Deep neural network (DNN) verification is an emerging field, with diverse verification engines quickly becoming available. Demonstrating the effectiveness of these engines on real-world DNNs is an important step towards their wider adoption. We present a tool that can leverage existing verification engines in performing a novel application: neural network simplification, through the reduction of the size of a DNN without harming its accuracy. We report on the work-flow of the simplification process, and demonstrate its potential significance and applicability on a family of real-world DNNs for aircraft collision avoidance, whose sizes we were able to reduce by as much as 10%.
|Original language||American English|
|Title of host publication||NASA Formal Methods - 12th International Symposium, NFM 2020, Proceedings|
|Editors||Ritchie Lee, Susmit Jha, Anastasia Mavridou|
|Number of pages||9|
|State||Published - 2020|
|Event||12th International Symposium on NASA Formal Methods, NFM 2020 - Moffett Field, United States|
Duration: 11 May 2020 → 15 May 2020
|Name||Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)|
|Conference||12th International Symposium on NASA Formal Methods, NFM 2020|
|Period||11/05/20 → 15/05/20|
Bibliographical noteFunding Information:
Acknowledgements. This project was partially supported by grants from the Binational Science Foundation (2017662), the Israel Science Foundation (683/18), and the National Science Foundation (1814369).
© 2020, Springer Nature Switzerland AG.
- Deep neural networks