Rust is a programming language that combines memory safety and low-level control, providing C-like performance while guaranteeing the absence of undefined behaviors by default. Rust’s growing popularity has prompted research on correct and idiomatic transpiling of existing code-bases to Rust. Existing work falls into two categories: rule-based and large language model (LLM)-based. While rule-based approaches are theoretically sound, they often yield unidiomatic and unsafe Rust code, and are limited to few source languages, which hinders maintainability and industrial application. By contrast, LLM-based approaches, while providing no guarantees, are polyglot and typically produce more idiomatic and safe Rust code. In this work, we present VERT, a formally correct, polyglot Rust translator with more idiomatic outputs. VERT supports any language that compiles to Web Assembly. Using the Web Assembly compiler, VERT obtains an oracle Rust program. Leveraging the LLM, VERT generates an idiomatic candidate Rust program. This candidate is verified against the oracle with model-checking to ensure equivalence.

Large language models (LLMs) show promise in code translation due to their ability to generate idiomatic code. However, a significant limitation when using LLMs for code translation is scalability: existing works have shown a drop in translation success rates for code exceeding around 100 lines. We overcome this limitation by developing a modular approach to translation, where we partition the code into small code fragments which can be translated independently and semantically validated (that is, by checking I/O equivalence). When this approach is applied naively, we discover that LLMs are unreliable when translating features of the source language that do not have a direct mapping to the target language, and that the LLM often gets stuck in repair loops when attempting to fix errors. To address these issues, we introduce two key concepts: (1) feature mapping , which integrates predefined translation rules with LLM-based translation to guide the LLM in navigating subtle language differences and producing semantically accurate code; and (2) type-compatibility , which facilitates localized checks at the function signature level to detect errors early, thereby narrowing the scope of potential repairs. We apply our approach to translating real-world Go codebases to Rust, demonstrating that we can consistently generate reliable Rust translations for projects up to 9,700 lines of code and 780 functions, with an average of 73% of functions successfully validated for I/O equivalence, considerably higher than any existing work.

Abstract Linear approximations of nonlinear functions have a wide range of applications such as rigorous global optimization and, recently, verification problems involving neural networks. In the latter case, a linear approximation must be hand-crafted for the neural network’s activation functions. This hand-crafting is tedious, potentially error-prone, and requires an expert to prove the soundness of the linear approximation. Such a limitation is at odds with the rapidly advancing deep learning field – current verification tools either lack the necessary linear approximation, or perform poorly on neural networks with state-of-the-art activation functions. In this work, we consider the problem of automatically synthesizing sound linear approximations for a given neural network activation function. Our approach is example-guided : we develop a procedure to generate examples, and then we leverage machine learning techniques to learn a (static) function that outputs linear approximations. However, since the machine learning techniques we employ do not come with formal guarantees, the resulting synthesized function may produce linear approximations with violations. To remedy this, we bound the maximum violation using rigorous global optimization techniques, and then adjust the synthesized linear approximation accordingly to ensure soundness. We evaluate our approach on several neural network verification tasks. Our evaluation shows that the automatically synthesized linear approximations greatly improve the accuracy (i.e., in terms of the number of verification problems solved) compared to hand-crafted linear approximations in state-of-the-art neural network verification tools. An artifact with our code and experimental scripts is available at: https://zenodo.org/record/6525186#.Yp51L9LMIzM . "Image missing" "Image missing"

Abstract The most scalable approaches to certifying neural network robustness depend on computing sound linear lower and upper bounds for the network’s activation functions. Current approaches are limited in that the linear bounds must be handcrafted by an expert, and can be sub-optimal, especially when the network’s architecture composes operations using, for example, multiplication such as in LSTMs and the recently popularSwishactivation. The dependence on an expert prevents the application of robustness certification to developments in the state-of-the-art of activation functions, and furthermore the lack of tightness guarantees may give a false sense of insecurity about a particular model. To the best of our knowledge, we are the first to consider the problem ofautomaticallysynthesizingtightlinear bounds for arbitrary n-dimensional activation functions. We propose the first fully automated method that achieves tight linear bounds while only leveraging the mathematical definition of the activation function itself. Our method leverages an efficient heuristic technique to synthesize bounds that are tight andusually sound, and then verifies the soundness (and adjusts the bounds if necessary) using the highly optimized branch-and-bound SMT solver,dReal. Even though our method depends on an SMT solver, we show that the runtime is reasonable in practice, and, compared with state of the art, our method often achieves 2-5X tighter final output bounds and more than quadruple certified robustness.

As deep neural networks are increasingly being deployed in practice, their efficiency has become an important issue. While there are compression techniques for reducing the network's size, energy consumption and computational requirement, they only demonstrate empirically that there is no loss of accuracy, but lack formal guarantees of the compressed network, e.g., in the presence of adversarial examples. Existing verification techniques such as Reluplex, ReluVal, and DeepPoly provide formal guarantees, but they are designed for analyzing a single network instead of the relationship between two networks. To fill the gap, we develop a new method for differential verification of two closely related networks. Our method consists of a fast but approximate forward interval analysis pass followed by a backward pass that iteratively refines the approximation until the desired property is verified. We have two main innovations. During the forward pass, we exploit structural and behavioral similarities of the two networks to more accurately bound the difference between the output neurons of the two networks. Then in the backward pass, we leverage the gradient differences to more accurately compute the most beneficial refinement. Our experiments show that, compared to state-of-the-art verification tools, our method can achieve orders-of-magnitude speedup and prove many more properties than existing tools.

As neural networks make their way into safety-critical systems, where misbehavior can lead to catastrophes, there is a growing interest in certifying the equivalence of two structurally similar neural networks - a problem known as differential verification. For example, compression techniques are often used in practice for deploying trained neural networks on computationally- and energy-constrained devices, which raises the question of how faithfully the compressed network mimics the original network. Unfortunately, existing methods either focus on verifying a single network or rely on loose approximations to prove the equivalence of two networks. Due to overly conservative approximation, differential verification lacks scalability in terms of both accuracy and computational cost. To overcome these problems, we propose NeuroDiff, a symbolic and fine-grained approximation technique that drastically increases the accuracy of differential verification on feed-forward ReLU networks while achieving many orders-of-magnitude speedup. NeuroDiff has two key contributions. The first one is new convex approximations that more accurately bound the difference of two networks under all possible inputs. The second one is judicious use of symbolic variables to represent neurons whose difference bounds have accumulated significant error. We find that these two techniques are complementary, i.e., when combined, the benefit is greater than the sum of their individual benefits. We have evaluated NeuroDiff on a variety of differential verification tasks. Our results show that NeuroDiff is up to 1000X faster and 5X more accurate than the state-of-the-art tool.

Flash Crowd Attacks (FCAs) are DDoS attacks that flood victim services, such as Web servers, with well-formed requests, generated by numerous bots. It is hard to detect and filter such attacks because both legitimate and attack requests look identical. In our previous work [1], we proposed models of how human users interact with Web servers, and also showed in simulation that these models can detect naive FCA attacks. We significantly extend these proposed models to make them more robust, simpler, and applicable to a wider variety of FCA attacks in this paper. We implement the models in a system called FRADE, and evaluate it on three Web servers with different server applications and different content. We show that FRADE can detect both naive and sophisticated bots within seconds and successfully filters out attack traffic. Therefore, FRADE significantly raises the bar for a successful attack by requiring attackers to deploy botnets that are at least three orders of magnitude larger than the botnets today.

Compression is an emerging source of exploitable side-channel leakage that threatens data security, particularly in web applications where compression is indispensable for performance reasons. Current approaches to mitigating compression side channels have drawbacks in that they either degrade compression ratio drastically or require too much effort from developers to be widely adopted. To bridge the gap, we develop Debreach, a static analysis and program transformation based approach to mitigating compression side channels. Debreach consists of two steps. First, it uses taint analysis to soundly identify flows of sensitive data in the program and uses code instrumentation to annotate data before feeding them to the compressor. Second, it enhances the compressor to exploit the freedom to not compress of standard compression protocols, thus removing the dependency between sensitive data and the size of the compressor's output. Since Debreach automatically instruments applications and does not change the compression protocols, it has the advantage of being non-disruptive and compatible with existing systems. We have evaluated Debreach on a set of web server applications written in PHP. Our experiments show that, while ensuring leakage-freedom, Debreach can achieve significantly higher compression performance than state-of-the-art approaches.

A unified modeling framework for non-functional properties of a program is essential for research in software analysis and verification, since it reduces burdens on individual researchers to implement new approaches and compare existing approaches. We present CANAL, a framework that models the cache behaviors of a program by transforming its intermediate representation in the LLVM compiler. CANAL inserts auxiliary variables and instructions over these variables, to allow standard verification tools to handle a new class of cache related properties, e.g., for computing the worst-case execution time and detecting side-channel leaks.