TY - GEN
T1 - Approximate Bisimulation Relations for Neural Networks and Application to Assured Neural Network Compression
AU - Xiang, Weiming
AU - Shao, Zhongzhu
N1 - Funding Information:
This research was supported by the National Science Foundation, under NSF CAREER Award 2143351. W. Xiang is with the School of Computer and Cyber Sciences, Augusta University, Augusta GA 30912 USA. Email: wxiang@augusta.edu Z. Shao is with Department of Electrical Engineering, Southwest Jiaotong University, Chengdu, China.
Publisher Copyright:
© 2022 American Automatic Control Council.
PY - 2022
Y1 - 2022
N2 - In this paper, we propose a concept of approximate bisimulation relation for feedforward neural networks. In the framework of approximate bisimulation relation, a novel neural network merging method is developed to compute the approximate bisimulation error between two neural networks based on reachability analysis of neural networks. The developed method is able to quantitatively measure the distance between the outputs of two neural networks with the same inputs. Then, we apply the approximate bisimulation relation results to perform neural networks model reduction and compute the compression precision, i.e., assured neural networks compression. At last, using the assured neural network compression, we accelerate the verification processes of ACAS Xu neural networks to illustrate the effectiveness and advantages of our proposed approximate bisimulation approach.
AB - In this paper, we propose a concept of approximate bisimulation relation for feedforward neural networks. In the framework of approximate bisimulation relation, a novel neural network merging method is developed to compute the approximate bisimulation error between two neural networks based on reachability analysis of neural networks. The developed method is able to quantitatively measure the distance between the outputs of two neural networks with the same inputs. Then, we apply the approximate bisimulation relation results to perform neural networks model reduction and compute the compression precision, i.e., assured neural networks compression. At last, using the assured neural network compression, we accelerate the verification processes of ACAS Xu neural networks to illustrate the effectiveness and advantages of our proposed approximate bisimulation approach.
UR - http://www.scopus.com/inward/record.url?scp=85138495295&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85138495295&partnerID=8YFLogxK
U2 - 10.23919/ACC53348.2022.9867845
DO - 10.23919/ACC53348.2022.9867845
M3 - Conference contribution
AN - SCOPUS:85138495295
T3 - Proceedings of the American Control Conference
SP - 3248
EP - 3253
BT - 2022 American Control Conference, ACC 2022
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2022 American Control Conference, ACC 2022
Y2 - 8 June 2022 through 10 June 2022
ER -