Leveraging the Graph Structure of Neural Network Training Dynamics

Vahedian, Fatemeh; Li, Ruiyu; Trivedi, Puja; Jin, Di; Koutra, Danai

doi:10.1145/3511808.3557628

Cited by 2 publications

(2 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They proposed "Unroll", and converting convolutional layers into graphs. Then, Vahedian et al [34] proposed a "Rolled" graph representation of convolutional layers to solve the DNN performance prediction problem by capturing the early DNN dynamics during the training phase. To maintain the semantic meaning of the convolutional layers, they represent each filter as a node and link the filters in successive layers by weighted edges.…”

Section: Graph Structure Of Dnnsmentioning

confidence: 99%

See 1 more Smart Citation

ActGraph: prioritization of test cases based on deep neural network activation graph

Chen,

Ge,

Zheng

2023

Autom Softw Eng

View full text Add to dashboard Cite

Widespread applications of deep neural networks (DNNs) benefit from DNN testing to guarantee their quality. In the DNN testing, numerous test cases are fed into the model to explore potential vulnerabilities, but they require expensive manual cost to check the label. Therefore, test case prioritization is proposed to solve the problem of labeling cost, e.g., activation-based and mutation-based prioritization methods. However, most of them suffer from limited scenarios (i.e. high confidence adversarial or false positive cases) and high time complexity. To address these challenges, we propose the concept of the activation graph from the perspective of the spatial relationship of neurons. We observe that the activation graph of cases that triggers the model's misbehavior significantly differs from that of normal cases. Motivated by it, we design a test case prioritization method based on the activation graph, ActGraph, by extracting the high-order node features of the activation graph for prioritization. ActGraph explains the difference between the test cases to solve the problem of scenario limitation. Without mutation operations, ActGraph is easy to implement, leading to lower time complexity. Extensive experiments on three datasets and four models demonstrate that ActGraph has the following key characteristics. (i) Effectiveness and generalizability: ActGraph shows competitive performance in all of the natural, adversarial and mixed scenarios, especially in RAUC-100 improvement (∼ ×1.40). (ii) Efficiency: Act-Graph does not use complex mutation operations and runs in less time ActGraph (∼ ×1/50) than the state-of-the-art method. The code of ActGraph is open-sourced at https:// github.com/ Embed-Debuger/ ActGraph.

show abstract

Section: Graph Structure Of Dnnsmentioning

confidence: 99%

“…Existing DAG studies have been able to convert the DNN into the graph, discussing the dynamic properties of the DNN model [32][33][34][35]. They used an undirected weighted graph to treat neurons as nodes and the weights of the model as weighted edges between nodes.…”

Section: Overviewmentioning

confidence: 99%

ActGraph: prioritization of test cases based on deep neural network activation graph

Chen,

Ge,

Zheng

2023

Autom Softw Eng

View full text Add to dashboard Cite

show abstract

ActGraph: Prioritization of Test Cases Based on Deep Neural Network Activation Graph

Chen

Zheng

2023

Preprint

View full text Add to dashboard Cite

Widespread applications of deep neural networks (DNNs) benefit fromDNN testing to guarantee their quality. In the DNN testing, numeroustest cases are fed into the model to explore potential vulnerabilities,but they require expensive manual cost to check the label. Therefore,test case prioritization is proposed to solve the problem of labeling cost,e.g., activation-based and mutation-based prioritization methods. How-ever, most of them suffer from limited scenarios (i.e. high confidenceadversarial or false positive cases) and high time complexity. To addressthese challenges, we propose the concept of the activation graph fromthe perspective of the spatial relationship of neurons. We observe thatthe activation graph of cases that triggers the model’s misbehaviorsignificantly differs from that of normal cases. Motivated by it, wedesign a test case prioritization method based on the activation graph,ActGraph, by extracting the high-order node features of the activationgraph for prioritization. ActGraph explains the difference between thetest cases to solve the problem of scenario limitation. Without muta-tion operations, ActGraph is easy to implement, leading to lower timecomplexity. Extensive experiments on three datasets and four mod-els demonstrate that ActGraph has the following key characteristics.(i) Effectiveness and generalizability: ActGraph shows competitiveperformance in all of the natural, adversarial and mixed scenarios,especially in RAUC-100 improvement (∼ ×1.40). (ii) Efficiency: Act-Graph does not use complex mutation operations and runs in less time (∼ ×1/50) than the state-of-the-art method. The code of ActGraph is open-sourced at https:// github.com/ Embed-Debuger/ ActGraph.

show abstract

Leveraging the Graph Structure of Neural Network Training Dynamics

Cited by 2 publications

References 11 publications

ActGraph: prioritization of test cases based on deep neural network activation graph

ActGraph: prioritization of test cases based on deep neural network activation graph

ActGraph: Prioritization of Test Cases Based on Deep Neural Network Activation Graph

Contact Info

Product

Resources

About