Fully supervised polyp segmentation relies on costly pixel-level annotations. Although semi- and weakly supervised methods reduce annotation requirements, they still depend on partial mask supervision. Text-supervised segmentation is a promising alternative; however, for polyps, the key challenge is to ground instance-specific phrases to the correct lesion region under cluttered backgrounds and large appearance variations. Existing approaches often rely on coarse text-image alignment, limiting precise region-level semantic correspondence. In this paper, we propose Text-Image Co-Alignment (TICoA), a text-supervised framework for polyp segmentation. TICoA leverages large language models (LLMs)-generated structured clinical descriptions as weak supervision and formulates segmentation as a fine-grained phrase-region co-alignment problem. Through contrastive learning, TICoA explicitly associates query phrases with corresponding image regions to achieve robust semantic grounding under weak supervision. Architecturally, we adopt a State-Space Model (Mamba) to efficiently model long-range dependencies with linear computational complexity. To support effective cross-modal interaction, we further design a dedicated Mamba Fusion module with a Bi-Dimension Fusion (BiDF) strategy, which progressively propagates information along spatial and channel dimensions. Experiments on polyp datasets, with additional validation on skin lesion segmentation, demonstrate that TICoA is competitive with state-of-the-art weakly supervised methods. Our code and data are available at https://github.com/silentyuchen/TICoA.

While attention-based models have demonstrated the remarkable ability of in-context learning (ICL), the theoretical understanding of how these models acquired this ability through gradient descent training is still preliminary. Towards answering this question, we study the gradient descent dynamics of multi-head linear self-attention trained for in-context linear regression. We examine two parametrizations of linear self-attention: one with the key and query weights merged as a single matrix (common in theoretical studies), and one with separate key and query matrices (closer to practical settings). For the merged parametrization, we show that the training dynamics has two fixed points and the loss trajectory exhibits a single, abrupt drop. We derive an analytical time-course solution for a certain class of datasets and initialization. For the separate parametrization, we show that the training dynamics has exponentially many fixed points and the loss exhibits saddle-to-saddle dynamics, which we reduce to scalar ordinary differential equations. During training, the model implements principal component regression in context with the number of principal components increasing over training time. Overall, we provide a theoretical description of how ICL abilities evolve during gradient descent training of linear attention, revealing abrupt acquisition or progressive improvements depending on how the key and query are parametrized.

Large language models (LLMs) have emerged as a dominant AI paradigm due to their exceptional text understanding and generation capabilities. However, their tendency to generate inconsistent or erroneous outputs challenges their reliability, especially in high-stakes domains requiring accuracy and trustworthiness. Existing research primarily focuses on detecting and mitigating model misbehavior in general-purpose scenarios, often overlooking the potential of integrating domain-specific knowledge. In this work, we advance misbehavior detection by incorporating domain knowledge. The core idea is to design a general specification language that enables domain experts to customize domain-specific constraints in a lightweight and intuitive manner, supporting later runtime monitoring of LLM outputs. To achieve this, we design a novel specification language ESL, and introduce a runtime verification framework RvLLM to validate LLM output against domain-specific constraints defined in ESL. RvLLM operates in two main stages: interpretation and reasoning. During interpretation, it derives interpretations of the specification based on the context, which then guide the reasoning process to identify inconsistencies. When new knowledge is derived, RvLLM issues a follow-up query to the LLM to further verify the consistency. We evaluate RvLLM on three representative tasks: violation detection against Singapore Rapid Transit Systems Act, numerical comparison, and inequality solving. Experimental results show that RvLLM effectively detects erroneous outputs across various LLMs in a lightweight and flexible manner. The results reveal that despite their impressive capabilities, LLMs remain prone to low-level errors due to a lack of formal guarantees during inference, and our framework offers a potential long-term solution by leveraging expert domain knowledge to rigorously and efficiently verify LLM outputs.

In the rapidly evolving landscape of neural network security, the resilience of neural networks against bit-flip attacks (i.e., an attacker maliciously flips an extremely small amount of bits within its parameter storage memory system to induce harmful behavior), has emerged as a relevant area of research. Existing studies suggest that quantization may serve as a viable defense against such attacks. Recognizing the documented susceptibility of real-valued neural networks to such attacks and the comparative robustness of quantized neural networks (QNNs), in this work, we introduce BFAVerifier, the first verification framework designed to formally verify the absence of bit-flip attacks against QNNs or to identify all vulnerable parameters in a sound and rigorous manner. BFAVerifier comprises two integral components: an abstraction-based method and an MILP-based method. Specifically, we first conduct a reachability analysis with respect to symbolic parameters that represent the potential bit-flip attacks, based on a novel abstract domain with a sound guarantee. If the reachability analysis fails to prove the resilience of such attacks, then we encode this verification problem into an equivalent MILP problem which can be solved by off-the-shelf solvers. Therefore, BFAVerifier is sound, complete, and reasonably efficient. We conduct extensive experiments, which demonstrate its effectiveness and efficiency across various activation functions, quantization bit-widths, and adversary capabilities.

Quantization replaces floating point arithmetic with integer arithmetic in deep neural network models, providing more efficient on-device inference with less power and memory. In this work, we propose a framework for formally verifying the properties of quantized neural networks. Our baseline technique is based on integer linear programming which guarantees both soundness and completeness. We then show how efficiency can be improved by utilizing gradient-based heuristic search methods and also bound-propagation techniques. We evaluate our approach on perception networks quantized with PyTorch. Our results show that we can verify quantized networks with better scalability and efficiency than the previous state of the art.

Abstract Quantization plays an important role in deploying neural networks on embedded, real-time systems with limited computing and storage resources (e.g., edge devices). It significantly reduces the model storage cost and improves inference efficiency by using fewer bits to represent the parameters. However, it was recently shown that critical properties may be broken after quantization, such as robustness and backdoor-freeness. In this work, we introduce the first method for synthesizing quantization strategies that verifiably maintain desired properties after quantization, leveraging a key insight that quantization leads to a data distribution shift in each layer. We propose to compute the preimage for each layer based on which the preceding layer is quantized, ensuring that the quantized reachable region of the preceding layer remains within the preimage. To tackle the challenge of computing the exact preimage, we propose an MILP-based method to compute its under-approximation. We implement our method into a tool and demonstrate its effectiveness and efficiency by providing certified quantization that successfully preserves model robustness and backdoor-freeness.

Using multiple input streams simultaneously to train multimodal neural networks is intuitively advantageous but practically challenging. A key challenge is unimodal bias, where a network overly relies on one modality and ignores others during joint training. We develop a theory of unimodal bias with multimodal deep linear networks to understand how architecture and data statistics influence this bias. This is the first work to calculate the duration of the unimodal phase in learning as a function of the depth at which modalities are fused within the network, dataset statistics, and initialization. We show that the deeper the layer at which fusion occurs, the longer the unimodal phase. A long unimodal phase can lead to a generalization deficit and permanent unimodal bias in the overparametrized regime. Our results, derived for multimodal linear networks, extend to nonlinear networks in certain settings. Taken together, this work illuminates pathologies of multimodal learning under joint training, showing that late and intermediate fusion architectures can give rise to long unimodal phases and permanent unimodal bias. Our code is available at: https://yedizhang.github.io/unimodal-bias.html.

Collaborative software development significantly enhances development productivity by enabling multiple contributors to work concurrently on different branches. Despite these advantages, such collaboration often increases the likelihood of causing conflicts. Resolving these conflicts brings huge challenges, primarily due to the necessity of comprehending the differences between conflicting versions. Researchers have explored various automatic conflict resolution techniques, including unstructured, structured, and learning-based approaches. However, these techniques are mostly heuristic-based or black-box in nature, which means they do not attempt to solve the root cause of the conflicts, i.e., the existence of different program behaviors exhibited by the conflicting versions.

Abstract To alleviate the practical constraints for deploying deep neural networks (DNNs) on edge devices, quantization is widely regarded as one promising technique. It reduces the resource requirements for computational power and storage space by quantizing the weights and/or activation tensors of a DNN into lower bit-width fixed-point numbers, resulting in quantized neural networks (QNNs). While it has been empirically shown to introduce minor accuracy loss, critical verified properties of a DNN might become invalid once quantized. Existing verification methods focus on either individual neural networks (DNNs or QNNs) or quantization error bound for partial quantization. In this work, we propose a quantization error bound verification method, named , where both weights and activation tensors are quantized. consists of two parts, i.e., a differential reachability analysis (DRA) and a mixed-integer linear programming (MILP) based verification method. DRA performs difference analysis between the DNN and its quantized counterpart layer-by-layer to compute a tight quantization error interval efficiently. If DRA fails to prove the error bound, then we encode the verification problem into an equivalent MILP problem which can be solved by off-the-shelf solvers. Thus, is sound, complete, and reasonably efficient. We implement and conduct extensive experiments, showing its effectiveness and efficiency.

Masking is one of the most effective countermeasures for securely implementing cryptographic algorithms against power side-channel attacks, the design of which however turns out to be intricate and error-prone. While techniques have been proposed to rigorously verify implementations of cryptographic algorithms, currently they are limited in scalability. To address this issue, compositional approaches have been investigated, but insofar they fail to prove the security of recent efficient implementations. To fill this gap, we propose a novel compositional verification approach. In particular, we introduce two new language-level security notions based on which we propose composition strategies and verification algorithms. Our approach is able to prove efficient implementations, which cannot be done by prior compositional approaches. We implement our approach as a tool CONVINCE and conduct extensive experiments to confirm its efficacy. We also use CONVINCE to further explore the design space of the AES Sbox with least refreshing by replacing its implementation for finite-field multiplication with more efficient counterparts. We automatically prove leakage-freeness of these new versions. As a result, we can effectively reduce 1,600 randomness and 3,200 XOR-operations of the state-of-the-art AES implementation.

As a new programming paradigm, neural-network-based machine learning has expanded its application to many real-world problems. Due to the black-box nature of neural networks, verifying and explaining their behavior are becoming increasingly important, especially when they are deployed in safety-critical applications. Existing verification work mostly focuses on qualitative verification, which asks whether there exists an input (in a specified region) for a neural network such that a property (e.g., local robustness) is violated. However, in many practical applications, such an (adversarial) input almost surely exists, which makes a qualitative answer less meaningful. In this work, we study a more interesting yet more challenging problem, i.e.,quantitativeverification of neural networks, which asks how often a property is satisfied or violated. We target binarized neural networks (BNNs), the 1-bit quantization of general neural networks. BNNs have attracted increasing attention in deep learning recently, as they can drastically reduce memory storage and execution time with bit-wise operations, which is crucial in recourse-constrained scenarios, e.g., embedded devices for Internet of Things. Toward quantitative verification of BNNs, we propose a novel algorithmic approach for encoding BNNs as Binary Decision Diagrams (BDDs), a widely studied model in formal verification and knowledge representation. By exploiting the internal structure of the BNNs, our encoding translates the input-output relation of blocks in BNNs to cardinality constraints, which are then encoded by BDDs. Based on the new BDD encoding, we develop a quantitative verification framework for BNNs where precise and comprehensive analysis of BNNs can be performed. To improve the scalability of BDD encoding, we also investigate parallelization strategies at various levels. We demonstrate applications of our framework by providing quantitative robustness verification and interpretability for BNNs. An extensive experimental evaluation confirms the effectiveness and efficiency of our approach.

Deep learning has become a promising programming paradigm in software development, owing to its surprising performance in solving many challenging tasks. Deep neural networks (DNNs) are increasingly being deployed in practice, but are limited on resource-constrained devices owing to their demand for computational power. Quantization has emerged as a promising technique to reduce the size of DNNs with comparable accuracy as their floating-point numbered counterparts. The resulting quantized neural networks (QNNs) can be implemented energy-efficiently. Similar to their floating-point numbered counterparts, quality assurance techniques for QNNs, such as testing and formal verification, are essential but are currently less explored. In this work, we propose a novel and efficient formal verification approach for QNNs. In particular, we are the first to propose an encoding that reduces the verification problem of QNNs into the solving of integer linear constraints, which can be solved using off-the-shelf solvers. Our encoding is both sound and complete. We demonstrate the application of our approach on local robustness verification and maximum robustness radius computation. We implement our approach in a prototype tool QVIP and conduct a thorough evaluation. Experimental results on QNNs with different quantization bits confirm the effectiveness and efficiency of our approach, e.g., two orders of magnitude faster and able to solve more verification tasks in the same time limit than the state-of-the-art methods.

Abstract Verifying and explaining the behavior of neural networks is becoming increasingly important, especially when they are deployed in safety-critical applications. In this paper, we study verification and interpretability problems for Binarized Neural Networks (BNNs), the 1-bit quantization of general real-numbered neural networks. Our approach is to encode BNNs into Binary Decision Diagrams (BDDs), which is done by exploiting the internal structure of the BNNs. In particular, we translate the input-output relation of blocks in BNNs to cardinality constraints which are in turn encoded by BDDs. Based on the encoding, we develop a quantitative framework for BNNs where precise and comprehensive analysis of BNNs can be performed. We demonstrate the application of our framework by providing quantitative robustness analysis and interpretability for BNNs. We implement a prototype tool and carry out extensive experiments, confirming the effectiveness and efficiency of our approach.

Reasoning about strategic abilities is key to an AI system consisting of multiple agents with random behaviors. We propose a probabilistic extension of Alternating µ-Calculus (AMC), named PAMC, for reasoning about strategic abilities of agents in stochastic multi-agent systems. PAMC subsumes existing logics AMC and PµTL. The usefulness of PAMC is exemplified by applications in genetic regulatory networks. We show that, for PAMC, the model checking problem is in UP∩co-UP, and the satisfiability problem is EXPTIME-complete, both of which are the same as those for AMC. Moreover, PAMC admits the small model property. We implement the satisfiability checking procedure in a tool PAMCSolver.