dnnf - v0.1.7

Use the latest DNNV release.

- Python
Published by dlshriver almost 4 years ago

dnnf - v0.1.6

This release fixes a bug which required interval constraints for all inputs, even if the input region was a bounded polytope. It also adds a new custom falsification method based on the newton-raphson method root finding method.

- Python
Published by dlshriver almost 4 years ago

dnnf - v0.1.5

Small bugfixes and improvements to code quality. We have adopted black and isort as part of our code formatting style requirements. We have also updated the installation script and Dockerfile. The Dockerfile now installs the verification and falsification benchmarks from the dlshriver/dnnv-benchmarks repo.

Fixes: - At some point a change to dnnv broke multi-processing in dnnf. This release should fix that issue (fixed in 3d21eeb). - The pytorch converter's flatten implementation could return incorrect shapes due to incorrect computation of the new shape. This was fixed in b40f1f7. - Split operations do not yet exist in dnnv, which could cause dnnf to fail to run. This was fixed in 674082a.

- Python
Published by dlshriver about 4 years ago

dnnf - v0.1.4

Enhancement to pytorch converter to support group parameter for Conv operators (contributed by @MissMeriel).

- Python
Published by dlshriver about 4 years ago

dnnf - v0.1.3

Fixes to work with the latest DNNV version. Updates dependency versions and now show the available backends with the --long-help option.

- Python
Published by dlshriver over 4 years ago

dnnf - v0.1.2

Fixes some bugs from upgrading DNNV to version 0.5.0

- Python
Published by dlshriver over 4 years ago

dnnf - v0.1.1

Update to use DNNV v0.5.0.

- Python
Published by dlshriver over 4 years ago

dnnf - v0.1.0

• Updated install process
• Support mixed input/output constraints (support for full reduction from paper)
• Updates for VNN-Comp
• Support VNN-LIB

- Python
Published by dlshriver almost 5 years ago