GCNII is an extension of a Graph Convolution Networks with two new techniques, initial residual and identify mapping, to tackle the problem of oversmoothing -- where stacking more layers and adding non-linearity tends to degrade performance. At each layer, initial residual constructs a skip connection from the input layer, while identity mapping adds an identity matrix to the weight matrix.
Source: Simple and Deep Graph Convolutional NetworksPaper | Code | Results | Date | Stars |
---|
Task | Papers | Share |
---|---|---|
Node Classification | 5 | 35.71% |
Link Prediction | 2 | 14.29% |
Classification | 1 | 7.14% |
Graph Attention | 1 | 7.14% |
Link Sign Prediction | 1 | 7.14% |
Generalization Bounds | 1 | 7.14% |
Graph Classification | 1 | 7.14% |
Graph Regression | 1 | 7.14% |
Node Classification on Non-Homophilic (Heterophilic) Graphs | 1 | 7.14% |