Search

Deep Neural Networks (DNNs) have emerged as an effective approach to tackling real-world problems. However, like human-written software, automatically-generated DNNs can have bugs and be attacked. This thus attracts many recent interests in developing effective and scalable DNN verification techniques and tools. In this work, we introduce a NeuralSAT, a new constraint solving approach to DNN verification. The design of NeuralSAT follows the DPLL(T) algorithm used modern SMT solving, which includes (conflict) clause learning, abstraction, and theory solving, and thus NeuralSAT can be considered as an SMT framework for DNNs. Preliminary results show that the NeuralSAT prototype is competitive to the state-of-the-art. We hope, with proper optimization and engineering, NeuralSAT will carry the power and success of modern SAT/SMT solvers to DNN verification. NeuralSAT is avaliable from: https://github.com/dynaroars/neuralsat-solver, 27 pages, 8 figures. NeuralSAT is avaliable from: https://github.com/dynaroars/neuralsat-solver

3D vision with real-time LiDAR-based point cloud data became a vital part of autonomous system research, especially perception and prediction modules use for object classification, segmentation, and detection. Despite their success, point cloud-based network models are vulnerable to multiple adversarial attacks, where the certain factor of changes in the validation set causes significant performance drop in well-trained networks. Most of the existing verifiers work perfectly on 2D convolution. Due to complex architecture, dimension of hyper-parameter, and 3D convolution, no verifiers can perform the basic layer-wise verification. It is difficult to conclude the robustness of a 3D vision model without performing the verification. Because there will be always corner cases and adversarial input that can compromise the model's effectiveness. In this project, we describe a point cloud-based network verifier that successfully deals state of the art 3D classifier PointNet verifies the robustness by generating adversarial inputs. We have used extracted properties from the trained PointNet and changed certain factors for perturbation input. We calculate the impact on model accuracy versus property factor and can test PointNet network's robustness against a small collection of perturbing input states resulting from adversarial attacks like the suggested hybrid reverse signed attack. The experimental results reveal that the resilience property of PointNet is affected by our hybrid reverse signed perturbation strategy, 11 pages, 12 figures

The field of software verification has produced a wide array of algorithmic techniques that can prove a variety of properties of a given program. It has been demonstrated that the performance of these techniques can vary up to 4 orders of magnitude on the same verification problem. Even for verification experts, it is difficult to decide which tool will perform best on a given problem. For general users, deciding the best tool for their verification problem is effectively impossible. In this work, we present Graves, a selection strategy based on graph neural networks (GNNs). Graves generates a graph representation of a program from which a GNN predicts a score for a verifier that indicates its performance on the program. We evaluate Graves on a set of 10 verification tools and over 8000 verification problems and find that it improves the state-of-the-art in verification algorithm selection by 11\%. We conjecture this is in part due to Graves' use of GNNs with attention mechanisms. Through a qualitative study on model interpretability, we find strong evidence that the Graves' GNN-based model learns to base its predictions on factors that relate to the unique features of the algorithmic techniques., 29 pages, 7 figures, 5 tables, submitted to ACM Transactions on Software Engineering and Methodology; Updated Symbiotic reference and description in Table 1

Recent literature has seen growing interest in using black-box strategies like CheckList for testing the behavior of NLP models. Research on white-box testing has developed a number of methods for evaluating how thoroughly the internal behavior of deep models is tested, but they are not applicable to NLP models. We propose a set of white-box testing methods that are customized for transformer-based NLP models. These include Mask Neuron Coverage (MNCOVER) that measures how thoroughly the attention layers in models are exercised during testing. We show that MNCOVER can refine testing suites generated by CheckList by substantially reduce them in size, for more than 60\% on average, while retaining failing tests -- thereby concentrating the fault detection power of the test suite. Further we show how MNCOVER can be used to guide CheckList input generation, evaluate alternative NLP testing methods, and drive data augmentation to improve accuracy., Findings of NAACL 2022 submission, 12 pages

Code coverage is a popular and widespread test adequacy metric that measures the percentage of program codes executed by a test suite. Despite its popularity, code coverage has several limitations. One of the major limitations is that it does not provide any insights into the quality or quantity of test oracles, a core component of testing. Due to this limitation, several studies have suggested that coverage is a poor test adequacy metric; therefore, it should not be used as an indicator of a test suite's fault detection effectiveness. To address this limitation, researchers have proposed extensions to traditional structural code coverage to explicitly consider the quality of test oracles. We refer to these extensions as oracle-based code coverage. This survey paper studies oracle-based coverage techniques published since their inception in 2007. We discuss each metric's definition, methodology, experimental studies, and research findings. Even though oracle-based coverage metrics are proven to be more effective than traditional coverage in detecting faults, they have received little attention in the software engineering community. We present all existing oracle-based adequacy metrics in this paper and compare the critical features against each other. We observe that different oracle-based adequacy metrics operate on different coverage domains and use diverse underlying analysis techniques, enabling a software tester to choose the appropriate metric based on the testing requirements. Our paper provides valuable information regarding the limitations of oracle-based methods, addressing which may help their broader adoption in software testing automation., Comment: 7 pages

Despite the large number of sophisticated deep neural network (DNN) verification algorithms, DNN verifier developers, users, and researchers still face several challenges. First, verifier developers must contend with the rapidly changing DNN field to support new DNN operations and property types. Second, verifier users have the burden of selecting a verifier input format to specify their problem. Due to the many input formats, this decision can greatly restrict the verifiers that a user may run. Finally, researchers face difficulties in re-using benchmarks to evaluate and compare verifiers, due to the large number of input formats required to run different verifiers. Existing benchmarks are rarely in formats supported by verifiers other than the one for which the benchmark was introduced. In this work we present DNNV, a framework for reducing the burden on DNN verifier researchers, developers, and users. DNNV standardizes input and output formats, includes a simple yet expressive DSL for specifying DNN properties, and provides powerful simplification and reduction operations to facilitate the application, development, and comparison of DNN verifiers. We show how DNNV increases the support of verifiers for existing benchmarks from 30% to 74%.

Monitoring programs for finite state properties is challenging due to high memory and execution time overheads it incurs. Some events if skipped or lost naturally can reduce both overheads, but lead to uncertainty about the current monitor state. In this work, we present a theoretical framework to model these lossy event streams and provide a construction for a monitor which observes them without producing false positives. The constructed monitor is optimally sound among all complete monitors. We model several loss types of practical relevance using our framework and provide construction of smaller approximate monitors for properties with a large number of states.

Given the current transformative potential of research that sits at the intersection of Deep Learning (DL) and Software Engineering (SE), an NSF-sponsored community workshop was conducted in co-location with the 34th IEEE/ACM International Conference on Automated Software Engineering (ASE'19) in San Diego, California. The goal of this workshop was to outline high priority areas for cross-cutting research. While a multitude of exciting directions for future work were identified, this report provides a general summary of the research areas representing the areas of highest priority which were discussed at the workshop. The intent of this report is to serve as a potential roadmap to guide future work that sits at the intersection of SE & DL., Community Report from the 2019 NSF Workshop on Deep Learning & Software Engineering, 37 pages

Deep neural networks (DNN) are growing in capability and applicability. Their effectiveness has led to their use in safety critical and autonomous systems, yet there is a dearth of cost-effective methods available for reasoning about the behavior of a DNN. In this paper, we seek to expand the applicability and scalability of existing DNN verification techniques through DNN refactoring. A DNN refactoring defines (a) the transformation of the DNN's architecture, i.e., the number and size of its layers, and (b) the distillation of the learned relationships between the input features and function outputs of the original to train the transformed network. Unlike with traditional code refactoring, DNN refactoring does not guarantee functional equivalence of the two networks, but rather it aims to preserve the accuracy of the original network while producing a simpler network that is amenable to more efficient property verification. We present an automated framework for DNN refactoring, and demonstrate its potential effectiveness through three case studies on networks used in autonomous systems.

We introduce a new technique for inferring program invariants that uses symbolic states generated by symbolic execution. Symbolic states, which consist of path conditions and constraints on local variables, are a compact description of sets of concrete program states and they can be used for both invariant inference and invariant verification. Our technique uses a counterexample-based algorithm that creates concrete states from symbolic states, infers candidate invariants from concrete states, and then verifies or refutes candidate invariants using symbolic states. The refutation case produces concrete counterexamples that prevent spurious results and allow the technique to obtain more precise invariants. This process stops when the algorithm reaches a stable set of invariants. We present SymInfer, a tool that implements these ideas to automatically generate invariants at arbitrary locations in a Java program. The tool obtains symbolic states from Symbolic PathFinder and uses existing algorithms to infer complex (potentially nonlinear) numerical invariants. Our preliminary results show that SymInfer is effective in using symbolic states to generate precise and useful invariants for proving program safety and analyzing program runtime complexity. We also show that SymInfer outperforms existing invariant generation systems.

In order to satisfy safety conditions, an agent may be constrained from acting freely. A safe controller can be designed a priori if an environment is well understood, but not when learning is employed. In particular, reinforcement learned (RL) controllers require exploration, which can be hazardous in safety critical situations. We study the benefits of giving structure to the constraints of a constrained Markov decision process by specifying them in formal languages as a step towards using safety methods from software engineering and controller synthesis. We instantiate these constraints as finite automata to efficiently recognise constraint violations. Constraint states are then used to augment the underlying MDP state and to learn a dense cost function, easing the problem of quickly learning joint MDP/constraint dynamics. We empirically evaluate the effect of these methods on training a variety of RL algorithms over several constraints specified in Safety Gym, MuJoCo, and Atari environments., Comment: NeurIPS 2019 Workshop on Safety and Robustness in Decision Making