Další formáty:
BibTeX
LaTeX
RIS
@inproceedings{2392233,
author = {Chau, Calvin and Křetínský, Jan and Mohr, Stefanie},
address = {Singapore},
booktitle = {Automated Technology for Verification and Analysis. ATVA 2023},
doi = {http://dx.doi.org/10.1007/978-3-031-45329-8_19},
keywords = {Abstraction; Machine learning; Neural network},
howpublished = {elektronická verze "online"},
language = {eng},
location = {Singapore},
isbn = {978-3-031-45328-1},
pages = {401-421},
publisher = {Springer},
title = {Syntactic vs Semantic Linear Abstraction and Refinement of Neural Networks},
year = {2023}
}
TY - JOUR
ID - 2392233
AU - Chau, Calvin - Křetínský, Jan - Mohr, Stefanie
PY - 2023
TI - Syntactic vs Semantic Linear Abstraction and Refinement of Neural Networks
PB - Springer
CY - Singapore
SN - 9783031453281
KW - Abstraction
KW - Machine learning
KW - Neural network
N2 - Abstraction is a key verification technique to improve scalability. However, its use for neural networks is so far extremely limited. Previous approaches for abstracting classification networks replace several neurons with one of them that is similar enough. We can classify the similarity as defined either syntactically (using quantities on the connections between neurons) or semantically (on the activation values of neurons for various inputs). Unfortunately, the previous approaches only achieve moderate reductions, when implemented at all. In this work, we provide a more flexible framework, where a neuron can be replaced with a linear combination of other neurons, improving the reduction. We apply this approach both on syntactic and semantic abstractions, and implement and evaluate them experimentally. Further, we introduce a refinement method for our abstractions, allowing for finding a better balance between reduction and precision.
ER -
CHAU, Calvin, Jan KŘETÍNSKÝ a Stefanie MOHR. Syntactic vs Semantic Linear Abstraction and Refinement of Neural Networks. Online. In \textit{Automated Technology for Verification and Analysis. ATVA 2023}. Singapore: Springer, 2023, s.~401-421. ISBN~978-3-031-45328-1. Dostupné z: https://dx.doi.org/10.1007/978-3-031-45329-8\_{}19.
|
