This paper explores formal verification for autonomous systems, identifying limitations and proposing improvements.
problem Ensuring safety of autonomous systems like self-driving cars and drones.
method Formal verification techniques based on formal methods, analyzing three assumptions and their limitations.
result Preliminary work to improve the strength of evidence provided by formal verification.
The paper reviews machine learning safety techniques for autonomous vehicles.
problem Challenges in machine learning safety for autonomous vehicles.
method Organizes practical safety techniques to complement engineering safety.
result Enhances dependability and safety of machine learning algorithms in autonomous vehicles.
Classifier-based AI safety gates fail in self-improvement, even with advanced verification methods.
problem Maintaining reliable oversight of AI systems as they improve over iterations.
method Comprehensive empirical testing on neural controllers and MuJoCo benchmarks, using various classifiers and verification methods.
result Classifier-based safety gates fail in maintaining reliable oversight, even with advanced verification methods.
PEREGRiNN verifies safety of ReLU NNs by penalizing relaxation in a greedy manner.
problem Formal verification of safety specifications for ReLU NNs.
method Uses a relaxed convex program to verify polytopic input/output constraints, penalizing relaxation and forcing largest relaxations to early layers.
result Significantly faster and more properties verified compared to other approaches.
Abstract Neural Networks (ANNs) improve DNN verification efficiency.
problem Efficiently verify safety-critical DNNs without slowing exponentially.
method Introduces ANNs that use abstract domains and activation functions to overapproximate DNNs.
result ANNs can soundly overapproximate DNNs with fewer nodes, improving verification efficiency.
Develops first robustness verification for complex Transformers.
problem Certify prediction behavior of Transformers with complex self-attention layers.
method Resolves challenges of cross-nonlinearity and cross-position dependency in Transformers.
result Certified robustness bounds are significantly tighter than Interval Bound Propagation.
Recent advances in machine learning and artificial intelligence are now being considered in safety-critical autonomous systems where software defects may cause severe harm to humans and the environment. Design organizations in these domains are currently unable to provide convincing arguments that their systems are saf…
Despite the tremendous advances that have been made in the last decade on developing useful machine-learning applications, their wider adoption has been hindered by the lack of strong assurance guarantees that can be made about their behavior. In this paper, we consider how formal verification techniques developed for …
Theoretical limits on verifying self-improving systems without risking unbounded utility.
problem Formalizing and proving the limits of safety verification for self-improving systems.
method Developed dual conditions and used Holder's inequality, NP counting method, and Lipschitz bounds to establish impossibility and ceiling results.
result A classifier-based safety gate cannot simultaneously permit unbounded beneficial self-modification and bounded cumulative risk.
The safety and resilience of fully autonomous vehicles (AVs) are of significant concern, as exemplified by several headline-making accidents. While AV development today involves verification, validation, and testing, end-to-end assessment of AV systems under accidental faults in realistic driving scenarios has been lar…
DIFFRNN verifies RNNs for safety and equivalence.
problem Certifying safety and equivalence of RNNs for resource-constrained platforms.
method Bounding nonlinear activation functions with linear constraints, solving constrained optimization problems, and using dReal SMT solver.
result DIFFRNN certifies the equivalence of optimized and original RNNs.
GPUPoly verifies large neural networks robustly on GPUs.
problem Proving robustness of deep neural networks is crucial but challenging.
method Custom polyhedra algorithms on GPUs.
result GPUPoly can verify 1M neuron networks in 34.5 ms.
Deep neural networks have achieved impressive experimental results in image classification, but can surprisingly be unstable with respect to adversarial perturbations, that is, minimal changes to the input image that cause the network to misclassify it. With potential applications including perception modules and end-t…
Safe learning of stochastic dynamics with safety constraints.
problem Learning controlled stochastic dynamics with safety constraints.
method Iterative expansion of a safe control set using kernel-based confidence bounds.
result The method ensures safe exploration and efficient estimation of system dynamics.
Survey of algorithms for testing AI-driven CPS safety.
problem Testing AI-driven CPS for safety in complex environments.
method Survey of applied algorithms for safety validation.
result Survey of existing tools and techniques for safety validation.
To use neural networks in safety-critical settings it is paramount to provide assurances on their runtime operation. Recent work on ReLU networks has sought to verify whether inputs belonging to a bounded box can ever yield some undesirable output. Input-splitting procedures, a particular type of verification mechanism…
A neural network learns to improve verification of neural networks.
problem Efficient verification of neural networks for safety-critical applications.
method A graph neural network (GNN) learns to imitate strong branching heuristics for effective branching in the Branch and Bound (BaB) formulation.
result Reduces the number of branches and verification time by roughly 50% compared to hand-designed strategies.
Accelerates DNN robustness verification with target labels.
problem Improving the robustness of deep neural networks against adversarial attacks.
method Guiding robustness verification with target labels, reducing search space and using symbolic interval propagation and linear relaxation.
result Significantly improves DNN verification speed by 36X, especially when perturbation distance is reasonable.
NeuroDiff improves neural network equivalence verification with fine-grained approximations.
problem Verifying the equivalence of compressed neural networks.
method Symbolic and fine-grained approximation technique for differential verification.
result NeuroDiff achieves up to 1000X speedup and 5X accuracy improvement.
The success of Deep Learning and its potential use in many safety-critical applications has motivated research on formal verification of Neural Network (NN) models. In this context, verification involves proving or disproving that an NN model satisfies certain input-output properties. Despite the reputation of learned …
Artificial Neural Networks (ANNs) have demonstrated remarkable utility in various challenging machine learning applications. While formally verified properties of their behaviors are highly desired, they have proven notoriously difficult to derive and enforce. Existing approaches typically formulate this problem as a p…
Verifying correctness of deep neural networks (DNNs) is challenging. We study a generic reachability problem for feed-forward DNNs which, for a given set of inputs to the network and a Lipschitz-continuous function over its outputs, computes the lower and upper bound on the function values. Because the network and the …
Paper proposes a dataset quality process for ML systems.
problem Inadequate standards for ML datasets in safety-critical systems.
method Proposes a dataset specification and verification process.
result Demonstrates the process on a railway signal recognition system.
This paper provides a comprehensive benchmark and taxonomy for certifiably robust DNN defenses.
problem Certifiably robust defenses against adversarial attacks for deep neural networks.
method Taxonomy and benchmark of certifiably robust approaches.
result First comprehensive benchmark of certifiably robust approaches on different datasets.
DeepDyve uses simpler neural networks to verify DNNs for faults.
problem Faults in DNNs, especially those not crafted by attackers.
method Pre-trained, simpler neural networks for dynamic verification.
result Reduces 90% of risks with 10% overhead.
Advances in the field of Machine Learning and Deep Neural Networks (DNNs) has enabled rapid development of sophisticated and autonomous systems. However, the inherent complexity to rigorously assure the safe operation of such systems hinders their real-world adoption in safety-critical domains such as aerospace and med…
Revel tackles safe exploration in RL with verified symbolic policies.
problem Computational infeasibility of verifying neural networks in RL learning loops.
method Two policy classes: neurosymbolic with approximate gradients and symbolic policies for efficient verification. Mirror descent over policies to safely update and project policies.
result Revel discovers policies that outperform prior approaches to verified exploration.
Generalizes neural network verification by adding arbitrary cutting planes.
problem Handling general cutting plane constraints in neural network verification.
method Generalized bound propagation method (GCP-CROWN) that allows arbitrary cutting plane constraints.
result GCP-CROWN significantly improves neural network verification performance.
Debona improves neural network verification by faster and tighter bounds.
problem Proving adversarial robustness of neural networks is computationally hard.
method Decouples upper and lower bounds computation, re-implements Neurify.
result 94% reduction in runtime for proving robustness of max-pooling layers.
Reinforcement learning is a powerful paradigm for learning optimal policies from experimental data. However, to find optimal policies, most reinforcement learning algorithms explore all possible actions, which may be harmful for real-world systems. As a consequence, learning algorithms are rarely applied on safety-crit…
Study speaker verification security using hierarchical Bayesian modeling.
problem Estimating false alarm rate in ASV systems for large speaker databases.
method Hierarchical Bayesian modeling of ASV scores to assess security against closest impostors.
result Neither i-vector nor x-vector systems are secure against increased impostor database size.
Paper proves neural networks can be approximated using interval bounds.
problem Verifying safety and robustness of neural networks.
method Introduces interval universal approximation (IUA) theorem for neural networks.
result Neural networks can be approximated using interval bounds for any continuous function and squashable activation functions.
Two impossibility theorems show formal alignment certification is impossible for AI systems.
problem Formal certification of AI alignment over open-ended domains is impossible.
method Two independent impossibility theorems: Semantic and Statistical barriers.
result No procedure can simultaneously satisfy soundness, completeness, and tractability.
Formalizes weak and strong verification for LLMs, controlling errors without assumptions.
problem Balancing cost and reliability in reasoning with LLMs.
method Formalizes weak-strong verification policies, introduces metrics, develops online algorithm.
result Optimal policies admit a two-threshold structure, and calibration and sharpness govern value of weak verifiers.
Despite the improved accuracy of deep neural networks, the discovery of adversarial examples has raised serious safety concerns. In this paper, we study two variants of pointwise robustness, the maximum safe radius problem, which for a given input sample computes the minimum distance to an adversarial example, and the …
PRoA assesses deep learning robustness against practical functional perturbations.
problem Inadequate practical robustness verification methods for deep learning systems.
method Probabilistic robustness assessment based on adaptive concentration.
result Statistical guarantees on probabilistic robustness against functional perturbations.
Deep neural networks (DNNs) are increasingly being adopted for sensing and control functions in a variety of safety and mission-critical systems such as self-driving cars, autonomous air vehicles, medical diagnostics, and industrial robotics. Failures of such systems can lead to loss of life or property, which necessit…
ANNs predict SAFARI-1 neutron fluxes with uncertainties.
problem Uncertainty quantification in ANN predictions for SAFARI-1.
method Deep Neural Networks (DNNs) with Monte Carlo Dropout (MCD) and Bayesian Neural Networks (BNN VI) for uncertainty quantification.
result Uncertainty bands envelop noisy measurement data points, indicating good prediction and generalization.
Paper develops PAC verification for hypothesis classes and statistical algorithms.
problem Verifying machine learning models interactively.
method Develops interactive proof for PAC verification, proves lower bounds, and introduces a generalization.
result Improved protocol for verifying unions of intervals and statistical query algorithms.
Hypothesis testing is an important problem with applications in target localization, clinical trials etc. Many active hypothesis testing strategies operate in two phases: an exploration phase and a verification phase. In the exploration phase, selection of experiments is such that a moderate level of confidence on the …
We explore the concept of co-design in the context of neural network verification. Specifically, we aim to train deep neural networks that not only are robust to adversarial perturbations but also whose robustness can be verified more easily. To this end, we identify two properties of network models - weight sparsity a…
New methods combat data poisoning attacks in bandit algorithms using limited verification.
problem Data poisoning attacks on bandit algorithms, especially in the UCB and ETC types.
method Verification-based mechanisms to restore optimal regret with limited verifications.
result A simple modified ETC type bandit algorithm can restore optimal regret with O(logT) verifications. Improved speaker verification with condition-aware backend.
problem Speaker verification calibration issues under unknown conditions.
method Discriminative PLDA model with joint training and condition integration.
result Out-of-the-box excellent calibration performance.
In this paper we demonstrate that performance of a speaker verification system can be improved by concatenating electroencephalography (EEG) signal features with speech signal features or only using EEG signal features. We use state-of-the-art end-to-end deep learning model for performing speaker verification and we de…
Improves neural network verification by merging abstract domains and Lagrangian methods.
problem Prove provable bounds for neural network outputs given input ranges.
method Uses zonotopes within a Lagrangian decomposition to verify deep neural networks.
result Yields bounds that improve upon existing techniques in both time and tightness.
Paper develops a model for verifying facts in tables without pre-retrieved evidence.
problem Verification of factual claims in structured data, especially in open-domain settings.
method Joint reranking-and-verification model that fuses evidence documents.
result Model achieves comparable performance to closed-domain state-of-the-art on TabFact dataset.
The increasing inclusion of Machine Learning (ML) models in safety critical systems like autonomous cars have led to the development of multiple model-based ML testing techniques. One common denominator of these testing techniques is their assumption that training programs are adequate and bug-free. These techniques on…
Efficiently verifies neural networks by handling neuron splits, improving speed and accuracy.
problem Handling neuron split constraints in incomplete neural network verification.
method β-CROWN, which optimizes parameters β to encode neuron splits and uses them in bound propagation.
result β-CROWN significantly speeds up verification while maintaining high accuracy.