Scalable Polyhedral Verification of Recurrent Neural Networks

Ryou, Wonryong; Chen, Jiayu; Balunović, Mislav; Singh, Gagandeep; Dan, Andrei; Vechev, Martin

doi:10.1007/978-3-030-81685-8_10

Cited by 26 publications

(15 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We leave as future work a further investigation of variations of ADMM that can improve convergence rates in deep learning-sized problem instances, as well as extensions beyond the LP setting. Furthermore, it would be interesting to extend the proposed method to verification of recurrent neural neworks (RNNs) such as vanilla RNNs, LSTMs 9 , and GRUs 10 [46], [47], [48]. where T f (x) ∈ ∂f (x) denotes a subgradient.…”

Section: Discussionmentioning

confidence: 99%

DeepSplit: Scalable Verification of Deep Neural Networks via Operator Splitting

Chen

Wong

Kolter

et al. 2022

IEEE Open J. Control. Syst.

View full text Add to dashboard Cite

Analyzing the worst-case performance of deep neural networks against input perturbations amounts to solving a large-scale non-convex optimization problem, for which several past works have proposed convex relaxations as a promising alternative. However, even for reasonably-sized neural networks, these relaxations are not tractable, and so must be replaced by even weaker relaxations in practice. In this work, we propose a novel operator splitting method that can directly solve a convex relaxation of the problem to high accuracy, by splitting it into smaller sub-problems that often have analytical solutions. The method is modular, scales to very large problem instances, and compromises of operations that are amenable to fast parallelization with GPU acceleration. We demonstrate our method in bounding the worst-case performance of large convolutional networks in image classification and reinforcement learning settings, and in reachability analysis of neural network dynamical systems.

show abstract

Section: Discussionmentioning

confidence: 99%

DeepSplit: Scalable Verification of Deep Neural Networks via Operator Splitting

Chen

Wong

Kolter

et al. 2022

IEEE Open J. Control. Syst.

View full text Add to dashboard Cite

show abstract

“…Following that they represent all the perturbation sets as a hyperrectangle and pass the hyperrectangle through the remaining network using IBP technique [29]. Following the similar direction the work in [88] presents Polyhedral Robustness Verifier of RNNs (PROVER) which represents the perturbations in input data in the form of polyhedral which is passed through a LSTM network to obtain a certifiable verified network for a more general sequential data. In this line [26] proposed robust certified abstract transformer AI2 using zonotopes as abstract representation of input perturbation where transformer minimize the zonotope projections to achieve certified robustness.…”

Section: Robustness By Certificationmentioning

confidence: 99%

A survey in Adversarial Defences and Robustness in NLP

Goyal¹,

Doddapaneni²,

Khapra³

et al. 2022

Preprint

View full text Add to dashboard Cite

In recent years, it has been seen that deep neural networks are lacking robustness and are likely to break in case of adversarial perturbations in input data. Strong adversarial attacks are proposed by various authors for computer vision and Natural Language Processing (NLP). As a counter-effort, several defense mechanisms are also proposed to save these networks from failing. In contrast with image data, generating adversarial attacks and defending these models is not easy in NLP because of the discrete nature of the text data. However, numerous methods for adversarial defense are proposed of late, for different NLP tasks such as text classification, named entity recognition, natural language inferencing, etc. These methods are not just used for defending neural networks from adversarial attacks, but also used as a regularization mechanism during training, saving the model from overfitting. The proposed survey is an attempt to review different methods proposed for adversarial defenses in NLP in the recent past by proposing a novel taxonomy. This survey also highlights the fragility of the advanced deep neural networks in NLP and the challenges in defending them.

show abstract

“…We also validate our approaches on more complex benchmarks compared with the aforementioned work. Related to abstraction refinement, [25,32] uses the post condition to refine the choice of slopes in forward over-approximations but do not consider the post condition as hard constraint. [37] combines forward abstract interpretation with MILP/LP solving for refinement, but only considers the pre-condition and the refinement is due to the precision gain of the MILP/LP solving.…”

Section: Related Workmentioning

confidence: 99%

“…To produce node, job, and global embeddings, we used neural networks with six fullyconnected layers each, containing [16,8,8,16,8,8] neurons in their layers, respectively. Finally, two neural network with four fully-connected layers, containing [32,16,8,1] neurons respectively, mapped the embeddings to actions-i.e., a node selected for scheduling and a number of executors to assign. Training was performed using the REINFORCE policy-gradient algorithm executed on 16 workers.…”

Section: G Trainingmentioning

confidence: 99%

Scalable Verification of GNN-based Job Schedulers

Wu,

Barrett,

Sharif

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Recently, Graph Neural Networks (GNNs) have been applied for scheduling jobs over clusters achieving better performance than hand-crafted heuristics. Despite their impressive performance, concerns remain over their trustworthiness when deployed in a real-world environment due to their black-box nature. To address these limitations, we consider formal verification of their expected properties such as strategy proofness and locality in this work. We address several domain specific challenges such as deeper networks and richer specifications not encountered by existing verifiers for image and NLP classifiers. We develop, GNN-Verify, the first general framework for verifying both single-step and multi-step properties of these schedulers based on carefully designed algorithms that combine abstractions, refinements, solvers, and proof transfer. Our experimental results on challenging benchmarks shows that our approach can provide precise and scalable formal guarantees on the trustworthiness of state-of-the-art GNN-based scheduler.

show abstract

Scalable Polyhedral Verification of Recurrent Neural Networks

Cited by 26 publications

References 22 publications

DeepSplit: Scalable Verification of Deep Neural Networks via Operator Splitting

DeepSplit: Scalable Verification of Deep Neural Networks via Operator Splitting

A survey in Adversarial Defences and Robustness in NLP

Scalable Verification of GNN-based Job Schedulers

Contact Info

Product

Resources

About