Kovásznai, Gergely and Kiss, Dorina Hedvig and Mlinkó, Péter (2023) Formal verification for quantized neural networks. Annales Mathematicae et Informaticae, 57. pp. 36-48. ISSN 1787-6117
|
Text
36_48.pdf - Published Version Download (868kB) | Preview |
Abstract
Despite of deep neural networks are being successfully used in many fields of computing, it is still challenging to verify their trustiness. Previously it has been shown that binarized neural networks can be verified by being encoded into Boolean constraints. In this paper, we generalize this encoding to quantized neural networks (QNNs). We demonstrate how to implement QNNs in Python, using the Tensorflow and Keras libraries. Also, we demonstrate how to implement a Boolean encoding of QNNs, as part of our tool that is able to run a variety of solvers to verify QNNs.
| Item Type: | Article |
|---|---|
| Uncontrolled Keywords: | Artificial Intelligence, Deep Learning, Neural Network, Formal Verification, SAT, SMT, Constraint Programming, Python, Keras |
| Subjects: | Q Science / természettudomány > QA Mathematics / matematika |
| Depositing User: | Tibor Gál |
| Date Deposited: | 11 Aug 2023 10:09 |
| Last Modified: | 11 Aug 2023 10:12 |
| URI: | http://real.mtak.hu/id/eprint/171291 |
Actions (login required)
 |
Edit Item |
