Robustness verification is a promising technique for rigorously proving Recurrent Neural Networks (RNNs) robustly. A key challenge is to over-approximate the nonlinear activation functions with linear constraints, which can transform the verification problem into an efficiently solvable linear programming problem. Existing methods over-approximate the nonlinear parts with linear bounding planes individually, which may cause significant over-estimation and lead to lower verification accuracy. In this paper, in order to tightly enclose the three-dimensional nonlinear surface generated by the Hadamard product, we propose a novel truncated rectangular prism formed by two linear relaxation planes and a refinement-driven method to minimize both its volume and surface area for tighter over-approximation. Based on this approximation, we implement a prototype DeepPrism for RNN robustness verification. The experimental results demonstrate that DeepPrism has significant improvement compared with the state-of-the-art approaches in various tasks of image classification, speech recognition and sentiment analysis.

Dynamic Facial Expression Recognition (DFER), as a crucial part of affective computing, has broad applications in many areas such as human-computer interaction and social media content analysis. Effectively integrating multimodal information, particularly audio-visual signals, remains the core challenge. However, existing approaches are generally constrained by two major challenges: (1) shallow and static fusion mechanisms, which fail to capture the dynamic co-evolution of audio-visual features during deep interaction; (2) implicit and coarse alignment strategies, which are insufficient to bridge the modality gap caused by heterogeneous feature distributions. To address these issues, we propose a novel framework, BHGap, which integrates deep iterative prompt generation with multi-stage feature alignment and fusion. The key idea is to reformulate audio-visual collaboration from a one-shot fusion event into a continuous, reciprocal generation process that spans every layer of frozen backbone encoders. Specifically, we design a State Space Model (SSM)-based cross-modal prompt generator that dynamically produces ''guidance prompts'' for the counterpart modality at each encoding layer, thereby enabling deep and fine-grained feature co-evolution. Beyond encoding, we further introduce a coarse-to-fine multi-stage alignment module: at the macro level, low-rank adversarial alignment is employed to establish spatio-temporal congruity between audio and video while reducing global distributional discrepancies; at the micro level, Maximum Mean Discrepancy (MMD) constraints combined with implicit differentiation optimization ensure fine-grained statistical consistency and semantic alignment. Extensive experiments on the public DFEW and MAFW datasets demonstrate that our method achieves state-of-the-art performance, offering a new paradigm of deep iterative fusion and explicit alignment for multimodal emotion recognition. Code is available at https://github.com/NDYZD666/-public-BHGap.

Text-to-3D (T23D) generation has transformed digital content creation, yet remains bottlenecked by blind trial-and-error prompting processes that yield unpredictable results. While visual prompt engineering has advanced in text-to-image domains, its application to 3D generation presents unique challenges requiring multi-view consistency evaluation and spatial understanding. We present Sel3DCraft, a visual prompt engineering system for T23D that transforms unstructured exploration into a guided visual process. Our approach introduces three key innovations: a dual-branch structure combining retrieval and generation for diverse candidate exploration; a multi-view hybrid scoring approach that leverages MLLMs with innovative high-level metrics to assess 3D models with human-expert consistency; and a prompt-driven visual analytics suite that enables intuitive defect identification and refinement. Extensive testing and a user study demonstrate that Sel3DCraft surpasses other T23D systems in supporting creativity for designers.

Exception handling is crucial but challenging in program development. It needs to identify and handle all potential exceptions within programs to ensure system security and stabilization. Traditional exception handling relies on the expertise and experience of programmers, which often leads to oversights. Therefore, identifying exceptional code and recommending handling solutions are hot research topics with significant practical value. This paper presents a model called CodeHunter for exception localization and type prediction. The model first utilizes BERT-based model to represent code features and then uses Bi-LSTM for sequence labeling to pinpoint exceptional code. Additionally, this model also considers contextual features of the exception code and learns weights for the code within the try block and its context through the self-attention mechanism. Subsequently, it performs exception localization and predicts exception types. We conduct experiments on three different datasets. The results demonstrate that in the task of exception localization, our model can achieve a maximum accuracy of 98.6%, exceeding SOTA baselines by 11.2%. In the task of exception type prediction, our model can surpass the accuracy of SOTA baselines by a maximum of 18.7%, achieving 92.0% Top-1 accuracy. The rationality of techniques used in our model is also proved by the ablation testing. The model is implemented as an IDE plugin for programming convenience.

FinTech software, crucial for both safety and timely market deployment, presents a compelling case for automated acceptance testing against regulatory business rules. However, the inherent challenges of comprehending unstructured natural language descriptions of these rules and crafting comprehensive test cases demand human intelligence. The emergence of Large Language Models (LLMs) holds promise for automated test case generation, leveraging their natural language processing capabilities. Yet, their dependence on human intervention for effective prompting hampers efficiency. In response, we introduce a groundbreaking, fully automated approach for generating high-coverage test cases from natural language business rules. Our methodology seamlessly integrates the versatility of LLMs with the predictability of algorithmic methods. We fine-tune pre-trained LLMs for improved information extraction accuracy and algorithmically generate comprehensive testable scenarios for the extracted business rules. Our prototype, LLM4Fin, is designed for testing real-world stock-trading software. Experimental results demonstrate LLM4Fin’s superiority over both state-of-the-art LLM, such as ChatGPT, and skilled testing engineers. We achieve remarkable performance, with up to 98.18% and an average of 20%−110% improvement on business scenario coverage, and up to 93.72% on code coverage, while reducing the time cost from 20 minutes to a mere 7 seconds. These results provide robust evidence of the framework’s practical applicability and efficiency, marking a significant advancement in FinTech software testing.

Trigger-Action Programming (TAP) is a new programming paradigm enabling end-users to customize their smart devices by defining simple trigger-action rules. While it offers appealing convenience to end-users, TAP renders devices vulnerable to operation chaos and security risk resulting from potential defects in the rules. Verifying TAP rules defined by end-users is thereby necessary to detect such vulnerabilities at the early stage. However, such rules are difficult to analyze because their executions are often device-specific and environment-driven. Existing approaches require modeling them with their host devices and running environments, which is labor-consuming and hard to be automated. Moreover, the composition of devices causes state explosion, rendering the conflict analysis time-consuming. In this paper, we first build a large corpus of TAP rules developed by end-users. Analyzing this corpus results in six types of conflicts and reveals that nearly 90% of end-users made conflicts in their customized rules, and on average, 3.7 rules contain a conflict, which concurs with the necessity of developing practical conflict analysis techniques. Empirical analysis motivates us to propose a lightweight SMT-based approach for conflict analysis from a programmatic perspective. Compared to the existing approaches, our approach does not require modeling devices; thus, it could be fully automatic and flexible in efficiently detecting various types of conflicts. We implement the approach in a tool TapChecker. We analyze 12,514 TAP rules collected from real-world TAP platforms (10,535) and laboratory experiments (1,979). Experimental results show that our approach outperforms the state-of-the-art tool regarding the number of detected conflicts and efficiency.

Knowledge Tracing (KT) is a crucial task in the field of online education, since it aims to predict students' performance on exercises based on their learning history. One typical solution for knowledge tracing is to combine the classic models in educational psychology, such as Item Response Theory (IRT) and Cognitive Diagnosis (CD), with Deep Neural Networks (DNN) technologies. In this solution, a student and related exercises are mapped into feature vectors based on the student's performance at the current time step, however, it does not consider the impact of historical behavior sequences, and the relationships between historical sequences and students. In this paper, we develop DAKTN, a novel model which assimilates the historical sequences to tackle this challenge for better knowledge tracing. To be specific, we apply a pooling layer to incorporate the student behavior sequence in the embedding layer. After that, we further design a local activation unit, which can adaptively calculate the representation vectors by taking the relevance of historical sequences into consideration with respect to candidate student and exercises. Through experimental results on three real-world datasets, DAKTN significantly outperforms state-of-the-art baseline models. We also present the reasonableness of DAKTN by ablation testing.

A software system is usually developed with multiple modules.However, its structure is continuously modified during software evolution, resulting in poor maintainability and understandability.Therefore, software evolution must accompany system refactoring.This paper describes an optimization approach for package structure according to complex network theory.First, we analyze the relations between classes and build the class dependency graph.Second, we propose a community detection algorithm to recombine the classes and optimize system cohesion and coupling without changing the external functionality.Third, by comparing the original and optimized package structure, the two dimensions of splitting the package and moving classes between packages identify package refactoring opportunities.In addition, we evaluate the impact of the above approach on package quality in terms of package reusability and instability.We design experiments on 10 open-source Java software projects to verify the effectiveness of our approach.

Knowledge tracing is a crucial task in intelligent tutoring systems.Aiming at the shortcomings of traditional knowledge tracing technology such as low prediction accuracy, overfitting and low utilization of multi-features, this paper proposes a knowledge tracing model SRGCA-M using multi-feature embedding with stacked residual GRU network.Compared with the traditional methods that only use the historical record of answering exercises, our approach utilizes a variety of features in the learning process of students to deep characterize students' learning.We increase the layers number of GRU network to expand the capacity of sequence learning and use residual connections to solve the problems of network degradation and vanishing gradient.We use the auto-encoder to solve the problem that the cross-feature encoding will rapidly increase the dimension of the input data.Comprehensive experimental results demonstrate that compared with various advanced techniques, our approach can not only achieve better performance of tracking knowledge changes of students but also fully utilize multi-feature information of students in the learning process.

Cognitive diagnosis is a crucial task in the field of educational measurement and psychology, which is aimed to mine and analyze the level of knowledge for a student in his or her learning process periodically. While a number of approaches and tools have been developed to diagnose the learning states of students, they do not fully learn the relationship between students, exercises and knowledge concepts in the learning system, or do not consider the traits that it is easier to complete diagnosis when focusing on a small part of knowledge concepts rather than all knowledge concepts. To address these limitations, we develop CDGK, a model based artificial neural network to deal with cognitive diagnosis. Our method not only captures non-linear interactions between exercise features, student scores, and their mastery on each knowledge concept, but also performs an aggregation of the knowledge concepts via converting them into graph structure, and only considering the leaf node in the knowledge concept tree, which can reduce the dimension of the model without accuracy loss. In our evaluation on two real-world datasets, CDGK outperforms the state-of-the-art related approaches in terms of accuracy, reasonableness and interpretability.

Deep Neural Networks (DNNs) have achieved high accuracy on image classification. However, a small disturbance to an input may fool the networks to misclassify the label, which can cause a series of security and social problems. Thus, the robustness of DNNs must be ensured, particularly to those safety-critical systems. In this paper, we focus on the problem of measuring the robustness of ReLU-based DNNs, which can be equivalently formulated to solve a Mixed Integer Linear Programming problem (MILP). The complexity of solving MILP is directly related to the number of integer variables. We propose an efficient method for robustness measurement and verification by pruning the search space of MILP problems. Particularly, we design a greedy algorithm based on linear programming (LP) to determine the reasonable boundary. Then the search space is pruned by setting the boundary to integer variables in MILP. The comparison experiments on five classifiers trained on MNIST and CIFAR-10 datasets show our method outperforms other related tools in terms of efficiency and accuracy.

Formula retrieval is an important research topic in Mathematical Information Retrieval (MIR).Most studies have focused on comparing formulae to determine the similarity between mathematical documents.However, two similar formulae may appear in completely different knowledge domains and have different meanings.Based on N-ary Tree-based Formula Embedding Model (NTFEM), we introduce a new hybrid retrieval model combining formula with its surrounding text for more accurate retrieval.Using keywords extraction technology, we extract keywords from text around the formula which can supplement the semantic information of formula.Then we get the representation vectors of keywords by FastText N-gram embedding model, and the representation vectors of formulae by NTFEM.Finally, documents are first sorted according to the similarity of keywords, and then the ranking results are optimized by formula similarity.Experimental results show that the accuracy of top-10 results is at least 20% higher than that of NTFEM and can be 50% in some specific topics.

Software quality is critical in the current information age, we need to find effective modeling methods to evaluate the quality of increasingly large software. Complex network is an important tool for modelling software systems from the macro perspective, and it expresses software as a complex network by analyzing the dependencies between software components. In this paper, we propose a new model called weighted class dependency network (WCDN) based on complex network theory to describe the characteristics of object-oriented software systems. WCDN model is based on class granularity, which considers not only the difference of multiple dependencies caused by object-oriented programming languages, but also the disparity of class importance. We apply WCDN to a refactoring algorithm based on three modular metrics, and evaluate the rationality and effectiveness of the network through disturbing-recovering experiments. After the comparative experiments of 12 open-source software systems, the results prove that WCDN can represent the architecture of the object-oriented software system correctly.

Formula retrieval is an important research topic in Mathematical Information Retrieval (MIR). Most studies have focused on formula comparison to determine the similarity between mathematical documents. However, two similar formulae may appear in entirely different knowledge domains and have different meanings. Based on N-ary Tree-based Formula Embedding Model (NTFEM, our previous work in [Y. Dai, L. Chen, and Z. Zhang, An N-ary tree-based model for similarity evaluation on mathematical formulae, in Proc. 2020 IEEE Int. Conf. Systems, Man, and Cybernetics, 2020, pp. 2578–2584.], we introduce a new hybrid retrieval model, NTFEM-K, which combines formulae with their surrounding keywords for more accurate retrieval. By using keywords extraction technology, we extract keywords from context, which can supplement the semantic information of the formula. Then, we get the vector representations of keywords by FastText N-gram embedding model and the vector representations of formulae by NTFEM. Finally, documents are sorted according to the similarity between keywords, and then the ranking results are optimized by formula similarity. For performance evaluation, NTFEM-K is not only compared with NTFEM but also hybrid retrieval models combining formulae with long text and hybrid retrieval models combining formulae with their keywords using other keyword extraction algorithms. Experimental results show that the accuracy of top-10 results of NTFEM-K is at least 20% higher than that of NTFEM and can be 50% in some specific topics.

Accurate and efficient measurements for evaluating the similarity between mathematical formulae play an important role in mathematical information retrieval. Most previous studies have focused on representing formulae in different types to catch their features and combining the traditional structure matching algorithms. This paper presents a new unsupervised model called N-ary Tree-based Formula Embedding Model (NTFEM) for the task of mathematical similarity evaluation. Using an n-ary tree structure to represent the formula, we convert the formula into a linear sequence that can be viewed as the input sentence and then embed the formula by using a word embedding model. Based on the characteristics of mathematical formulae, a weighting function is also used to get the final weighted average embedding vector. Through some experiments on NTCIR-12 Wikipedia Formula Browsing Task, our model can outperform previous formula search engines in Bpref prediction metrics. In addition, compared with traditional tree-based models, NTFEM not only improves the retrieval effect, but also greatly reduces the training time and improves training efficiency.

We consider prime fields of large characteristic, typically fitting on $k$ machine words, where k is a power of 2. When the characteristic of these fields is restricted to a subclass of the generalized Fermat numbers, we show that arithmetic operations in such fields offer attractive performance, both in terms of algebraic complexity and parallelism. In particular, these operations can be vectorized, leading to efficient implementation of fast Fourier transforms on graphics processing units.