ...
xabzakabecd's user avatar
  • 3,495
1 vote
0 answers
43 views

Understanding convolutional operators in graphs

In a Graph Convolutional Neural Network, a convolutional operation is $$ h_i^{(l)} = \sigma\left(W^{(l)} \cdot \text{Agg}\left\{ h_j^{(l-1)},\forall j \in \hat{N}(i)\right\} \right)$$ Is my intuition ...
Jia's user avatar
  • 31