Deep neural networks (DNNs) are the workhorses of deep learning, which constitutes the state of the art in numerous application domains. However, DNN-based decision rules are notoriously prone to poor generalization, i.e., may prove inadequate on inputs not encountered during training. This limitation poses a significant obstacle to employing deep learning for mission-critical tasks, and also in real-world environments that exhibit high variability. We propose a novel, verification-driven methodology for identifying DNN-based decision rules that generalize well to new input domains. Our approach quantifies generalization to an input domain by the extent to which decisions reached by independently trained DNNs are in agreement for inputs in this domain. We show how, by harnessing the power of DNN verification, our approach can be efficiently and effectively realized. We evaluate our verification-based approach on three deep reinforcement learning (DRL) benchmarks, including a system for Internet congestion control. Our results establish the usefulness of our approach. More broadly, our work puts forth a novel objective for formal verification, with the potential for mitigating the risks associated with deploying DNN-based systems in the wild.
|Original language||American English|
|Title of host publication||Computer Aided Verification - 35th International Conference, CAV 2023, Proceedings|
|Editors||Constantin Enea, Akash Lal|
|Publisher||Springer Science and Business Media Deutschland GmbH|
|Number of pages||18|
|State||Published - 2023|
|Event||Proceedings of the 35th International Conference on Computer Aided Verification, CAV 2023 - Paris, France|
Duration: 17 Jul 2023 → 22 Jul 2023
|Name||Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)|
|Conference||Proceedings of the 35th International Conference on Computer Aided Verification, CAV 2023|
|Period||17/07/23 → 22/07/23|
Bibliographical noteFunding Information:
The work of Amir, Zelazny, and Katz was partially supported by the Israel Science Foundation (grant number 683/18). The work of Amir was supported by a scholarship from the Clore Israel Foundation. The work of Maayan and Schapira was partially supported by funding from Huawei.
© 2023, The Author(s).