VNN-COMP & ARCH-COMP¶
NNV participates in two major annual verification competitions, contributing to community benchmarking and tool evaluation.
VNN-COMP (Verification of Neural Networks Competition)¶
Year |
Contribution |
Source |
|---|---|---|
2025 |
Star set reachability approach (competition contribution) |
|
2024 |
Full benchmark participation with multi-solver fallback |
|
2023 |
||
2021 |
NNV will participate in both VNN-COMP 2026 and ARCH-COMP 2026.
Competition strategy: NNV uses a three-phase approach:
Random falsification (100 samples) for quick counterexample search
Reachability analysis (approx-star) if no counterexample found
Multiple LP solver fallback (Gurobi → GLPK → linprog)
ARCH-COMP (Applied Verification of Continuous and Hybrid Systems)¶
NNV participates in the AINNCS (Artificial Intelligence and Neural Network Control Systems) category:
Year |
Report |
Source |
|---|---|---|
2025 |
||
2024 |
||
2023 |
||
2022 |
||
2021 |
||
2020 |
||
2019 |
Initial participation |
Other Conference Submissions¶
NNV has reproducible code for publications at:
Venue |
Year(s) |
Topics |
|---|---|---|
CAV |
2020, 2021, 2023 |
Core tool papers, ImageStar, semantic segmentation |
FM |
2019 |
Star-based reachability for DNNs |
FormaliSE |
2023, 2024, 2025 |
BNNs, malware detection, video verification |
FMICS |
2023 |
Time-dependent neural networks |
ICAIF |
2024 |
FairNNV fairness verification |
HSCC |
2022, 2023 |
RNN verification, evaluation methods |
NeurIPS |
2025 |
Probabilistic semantic segmentation verification |
All submission code is available in examples/Submission/.