Learning-Based Efficient Graph Similarity Computation via Multi-Scale Convolutional Set Matching

Bai, Yunsheng; Ding, Hao; Gu, Ken; Sun, Yizhou; Wang, Wei

doi:10.1609/aaai.v34i04.5720

Cited by 55 publications

(67 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The second category is the autoregressive method, which constructs a solution by iteratively extending a partial solution to obtain a solution of the CO problem. Table 2 MDS, MM, MVC GNN, non-autoregressive GAP [49] Graph partition GNN, non-autoregressive GMN [41] GED GNN, non-autoregressive SimGNN [2] GED GNN, non-autoregressive GRAPHSIM [3] GED GNN, non-autoregressive GNNGC [40] GColor GNN, non-autogressive SiameseGNN [51] Graph matching, TSP GNN, non-autogressive PCAGM [69] Graph matching GNN, non-autogressive IsoNN [44] Graph Iso. AutoEncoder, non-autogressive GNNTS [42] MIS, MVC, MC GNN, non-autoregressive Ptr-Net [67] TSP AutoEncoder, autoregressive LSTMGMatching [46] Graph matching AutoEncoder, autogressive S2V-DQN [14] MVC, MaxCut, TSP GNN, autoregressive CombOptZero [1] MVC, MaxCut, MC GNN, autoregressive RLMCS [4] MCS GNN, autoregressive CENALP [19] Graph alignment SkipGram, autoregressive TSPImprove [71] TSP AutoEncoder, autoregressive AM [37] TSP AutoEncoder, autoregressive…”

Section: Graph Learning-based Combinatorial Optimization Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Graph Learning for Combinatorial Optimization: A Survey of State-of-the-Art

Peng

Choi

2021

Data Sci. Eng.

View full text Add to dashboard Cite

Graphs have been widely used to represent complex data in many applications, such as e-commerce, social networks, and bioinformatics. Efficient and effective analysis of graph data is important for graph-based applications. However, most graph analysis tasks are combinatorial optimization (CO) problems, which are NP-hard. Recent studies have focused a lot on the potential of using machine learning (ML) to solve graph-based CO problems. Most recent methods follow the two-stage framework. The first stage is graph representation learning, which embeds the graphs into low-dimension vectors. The second stage uses machine learning to solve the CO problems using the embeddings of the graphs learned in the first stage. The works for the first stage can be classified into two categories, graph embedding methods and end-to-end learning methods. For graph embedding methods, the learning of the the embeddings of the graphs has its own objective, which may not rely on the CO problems to be solved. The CO problems are solved by independent downstream tasks. For end-to-end learning methods, the learning of the embeddings of the graphs does not have its own objective and is an intermediate step of the learning procedure of solving the CO problems. The works for the second stage can also be classified into two categories, non-autoregressive methods and autoregressive methods. Non-autoregressive methods predict a solution for a CO problem in one shot. A non-autoregressive method predicts a matrix that denotes the probability of each node/edge being a part of a solution of the CO problem. The solution can be computed from the matrix using search heuristics such as beam search. Autoregressive methods iteratively extend a partial solution step by step. At each step, an autoregressive method predicts a node/edge conditioned to current partial solution, which is used to its extension. In this survey, we provide a thorough overview of recent studies of the graph learning-based CO methods. The survey ends with several remarks on future research directions.

show abstract

Section: Graph Learning-based Combinatorial Optimization Methodsmentioning

confidence: 99%

“…The mean squared error between the predicted similarity with the ground truth is used as the loss of SimGNN. In the follow-up work GRAPHSIM [3], a CNN-based method is used to replace the histogram of SimGNN.…”

Section: B Graph Partitionmentioning

confidence: 99%

Graph Learning for Combinatorial Optimization: A Survey of State-of-the-Art

Peng

Choi

2021

Data Sci. Eng.

View full text Add to dashboard Cite

show abstract

“…By training the model to produce embeddings Z GQ and Z GT such that F (Z GQ , Z GT ) ≈ SED(Z GQ , Z GT ), we enforce a rich structure on the embedding space. We show that these embeddings satisfy many key properties of the SED (and GED) function that existing neural algorithms fail to do [2,47,3,53]. F is defined as follows:…”

Section: Pre-mlpmentioning

confidence: 98%

“…To mitigate this computational bottleneck, several heuristics [9,14,19,42,38] and index structures [22,51,35,54] have been proposed. Recently, graph neural networks have been shown to be effective in learning and predicting GED [2,47,34,3,53,48] and subgraph isomorphism [37]. The basic goal in all these algorithms is to learn a neural model from a training set of graph pairs and their distances, such that, at inference time, given an unseen graph pair, we are able to predict its distance accurately.…”

Section: Introduction and Related Workmentioning

confidence: 99%

A Neural Framework for Learning Subgraph and Graph Similarity Measures

Sharma¹,

Grover²,

Medya³

et al. 2021

Preprint

View full text Add to dashboard Cite

Subgraph similarity search is a fundamental operator in graph analysis. In this framework, given a query graph and a graph database, the goal is to identify subgraphs of the database graphs that are structurally similar to the query. Subgraph edit distance (SED) is one of the most expressive measures for subgraph similarity. In this work, we study the problem of learning SED from a training set of graph pairs and their SED values. Towards that end, we design a novel siamese graph neural network called NEUROSED, which learns an embedding space with a rich structure reminiscent of SED. With the help of a specially crafted inductive bias, NEUROSED not only enables high accuracy but also ensures that the predicted SED, like true SED, satisfies triangle inequality. The design is generic enough to also model graph edit distance (GED), while ensuring that the predicted GED space is metric, like the true GED space. Extensive experiments on real graph datasets, for both SED and GED, establish that NEUROSED achieves ≈ 2 times lower RMSE than the state of the art and is ≈ 18 times faster than the fastest baseline. Further, owing to its pair-independent embeddings and theoretical properties, NEUROSED allows ≈ 3 orders of magnitude faster retrieval of graphs and subgraphs.Preprint. Under review.

show abstract

“…However, this method can only works on x_86 platform. Bai et al [26] propose a graph similarity comparison method directly based on node embedding instead of the widely used graph-level embedding. It is a general framework for similarity computation which can work with other models, and does not take semantic feature of each node into consideration.…”

Section: Introductionmentioning

confidence: 99%

BEDetector: A Two-Channel Encoding Method to Detect Vulnerabilities Based on Binary Similarity

Shen

et al. 2021

IEEE Access

View full text Add to dashboard Cite

Applying neural network technology to binary similarity detection has become a promising search topic, and vulnerability detection is an important application field of binary similarity detection. When embedding binary code into matrix by neural network, the problem of feature representation also needs to be solved in vulnerability detection. However, most of the current researches extract the syntax or structural features of binary code, and take basic block as the minimum analysis unit, which is relatively coarse. In addition, the structural features of binary functions are usually represented by the dependency graph. In the embedding process, only the neighbour information of the node can be obtained, ignoring the global information of the graph. To solve these two problems, we propose a two-channel feature extraction method to obtain semantic feature in finer granularity and represent the structural features globally instead of locally. Inspired by natural language process, we propose a contextual semantic feature extraction method to obtain different granularity features of binary functions. It takes instruction as the minimum analysis unit and obtains the semantic relationship between instructions. Meanwhile, in order to represent the structural feature of each function, we propose a neural GAE model instead of the widely used structure2vec model. In this way, we can preserve and reconstruct the control dependencies between the basic blocks in the whole graph. We have implemented a prototype system BEDetector, evaluated the effectiveness of its neural model and compared the accuracy of vulnerability function detection with state-of-the-art system. Besides, we choose the real-world firmware files as the detection target and prove that BEDetector can achieve a relatively high detection rate. BEDetector could reach a precision of 88.8%, 86.7% and 100% when ranking top-50 candidate functions in the detection of the CVE vulnerability function ssl3_get_key_exchange, ssl3_get_new_session_ticket and udhcp_get_option, proving the efficiency of our method.

show abstract

Learning-Based Efficient Graph Similarity Computation via Multi-Scale Convolutional Set Matching

Cited by 55 publications

References 19 publications

Graph Learning for Combinatorial Optimization: A Survey of State-of-the-Art

Graph Learning for Combinatorial Optimization: A Survey of State-of-the-Art

A Neural Framework for Learning Subgraph and Graph Similarity Measures

BEDetector: A Two-Channel Encoding Method to Detect Vulnerabilities Based on Binary Similarity

Contact Info

Product

Resources

About