Current numerical abstract interpretation relies on fixed, hand-crafted, instruction-specific transformers tailored to each domain, giving rise to three significant limitations. First, extensibility is limited because transformers cannot be reused across domains and new transformers need to be designed for each new domain or operator. Second, precise compositional reasoning over instruction sequences is difficult as transformers are defined only at the instruction level. Third, all downstream tasks are forced to use the same fixed transformer, irrespective of their precision, efficiency, or task-specific requirements. To address these limitations, we propose the Evolving Abstract Transformer, a general transformer that replaces the fixed single-output design of traditional transformers with an adaptable search over a parametric space of sound outputs. This is achieved through two underlying algorithms we develop. First, the Universal Parametric Output Space Encoder (UPOSE) constructs a compact parametric space of sound outputs for any polyhedral numerical domain and any operator in the Quadratic-Bounded Guarded Operators (QGO) class which includes both individual instructions and structured sequences. Next, the Adaptive Gradient Guidance (AGG) algorithm leverages the differentiable structure of the space generated by UPOSE and uses gradient-based updates to efficiently search it according to downstream analysis objectives and available runtime, continually evolving the output as more time is provided. We implement these ideas in the AbsEvolve framework and evaluate their effectiveness across three numerical abstract domains: Zones, Octagons, and Polyhedra. Our results demonstrate that the evolving transformer works across domains and handles diverse instructions and sequences, allowing efficient adaptability in the precision–efficiency tradeoff by adjusting the number of gradient steps in the search, while also reaching the most precise invariants up to 3.2× faster than existing baselines.

How to construct globally sound abstract interpreters to safely approximate program behaviors remains a bottleneck in abstract interpretation. In this paper, we show the potential of using state-of-the-art LLMs to automate this tedious process. Focusing on the neural network verification area, we synthesize non-trivial sound abstract transformers across diverse abstract domains using LLMs to search within infinite space from scratch. We formalize the synthesis task as a constrained optimization problem, for which we design a novel mathematically grounded cost function that measures the degree of unsoundness of each generated candidate transformer, while enforcing hard syntactic and semantic validity constraints. Building on this formulation, we introduce SAIL, a novel unified framework that combines model generation, syntactic and semantic validation, and cost-function-based refinement to synthesize globally sound abstract transformers. Evaluation results show that SAIL not only matches the performance of manually designed transformers, but also is able to synthesize sound and high-precision transformers that do not exist in the literature for complex non-linear operators.

Large Language Models (LLMs) can produce biased responses that can cause representational harms. However, conventional studies are insufficient to thoroughly evaluate biases across LLM responses for different demographic groups (a.k.a. counterfactual bias), as they do not scale to large number of inputs and do not provide guarantees. Therefore, we propose the first framework, LLMCert-B that certifies LLMs for counterfactual bias on distributions of prompts. A certificate consists of high-confidence bounds on the probability of unbiased LLM responses for any set of counterfactual prompts - prompts differing by demographic groups, sampled from a distribution. We illustrate counterfactual bias certification for distributions of counterfactual prompts created by applying prefixes sampled from prefix distributions, to a given set of prompts. We consider prefix distributions consisting random token sequences, mixtures of manual jailbreaks, and perturbations of jailbreaks in LLM’s embedding space. We generate non-trivial certificates for SOTA LLMs, exposing their vulnerabilities over distributions of prompts generated from computationally inexpensive prefix distributions.

Large Language Models (LLMs) are widely used for tasks such as natural language and code generation, but their outputs often suffer from issues like hallucination, toxicity, and incorrect results. Current libraries for structured LLM generation rely on left-to-right decoding without support for backtracking, limiting the ability to correct or refine outputs mid-generation. To address this, we introduce IterGen, a user-friendly library for iterative, grammar-guided LLM generation that enables users to move both forward and backward within the generated output based on grammar symbols. By leveraging a symbol-to-position mapping and maintaining the key-value (KV) cache state, IterGen ensures efficient and structured generation while allowing for corrections during the process. We demonstrate IterGen's effectiveness in two important applications: reducing privacy leakage in LLM outputs, improving the accuracy of LLM-generated SQL and Vega-Lite queries. Our code and additional resources are available at https://structuredllm.com.

Variational Autoencoders (VAEs) have become increasingly popular and deployed in safety-critical applications. In such applications, we want to give certified probabilistic guarantees on performance under adversarial attacks. We propose a novel method, CIVET, for certified training of VAEs. CIVET depends on the key insight that we can bound worst-case VAE error by bounding the error on carefully chosen support sets at the latent layer. We show this point mathematically and present a novel training algorithm utilizing this insight. We show in an extensive evaluation across different datasets (in both the wireless and vision application areas), architectures, and perturbation magnitudes that our method outperforms SOTA methods achieving good standard performance with strong robustness guarantees.

Code generation, symbolic math reasoning, and other tasks require LLMs to produce outputs that are both syntactically and semantically correct. Constrained LLM generation is a promising direction to enforce adherence to formal grammar, but prior works have empirically observed that strict enforcement of formal constraints often diminishes the reasoning capabilities of LLMs. In this work, we first provide a theoretical explanation for why constraining LLM outputs to very restrictive grammars that only allow syntactically valid final answers reduces the reasoning capabilities of the model. Second, we demonstrate that by augmenting the output grammar with carefully designed additional rules, it is always possible to preserve the reasoning capabilities of the LLM while ensuring syntactic and semantic correctness in its outputs. Building on these theoretical insights, we propose a reasoning-augmented constrained decoding algorithm, CRANE, which effectively balances the correctness of constrained generation with the flexibility of unconstrained generation. Experiments on multiple open-source LLMs and benchmarks show that CRANE significantly outperforms both state-of-the-art constrained decoding strategies and standard unconstrained decoding, showing up to 10% points accuracy improvement over baselines on challenging symbolic reasoning benchmarks GSM-symbolic and FOLIO.

Autonomous agentic AI systems powered by vision-language models (VLMs) are rapidly advancing toward real-world deployment, yet their cross-modal reasoning capabilities introduce new attack surfaces for adversarial manipulation that exploit semantic reasoning across modalities. Existing adversarial attacks typically rely on visible pixel perturbations or require privileged model or environment access, making them impractical for stealthy, real-world exploitation. We introduce TRAP, a novel generative adversarial framework that manipulates the agent’s decision-making using diffusion-based semantic injections into the vision-language embedding space. Our method combines negative prompt–based degradation with positive semantic optimization, guided by a Siamese semantic network and layout-aware spatial masking. Without requiring access to model internals, TRAP produces visually natural images yet induces consistent selection biases in agentic AI systems. We evaluate TRAP on the Microsoft Common Objects in Context (COCO) dataset, building multi-candidate decision scenarios. Across these scenarios, TRAP consistently induces decision-level preference redirection on leading models, including LLaVA-34B, Gemma3, GPT-4o, and Mistral-3.2, significantly outperforming existing baselines such as SPSA, Bandit, and standard diffusion approaches. These findings expose a critical, generalized vulnerability: autonomous agents can be consistently misled through visually subtle, semantically-guided cross-modal manipulations. Overall, our results show the need for defense strategies beyond pixel-level robustness to address semantic vulnerabilities in cross-modal decision-making. The code for TRAP is accessible on GitHub at https://github.com/uiuc-focal-lab/TRAP.

Diffusion LLMs have emerged as a promising alternative to conventional autoregressive LLMs, offering substantial potential for improving runtime efficiency. However, existing diffusion models fail to provably enforce user-specified formal constraints, such as regular expressions, which makes them unreliable for tasks that require structured outputs, such as fixed-schema JSON generation. Unlike autoregressive models, which generate tokens sequentially, diffusion LLMs predict a block of tokens in parallel. This parallelism makes traditional constrained decoding algorithms, designed to enforce constraints with sequential token prediction, ineffective at preserving the true output distribution. To address this limitation, we propose DINGO, a dynamic programming-based constrained decoding strategy that is both efficient and provably distribution-preserving. DINGO enables sampling of output strings with the highest probability under the model’s predicted distribution while strictly adhering to any user-specified regular expression. On standard symbolic math and JSON generation benchmarks, DINGO achieves up to a $68$\% points of improvement over unconstrained inference. The code is available at [**DINGO**](https://github.com/uiuc-focal-lab/DINGO).

The uninterpretability of Deep Neural Networks (DNNs) hinders their use in safety-critical applications. Abstract Interpretation-based DNN certifiers provide promising avenues for building trust in DNNs. Unsoundness in the mathematical logic of these certifiers can lead to incorrect results. However, current approaches to ensure their soundness rely on manual, expert-driven proofs that are tedious to develop, limiting the speed of developing new certifiers. Automating the verification process is challenging due to the complexity of verifying certifiers for arbitrary DNN architectures and handling diverse abstract analyses. We introduce ProveSound , a novel verification procedure that automates the soundness verification of DNN certifiers for arbitrary DNN architectures. Our core contribution is the novel concept of a symbolic DNN, using which, ProveSound reduces the soundness property, a universal quantification over arbitrary DNNs, to a tractable symbolic representation, enabling verification with standard SMT solvers. By formalizing the syntax and operational semantics of ConstraintFlow , a DSL for specifying certifiers, ProveSound efficiently verifies both existing and new certifiers, handling arbitrary DNN architectures. Our code is available at https://github.com/uiuc-focal-lab/constraintflow.git

In recent years numerous methods have been developed to formally verify the robustness of deep neural networks (DNNs). Though the proposed techniques are effective in providing mathematical guarantees about the DNNs' behavior, it is not clear whether the proofs generated by these methods are human-understandable. In this paper, we bridge this gap by developing new concepts, algorithms, and representations to generate human understandable insights into the internal workings of DNN robustness proofs. Leveraging the proposed method, we show that the robustness proofs of standard DNNs rely more on spurious input features as compared to the proofs of DNNs trained to be robust. Robustness proofs of the provably robust DNNs filter out a larger number of spurious input features as compared to adversarially trained DNNs, sometimes even leading to the pruning of semantically meaningful input features. The proofs for the DNNs combining adversarial and provably robust training tend to achieve the middle ground

Randomized smoothing-based certification is an effective approach for obtaining robustness certificates of deep neural networks (DNNs) against adversarial attacks. This method constructs a smoothed DNN model and certifies its robustness through statistical sampling, but it is computationally expensive, especially when certifying with a large number of samples. Furthermore, when the smoothed model is modified (e.g., quantized or pruned), certification guarantees may not hold for the modified DNN, and recertifying from scratch can be prohibitively expensive. We present the first approach for incremental robustness certification for randomized smoothing, IRS. We show how to reuse the certification guarantees for the original smoothed model to certify an approximated model with very few samples. IRS significantly reduces the computational cost of certifying modified DNNs while maintaining strong robustness guarantees. We experimentally demonstrate the effectiveness of our approach, showing up to 4.1x certification speedup over the certification that applies randomized smoothing of the approximate model from scratch.

We focus on verifying relational properties defined over deep neural networks (DNNs) such as robustness against universal adversarial perturbations (UAP), certified worst-case hamming distance for binary string classifications, etc. Precise verification of these properties requires reasoning about multiple executions of the same DNN. However, most of the existing works in DNN verification only handle properties defined over single executions and as a result, are imprecise for relational properties. Though few recent works for relational DNN verification, capture linear dependencies between the inputs of multiple executions, they do not leverage dependencies between the outputs of hidden layers producing imprecise results. We develop a scalable relational verifier RACoon that utilizes cross-execution dependencies at all layers of the DNN gaining substantial precision over SOTA baselines on a wide range of datasets, networks, and relational properties.

Universal Adversarial Perturbations (UAPs) are imperceptible, image-agnostic vectors that cause deep neural networks (DNNs) to misclassify inputs with high probability. In practical attack scenarios, adversarial perturbations may undergo transformations such as changes in pixel intensity, scaling, etc. before being added to DNN inputs. Existing methods do not create UAPs robust to these real-world transformations, thereby limiting their applicability in practical attack scenarios. In this work, we introduce and formulate UAPs robust against real-world transformations. We build an iterative algorithm using probabilistic robustness bounds and construct UAPs robust to transformations generated by composing arbitrary sub-differentiable transformation functions. We perform an extensive evaluation on the popular CIFAR-10 and ILSVRC 2012 datasets measuring our UAPs’ robustness under a wide range common, real-world transformations such as rotation, contrast changes, etc. We further show that by using a set of primitive transformations our method generalizes well to unseen transformations such as fog, JPEG compression, etc. Our results show that our method can generate UAPs up to 23% more robust than state-of-the-art baselines.

We present NN4SysBench, a benchmark suite for neural network verification that is composed of applications from the domain of computer systems. We call these neural networks for computer systems or NN4Sys. NN4Sys is booming: there are many proposals for using neural networks in computer systems—for example, databases, OSes, and networked systems—many of which are safety critical. Neural network verification is a technique to formally verify whether neural networks satisfy safety properties. We however observe that NN4Sys has some unique characteristics that today’s verification tools overlook and have limited support. Therefore, this benchmark suite aims at bridging the gap between NN4Sys and the verification by using impactful NN4Sys applications as benchmarks to illustrate computer systems’ unique challenges. We also build a compatible version of NN4SysBench, so that today’s verifiers can also work on these benchmarks with approximately the same verification difficulties. The code is available at https://github.com/lydialin1212/NN4Sys_Benchmark.

We propose RABBit, a Branch-and-Bound-based verifier for verifying relational properties defined over Deep Neural Networks, such as robustness against universal adversarial perturbations (UAP). Existing SOTA complete $L_{\infty}$-robustness verifiers can not reason about dependencies between multiple executions and, as a result, are imprecise for relational verification. In contrast, existing SOTA relational verifiers only apply a single bounding step and do not utilize any branching strategies to refine the obtained bounds, thus producing imprecise results. We develop the first scalable Branch-and-Bound-based relational verifier, RABBit, which efficiently combines branching over multiple executions with cross-executional bound refinement to utilize relational constraints, gaining substantial precision over SOTA baselines on a wide range of datasets and networks. Our code is at https://github.com/uiuc-focal-lab/RABBit.

We consider the verification of input-relational properties defined over deep neural networks (DNNs) such as robustness against universal adversarial perturbations, monotonicity, etc. Precise verification of these properties requires reasoning about multiple executions of the same DNN. We introduce a novel concept of difference tracking to compute the difference between the outputs of two executions of the same DNN at all layers. We design a new abstract domain, DiffPoly for efficient difference tracking that can scale large DNNs. DiffPoly is equipped with custom abstract transformers for common activation functions (ReLU, Tanh, Sigmoid, etc.) and affine layers and can create precise linear cross-execution constraints. We implement an input-relational verifier for DNNs called RaVeN which uses DiffPoly and linear program formulations to handle a wide range of input-relational properties. Our experimental results on challenging benchmarks show that by leveraging precise linear constraints defined over multiple executions of the DNN, RaVeN gains substantial precision over baselines on a wide range of datasets, networks, and input-relational properties.

We present Pasado, a technique for synthesizing precise static analyzers for Automatic Differentiation. Our technique allows one to automatically construct a static analyzer specialized for the Chain Rule, Product Rule, and Quotient Rule computations for Automatic Differentiation in a way that abstracts all of the nonlinear operations of each respective rule simultaneously. By directly synthesizing an abstract transformer for the composite expressions of these 3 most common rules of AD, we are able to obtain significant precision improvement compared to prior works which compose standard abstract transformers together suboptimally. We prove our synthesized static analyzers sound and additionally demonstrate the generality of our approach by instantiating these AD static analyzers with different nonlinear functions, different abstract domains (both intervals and zonotopes) and both forward-mode and reverse-mode AD. We evaluate Pasado on multiple case studies, namely soundly computing bounds on a neural network’s local Lipschitz constant, soundly bounding the sensitivities of financial models, certifying monotonicity, and lastly, bounding sensitivities of the solutions of differential equations from climate science and chemistry for verified ranges of initial conditions and parameters. The local Lipschitz constants computed by Pasado on our largest CNN are up to 2750× more precise compared to the existing state-of-the-art zonotope analysis. The bounds obtained on the sensitivities of the climate, chemical, and financial differential equation solutions are between 1.31 − 2.81× more precise (on average) compared to a state-of-the-art zonotope analysis.

Complete verification of deep neural networks (DNNs) can exactly determine whether the DNN satisfies a desired trustworthy property (e.g., robustness, fairness) on an infinite set of inputs or not. Despite the tremendous progress to improve the scalability of complete verifiers over the years on individual DNNs, they are inherently inefficient when a deployed DNN is updated to improve its inference speed or accuracy. The inefficiency is because the expensive verifier needs to be run from scratch on the updated DNN. To improve efficiency, we propose a new, general framework for incremental and complete DNN verification based on the design of novel theory, data structure, and algorithms. Our contributions implemented in a tool named IVAN yield an overall geometric mean speedup of 2.4x for verifying challenging MNIST and CIFAR10 classifiers and a geometric mean speedup of 3.8x for the ACAS-XU classifiers over the state-of-the-art baselines.

Abstract Existing neural network verifiers compute a proof that each input is handled correctly under a given perturbation by propagating a symbolic abstraction of reachable values at each layer. This process is repeated from scratch independently for each input (e.g., image) and perturbation (e.g., rotation), leading to an expensive overall proof effort when handling an entire dataset. In this work, we introduce a new method for reducing this verification cost without losing precision based on a key insight that abstractions obtained at intermediate layers for different inputs and perturbations can overlap or contain each other. Leveraging our insight, we introduce the general concept of shared certificates, enabling proof effort reuse across multiple inputs to reduce overall verification costs. We perform an extensive experimental evaluation to demonstrate the effectiveness of shared certificates in reducing the verification cost on a range of datasets and attack specifications on image classifiers including the popular patch and geometric perturbations. We release our implementation at https://github.com/eth-sri/proof-sharing .

We present a novel abstraction for bounding the Clarke Jacobian of a Lipschitz continuous, but not necessarily differentiable function over a local input region. To do so, we leverage a novel abstract domain built upon dual numbers, adapted to soundly over-approximate all first derivatives needed to compute the Clarke Jacobian. We formally prove that our novel forward-mode dual interval evaluation produces a sound, interval domain-based over-approximation of the true Clarke Jacobian for a given input region. Due to the generality of our formalism, we can compute and analyze interval Clarke Jacobians for a broader class of functions than previous works supported – specifically, arbitrary compositions of neural networks with Lipschitz, but non-differentiable perturbations. We implement our technique in a tool called DeepJ and evaluate it on multiple deep neural networks and non-differentiable input perturbations to showcase both the generality and scalability of our analysis. Concretely, we can obtain interval Clarke Jacobians to analyze Lipschitz robustness and local optimization landscapes of both fully-connected and convolutional neural networks for rotational, contrast variation, and haze perturbations, as well as their compositions.

We present a novel, general construction to abstractly interpret higher-order automatic differentiation (AD). Our construction allows one to instantiate an abstract interpreter for computing derivatives up to a chosen order. Furthermore, since our construction reduces the problem of abstractly reasoning about derivatives to abstractly reasoning about real-valued straight-line programs, it can be instantiated with almost any numerical abstract domain, both relational and non-relational. We formally establish the soundness of this construction. We implement our technique by instantiating our construction with both the non-relational interval domain and the relational zonotope domain to compute both first and higher-order derivatives. In the latter case, we are the first to apply a relational domain to automatic differentiation for abstracting higher-order derivatives, and hence we are also the first abstract interpretation work to track correlations across not only different variables, but different orders of derivatives. We evaluate these instantiations on multiple case studies, namely robustly explaining a neural network and more precisely computing a neural network’s Lipschitz constant. For robust interpretation, first and second derivatives computed via zonotope AD are up to 4.76× and 6.98× more precise, respectively, compared to interval AD. For Lipschitz certification, we obtain bounds that are up to 11,850× more precise with zonotopes, compared to the state-of-the-art interval-based tool.

Formal verification of neural networks is critical for their safe adoption in real-world applications. However, designing a precise and scalable verifier which can handle different activation functions, realistic network architectures and relevant specifications remains an open and difficult challenge. In this paper, we take a major step forward in addressing this challenge and present a new verification framework, called PRIMA. PRIMA is both (i) general: it handles any non-linear activation function, and (ii) precise: it computes precise convex abstractions involving multiple neurons via novel convex hull approximation algorithms that leverage concepts from computational geometry. The algorithms have polynomial complexity, yield fewer constraints, and minimize precision loss. We evaluate the effectiveness of PRIMA on a variety of challenging tasks from prior work. Our results show that PRIMA is significantly more precise than the state-of-the-art, verifying robustness to input perturbations for up to 20%, 30%, and 34% more images than existing work on ReLU-, Sigmoid-, and Tanh-based networks, respectively. Further, PRIMA enables, for the first time, the precise verification of a realistic neural network for autonomous driving within a few minutes.

Developers of machine learning applications often apply post-training neural network optimizations, such as quantization and pruning, that approximate a neural network to speed up inference and reduce energy consumption, while maintaining high accuracy and robustness. Despite a recent surge in techniques for the robustness verification of neural networks, a major limitation of almost all state-of-the-art approaches is that the verification needs to be run from scratch every time the network is even slightly modified. Running precise end-to-end verification from scratch for every new network is expensive and impractical in many scenarios that use or compare multiple approximate network versions, and the robustness of all the networks needs to be verified efficiently. We present FANC, the first general technique for transferring proofs between a given network and its multiple approximate versions without compromising verifier precision. To reuse the proofs obtained when verifying the original network, FANC generates a set of templates – connected symbolic shapes at intermediate layers of the original network – that capture the proof of the property to be verified. We present novel algorithms for generating and transforming templates that generalize to a broad range of approximate networks and reduce the verification cost. We present a comprehensive evaluation demonstrating the effectiveness of our approach. We consider a diverse set of networks obtained by applying popular approximation techniques such as quantization and pruning on fully-connected and convolutional architectures and verify their robustness against different adversarial attacks such as adversarial patches, L 0 , rotation and brightening. Our results indicate that FANC can significantly speed up verification with state-of-the-art verifier, DeepZ by up to 4.1x.

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 whether these GNN-based job schedulers meet users’ expectations about other important properties, such as strategy-proofness, sharing incentive, and stability. In this work, we consider formal verification of GNN-based job schedulers. We address several domain-specific challenges such as networks that are deeper and specifications that are richer than those encountered when verifying image and NLP classifiers. We develop vegas, 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 show that vegas achieves significant speed-up when verifying important properties of a state-of-the-art GNN-based scheduler compared to previous methods.

Abstract We present a scalable and precise verifier for recurrent neural networks, calledProverbased on two novel ideas: (i) a method to compute a set of polyhedral abstractions for the non-convex and non-linear recurrent update functions by combining sampling, optimization, and Fermat’s theorem, and (ii) a gradient descent based algorithm for abstraction refinement guided by the certification problem that combines multiple abstractions for each neuron. UsingProver, we present the first study of certifying a non-trivial use case of recurrent neural networks, namely speech classification. To achieve this, we additionally develop custom abstractions for the non-linear speech preprocessing pipeline. Our evaluation shows thatProversuccessfully verifies several challenging recurrent models in computer vision, speech, and motion sensor data classification beyond the reach of prior work.

The use of deep 3D point cloud models in safety-critical applications, such as autonomous driving, dictates the need to certify the robustness of these models to real-world trans-formations. This is technically challenging, as it requires a scalable verifier tailored to point cloud models that handles a wide range of semantic 3D transformations. In this work, we address this challenge and introduce 3DCertify, the first verifier able to certify the robustness of point cloud models. 3DCertify is based on two key insights: (i) a generic relaxation based on first-order Taylor approximations, applicable to any differentiable transformation, and (ii) a precise relaxation for global feature pooling, which is more complex than pointwise activations (e.g., ReLU or sigmoid) but commonly employed in point cloud models. We demonstrate the effectiveness of 3DCertify by performing an extensive evaluation on a wide range of 3D transformations (e.g., rotation, twisting) for both classification and part segmentation tasks. For example, we can certify robustness against rotations by ±60° for 95.7% of point clouds, and our max pool relaxation increases certification by up to 15.6%.

Massive MIMO forms a crucial component for 5G because of its ability to improve quality of service and support multiple streams simultaneously. However, for real-world MIMO deployments, estimating the downlink wireless channel from each antenna on the base station to every client device is a critical bottleneck, especially for the widely used frequency duplexed designs that cannot utilize reciprocity. Typically, this channel estimation requires explicit feedback from client devices and is prohibitive for large antenna deployments. In this paper, we present FIRE, a system that uses an end-to-end machine learning approach to enable accurate channel estimation without requiring any feedback from client devices. FIRE is interpretable, accurate, and has low compute overhead. We show that FIRE can successfully support MIMO transmissions in a real-world testbed and achieves SNR improvement over 10 dB in MIMO transmissions compared to the current state-of-the-art.

We develop an effective generation of adversarial attacks on neural models that output a sequence of probability distributions rather than a sequence of single values. This setting includes the recently proposed deep probabilistic autoregressive forecasting models that estimate the probability distribution of a time series given its past and achieve state-of-the-art results in a diverse set of application domains. The key technical challenge we address is how to effectively differentiate through the Monte-Carlo estimation of statistics of the output sequence joint distribution. Additionally, we extend prior work on probabilistic forecasting to the Bayesian setting which allows conditioning on future observations, instead of only on past observations. We demonstrate that our approach can successfully generate attacks with small input perturbations in two challenging tasks where robust decision making is crucial – stock market trading and prediction of electricity consumption.

Numerical abstract domains are a key component of modern static analyzers. Despite recent advances, precise analysis with highly expressive domains remains too costly for many real-world programs. To address this challenge, we introduce a new data-driven method, called LAIT, that produces a faster and more scalable numerical analysis without significant loss of precision. Our approach is based on the key insight that sequences of abstract elements produced by the analyzer contain redundancy which can be exploited to increase performance without compromising precision significantly. Concretely, we present an iterative learning algorithm that learns a neural policy that identifies and removes redundant constraints at various points in the sequence. We believe that our method is generic and can be applied to various numerical domains.

The use of neural networks in safety-critical computer vision systems calls for their robustness certification against natural geometric transformations (e.g., rotation, scaling). However, current certification methods target mostly norm-based pixel perturbations and cannot certify robustness against geometric transformations. In this work, we propose a new method to compute sound and asymptotically optimal linear relaxations for any composition of transformations. Our method is based on a novel combination of sampling and optimization. We implemented the method in a system called DeepG and demonstrated that it certifies significantly more complex geometric transformations than existing methods on both defended and undefended networks while scaling to large architectures.

We propose a new parametric framework, called k-ReLU, for computing precise and scalable convex relaxations used to certify neural networks. The key idea is to approximate the output of multiple ReLUs in a layer jointly instead of separately. This joint relaxation captures dependencies between the inputs to different ReLUs in a layer and thus overcomes the convex barrier imposed by the single neuron triangle relaxation and its approximations. The framework is parametric in the number of k ReLUs it considers jointly and can be combined with existing verifiers in order to improve their precision. Our experimental results show that k-ReLU en- ables significantly more precise certification than existing state-of-the-art verifiers while maintaining scalability.

We present a novel method for scalable and precise certification of deep neural networks. The key technical insight behind our approach is a new abstract domain which combines floating point polyhedra with intervals and is equipped with abstract transformers specifically tailored to the setting of neural networks. Concretely, we introduce new transformers for affine transforms, the rectified linear unit (ReLU), sigmoid, tanh, and maxpool functions. We implemented our method in a system called DeepPoly and evaluated it extensively on a range of datasets, neural architectures (including defended networks), and specifications. Our experimental results indicate that DeepPoly is more precise than prior work while scaling to large networks. We also show how to combine DeepPoly with a form of abstraction refinement based on trace partitioning. This enables us to prove, for the first time, the robustness of the network when the input image is subjected to complex perturbations such as rotations that employ linear interpolation.

We show how to leverage reinforcement learning (RL) in order to speed up static program analysis. The key insight is to establish a correspondence between concepts in RL and those in analysis: a state in RL maps to an abstract program state in analysis, an action maps to an abstract transformer, and at every state, we have a set of sound transformers (actions) that represent different trade-offs between precision and performance. At each iteration, the agent (analysis) uses a policy learned offline by RL to decide on the transformer which minimizes loss of precision at fixpoint while improving analysis performance. Our approach leverages the idea of online decomposition (applicable to popular numerical abstract domains) to define a space of new approximate transformers with varying degrees of precision and performance. Using a suitably designed set of features that capture key properties of abstract program states and available actions, we then apply Q-learning with linear function approximation to compute an optimized context-sensitive policy that chooses transformers during analysis. We implemented our approach for the notoriously expensive Polyhedra domain and evaluated it on a set of Linux device drivers that are expensive to analyze. The results show that our approach can yield massive speedups of up to two orders of magnitude while maintaining precision at fixpoint.

We present a new method and system, called DeepZ, for certifying neural network robustness based on abstract interpretation. Compared to state-of-the-art automated verifiers for neural networks, DeepZ: (i) handles ReLU, Tanh and Sigmoid activation functions, (ii) supports feedforward and convolutional architectures, (iii) is significantly more scalable and precise, and (iv) and is sound with respect to floating point arithmetic. These benefits are due to carefully designed approximations tailored to the setting of neural networks. As an example, DeepZ achieves a verification accuracy of 97% on a large network with 88,500 hidden units under $L_{\infty}$ attack with $\epsilon = 0.1$ with an average runtime of 133 seconds.

Numerical abstract domains such as Polyhedra, Octahedron, Octagon, Interval, and others are an essential component of static program analysis. The choice of domain offers a performance/precision tradeoff ranging from cheap and imprecise (Interval) to expensive and precise (Polyhedra). Recently, significant speedups were achieved for Octagon and Polyhedra by manually decomposing their transformers to work with the Cartesian product of projections associated with partitions of the variable set. While practically useful, this manual process is time consuming, error-prone, and has to be applied from scratch for every domain. In this paper, we present a generic approach for decomposing the transformers of sub-polyhedra domains along with conditions for checking whether the decomposed transformers lose precision with respect to the original transformers. These conditions are satisfied by most practical transformers, thus our approach is suitable for increasing the performance of these transformers without compromising their precision. Furthermore, our approach is ``black box:'' it does not require changes to the internals of the original non-decomposed transformers or additional manual effort per domain. We implemented our approach and applied it to the domains of Zones, Octagon, and Polyhedra. We then compared the performance of the decomposed transformers obtained with our generic method versus the state of the art: the (non-decomposed) PPL for Polyhedra and the much faster ELINA (which uses manual decomposition) for Polyhedra and Octagon. Against ELINA we demonstrate finer partitions and an associated speedup of about 2x on average. Our results indicate that the general construction presented in this work is a viable method for improving the performance of sub-polyhedra domains. It enables designers of abstract domains to benefit from decomposition without re-writing all of their transformers from scratch as required by prior work.

Numerical abstract domains are an important ingredient of modern static analyzers used for verifying critical program properties (e.g., absence of buffer overflow or memory safety). Among the many numerical domains introduced over the years, Polyhedra is the most expressive one, but also the most expensive: it has worst-case exponential space and time complexity. As a consequence, static analysis with the Polyhedra domain is thought to be impractical when applied to large scale, real world programs.

Numerical abstract domains are a fundamental component in modern static program analysis and are used in a wide range of scenarios (e.g. computing array bounds, disjointness, etc). However, analysis with these domains can be very expensive, deeply affecting the scalability and practical applicability of the static analysis. Hence, it is critical to ensure that these domains are made highly efficient. In this work, we present a complete approach for optimizing the performance of the Octagon numerical abstract domain, a domain shown to be particularly effective in practice. Our optimization approach is based on two key insights: i) the ability to perform online decomposition of the octagons leading to a massive reduction in operation counts, and ii) leveraging classic performance optimizations from linear algebra such as vectorization, locality of reference, scalar replacement and others, for improving the key bottlenecks of the domain. Applying these ideas, we designed new algorithms for the core Octagon operators with better asymptotic runtime than prior work and combined them with the optimization techniques to achieve high actual performance. We implemented our approach in the Octagon operators exported by the popular APRON C library, thus enabling existing static analyzers using APRON to immediately benefit from our work. To demonstrate the performance benefits of our approach, we evaluated our framework on three published static analyzers showing massive speed-ups for the time spent in Octagon analysis (e.g., up to 146x) as well as significant end-to-end program analysis speed-ups (up to 18.7x). Based on these results, we believe that our framework can serve as a new basis for static analysis with the Octagon numerical domain.