### 技术控

今日:289| 主题:58108

# [其他] Graph Convolutional Networks – An introduction to neural networks on graphs

226 1
 Graph Convolutional Networks – An introduction to neural networks on graphs    Multi-layer Graph Convolutional Network (GCN) with first-order filters.     Overview   Many important real-world datasets come in the form of graphs or networks: social networks, knowledge graphs, protein-interaction networks, the World Wide Web, etc. (just to name a few). Yet, until recently, very little attention has been devoted to the generalization of neural network models to such structured datasets.    In the last couple of years, a number of papers re-visited this problem of generalizing neural networks to work on arbitrarily structured graphs ( Bruna et al. , ICLR 2014; Henaff et al. , 2015; Duvenaud et al. , NIPS 2015; Li et al. , ICLR 2016; Defferrard et al. , NIPS 2016; Kipf & Welling , 2016), some of them now achieving very promising results in domains that have previously been dominated by, e.g., kernel-based methods, graph-based regularization techniques and others.   In this post, I will give a brief overview of recent developments in this field and point out strengths and drawbacks of various approaches. The discussion here will mainly focus on two recent papers:        Kipf & Welling (2016), Semi-Supervised Classification with Graph Convolutional Networks (disclaimer: I'm the first author)    Defferrard et al. (NIPS 2016), Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering      and a review/discussion post by Ferenc Huszar: How powerful are Graph Convolutions? that discusses some limitations of these kinds of models. I wrote a short comment on Ferenc's review(at the very end of this post).   Outline        Short introduction to neural network models on graphs    Spectral graph convolutions and Graph Convolutional Networks (GCNs)    Demo: Graph embeddings with a simple 1st-order GCN model    GCNs as differentiable generalization of the Weisfeiler-Lehman algorithm      If you're already familiar with GCNs and related methods, you might want to jump directly toEmbedding the karate club network.   How powerful are Graph Convolutional Networks?   Recent literature    Generalizing well-established neural models like RNNs or CNNs to work on arbitrarily structured graphs is a challenging problem. Some recent papers introduce problem-specific specialized architectures (e.g. Duvenaud et al. , NIPS 2015; Li et al. , ICLR 2016), others make use of graph convolutions known from spectral graph theory( Bruna et al. , ICLR 2014; Henaff et al. , 2015) to define parameterized filters that are used in a multi-layer neural network model, akin to "classical" CNNs that we know and love.    More recent work focuses on bridging the gap between fast heuristics and the slow, but somewhat more principled, spectral approach. Defferrard et al. (NIPS 2016) approximate smooth filters in the spectral domain using Chebyshev polynomials with free parameters that are learned in a neural network-like model. They achieve convincing results on regular domains (like MNIST), closely approaching those of a simple 2D CNN model.    In Kipf & Welling (2016), we take a somewhat similar approach and start from the framework of spectral graph convolutions, yet introduce simplifications (we will get to those later in the post) that in many cases allow both for significantly faster training times and higher predictive accuracy, reaching state-of-the-art classification results on a number of benchmark graph datasets.   GCNs Part I: Definitions    Currently, most graph neural network models have a somewhat universal architecture in common. I will refer to these models as Graph Convolutional Networks (GCNs); convolutional, because filter parameters are typically shared over all locations in the graph (or a subset thereof as in Duvenaud et al. , NIPS 2015).    For these models, the goal is then to learn a function of signals/features on a graph $$\mathcal{G}=(\mathcal{V}, \mathcal{E})$$ which takes as input:        A feature description $$x_i$$ for every node $$i$$ ; summarized in a $$N\times D$$ feature matrix $$X$$ ( $$N$$ : number of nodes, $$D$$ : number of input features)    A representative description of the graph structure in matrix form; typically in the form of an adjacency matrix $$A$$ (or some function thereof)      and produces a node-level output $$Z$$ (an $$N\times F$$ feature matrix, where $$F$$ is the number of output features per node). Graph-level outputs can be modeled by introducing some form of pooling operation (see, e.g. Duvenaud et al. , NIPS 2015).    Every neural network layer can then be written as a non-linear function $H^{(l+1)} = f(H^{(l)}, A) \, ,$ with $$H^{(0)} = X$$ and $$H^{(L)} = Z$$ (or $$z$$ for graph-level outputs), $$L$$ being the number of layers. The specific models then differ only in how $$f(\cdot, \cdot)$$ is chosen and parameterized.   GCNs Part II: A simple example   As an example, let's consider the following very simple form of a layer-wise propagation rule:   $f(H^{(l)}, A) = \sigma\left( AH^{(l)}W^{(l)}\right) \, ,$    where $$W^{(l)}$$ is a weight matrix for the $$l$$ -th neural network layer and $$\sigma(\cdot)$$ is a non-linear activation function like the $$\text{ReLU}$$ . Despite its simplicity this model is already quite powerful (we'll come to that in a moment).    But first, let us address two limitations of this simple model: multiplication with $$A$$ means that, for every node, we sum up all the feature vectors of all neighboring nodes but not the node itself (unless there are self-loops in the graph). We can "fix" this by enforcing self-loops in the graph: we simply add the identity matrix to $$A$$ .    The second major limitation is that $$A$$ is typically not normalized and therefore the multiplication with $$A$$ will completely change the scale of the feature vectors (we can understand that by looking at the eigenvalues of $$A$$ ). Normalizing $$A$$ such that all rows sum to one, i.e. $$D^{-1}A$$ , where $$D$$ is the diagonal node degree matrix, gets rid of this problem. Multiplying with $$D^{-1}A$$ now corresponds to taking the average of neighboring node features. In practice, dynamics get more interesting when we use a symmetric normalization, i.e. $$D^{-\frac{1}{2}}AD^{-\frac{1}{2}}$$ (as this no longer amounts to mere averaging of neighboring nodes). Combining these two tricks, we essentially arrive at the propagation rule introduced in Kipf & Welling (2016):   $f(H^{(l)}, A) = \sigma\left( \hat{D}^{-\frac{1}{2}}\hat{A}\hat{D}^{-\frac{1}{2}}H^{(l)}W^{(l)}\right) \, ,$    with $$\hat{A} = A + I$$ , where $$I$$ is the identity matrix and $$\hat{D}$$ is the diagonal node degree matrix of $$\hat{A}$$ .    In the next section, we will take a closer look at how this type of model operates on a very simple example graph: Zachary's karate club network (make sure to check out the Wikipedia article !).   GCNs Part III: Embedding the karate club network       Graph Convolutional Networks – An introduction to neural networks on graphs     Karate club graph, colors denote communities obtained via modularity-based clustering ( Brandes et al. , 2008).      Let's take a look at how our simple GCN model (see previous section or Kipf & Welling , 2016) works on a well-known graph dataset: Zachary's karate club network (see Figure above).    We take a 3-layer GCN with randomly initialized weights. Now, even before training the weights, we simply insert the adjacency matrix of the graph and $$X = I$$ (i.e. the identity matrix, as we don't have any node features) into the model. The 3-layer GCN now performs three propagation steps during the forward pass and effectively convolves the 3rd-order neighborhood of every node (all nodes up to 3 "hops" away). Remarkably, the model produces an embedding of these nodes that closely resembles the community-structure of the graph (see Figure below). Remember that we have initialized the weights completely at random and have not yet performed any training updates (so far)!       Graph Convolutional Networks – An introduction to neural networks on graphs    GCN embedding (with random weights) for nodes in the karate club network.      This might seem somewhat surprising. A recent paper on a model called DeepWalk ( Perozzi et al. , KDD 2014) showed that they can learn a very similar embedding in a complicated unsupervised training procedure. How is it possible to get such an embedding more or less "for free" using our simple untrained GCN model?   We can shed some light on this by interpreting the GCN model as a generalized, differentiable version of the well-known Weisfeiler-Lehman algorithm on graphs. The (1-dimensional) Weisfeiler-Lehman algorithm works as follows:    For all nodes $$v_i\in \mathcal{G}$$ :        Get features $$\{h_{v_j}\}$$ of neighboring nodes $$\{v_j\}$$    Update node feature $$h_{v_i} \leftarrow \text{hash}\left(\sum_j h_{v_j}\right)$$ , where $$\text{hash}(\cdot)$$ is (ideally) an injective hash function      Repeat for $$k$$ steps or until convergence.    In practice, the Weisfeiler-Lehman algorithm assigns a unique set of features for most graphs. This means that every node is assigned a feature that uniquely describes its role in the graph. Exceptions are highly regular graphs like grids, chains, etc. For most irregular graphs, this feature assignment can be used as a check for graph isomorphism (i.e. whether two graphs are identical, up to a permutation of the nodes).   Going back to our Graph Convolutional layer-wise propagation rule (now in vector form):   $h^{(l+1)}_{v_i} = \sigma \left( \sum_{j} \frac{1}{c_{ij}}h^{(l)}_{v_j}W^{(l)} \right) \, ,$    where $$j$$ indexes the neighboring nodes of $$v_i$$ . $$c_{ij}$$ is a normalization constant for the edge $$(v_i,v_j)$$ which originates from using the symmetrically normalized adjacency matrix $$D^{-\frac{1}{2}}AD^{-\frac{1}{2}}$$ in our GCN model. We now see that this propagation rule can be interpreted as a differentiable and parameterized (with $$W^{(l)}$$ ) variant of the hash function used in the original Weisfeiler-Lehman algorithm. If we now choose an appropriate non-linearity and initialize the random weight matrix such that it is orthogonal (or e.g. using the initialization from Glorot & Bengio , AISTATS 2010), this update rule becomes stable in practice (also thanks to the normalization with $$c_{ij}$$ ). And we make the remarkable observation that we get meaningful smooth embeddings where we can interpret distance as (dis-)similarity of local graph structures!   GCNs Part IV: Semi-supervised learning    Since everything in our model is differentiable and parameterized, we can add some labels, train the model and observe how the embeddings react. We can use the semi-supervised learning algorithm for GCNs introduced in Kipf & Welling (2016). We simply label one node per class/community (highlighted nodes in the video below) and start training for a couple of iterations:                Semi-supervised classification with GCNs: Latent space dynamics for 300 training iterations with a single label per class. Labeled nodes are highlighted.      Note that the model directly produces a 2-dimensional latent space which we can immediately visualize. We observe that the 3-layer GCN model manages to linearly separate the communities, given only one labeled example per class. This is a somewhat remarkable result, given that the model received no feature description of the nodes. At the same time, initial node features could be provided, which is exactly what we do in the experiments described in our paper ( Kipf & Welling , 2016) to achieve state-of-the-art classification results on a number of graph datasets.   Conclusion   Research on this topic is just getting started. The past several months have seen exciting developments, but we have probably only scratched the surface of these types of models so far. It remains to be seen how neural networks on graphs can be further taylored to specific types of problems, like, e.g., learning on directed or relational graphs, and how one can use learned graph embeddings for further tasks down the line, etc. This list is by no means exhaustive and I expect further interesting applications and extensions to pop up in the near future. Let me know in the comments below if you have some exciting ideas or questions to share!   Thanks to the following people:   Max Welling, Taco Cohen, Chris Louizos and Karen Ullrich (for many discussions and feedback both on the paper and this blog post). Also I'd like to thank Ferenc Huszar for highlighting some drawbacks of these kinds of models.   A note on completeness   This blog post constitutes by no means an exhaustive review of the field of neural networks on graphs. I have left out a number of both recent and older papers to make this post more readable and to give it a coherent story line. The papers that I mentioned here will nonetheless serve as a good start if you want to dive deeper into this topic and get a complete overview of what is around and what has been tried so far.   Citation    If you want to use some of this in your own work, you can cite our paper on Graph Convolutional Networks: @article{kipf2016semi,  title={Semi-Supervised Classification with Graph Convolutional Networks},  author={Kipf, Thomas N and Welling, Max},  journal={arXiv preprint arXiv:1609.02907},  year={2016}}复制代码 Source code    We will release the source code of all experiments described here and in the paper very soon. Follow me on Twitter , if you want to make sure not to miss the release.   The issue with regular graphs

 回个帖子，下班咯~

• ## 万字长文 | 共享汽车的宿命

如今共享经济与各行各业都有扯不清的关系， [...]

• ## 2017超流行十款发色推荐，炎炎夏日就要“色

微信号 jnntf001 [...]

• ## 拆了那些格子间，梦想加想让你单纯的“享受

题图：梦想加望京店 “你见过10年前的豪车 [...]

• ## 睡不着？Sana智能护目镜让您摆脱失眠困扰

相信大部分人都有过失眠的体验，虽然失眠只是一 [...]

• ## 瞄准近千亿农业植保市场，弧光航空欲用大载

随着中国土地流转率的提升、农村年龄结构老化 [...]

• ## Blockchains are the new Linux, not the n

Cryptocurrencies are booming beyond belief [...]

• ## 很好，看完这个节目让我觉得自己就是个什么

微信号 sinaenterta [...]

• ## 暴击！93年的孙怡素颜像董子健妈！不保养后

微信号 aq8050 董子 [...]

• ## 你和时髦之间还差一件马甲！

微信号 ishepin 马 [...]

• ## US weighs banning laptop computers on in

A bombing at a concert in Mancheste [...]