Reinforcement learning (RL) has shown promise for enhancing code generation capabilities in large language models (LLMs), yet its effectiveness critically depends on high-quality test suites for reliable reward signals. Current approaches suffer from inadequate test case quantity and quality, leading to false positives (incorrect solutions passing verification) and slow positives (valid but suboptimal implementations), which corrupt RL training dynamics. We address these challenges through three key contributions: (1) We systematically analyze how low-quality test suites degrade Code RL performance via reward misalignment; (2) We propose Themis, an automated framework that transforms test case generation into code synthesis—first extracting problem constraints via template-guided parsing, then generating executable test generators through LLM-powered code synthesis, and finally validating tests through constraint-aware filtering; (3) We develop an error-guided test case reduction method that preserves error detection efficacy while reducing test set cardinality, thereby enhancing reinforcement learning training efficiency. Evaluated on programming competition datasets, Themis achieves 95 percent error detection rates, outperforming original test suites in most of the cases. When integrated into RL pipelines, models trained with Themis-generated tests demonstrate consistent 3-5 percent improvements across HumanEval, MBPP, and LiveCodeBench compared to the baseline, matching performance levels achieved with manually curated test suites. Our constraint-aware test synthesis framework ensures full automation while preserving semantic validity—critical for scaling RL training to complex code generation tasks. The framework's modular design also enables seamless integration with existing code data synthesis frameworks.

Given the immutability of biometric data, it is imperative to develop a biometric template protection method that guarantees the complete non-disclosure of any original biometric information while ensuring high recognition performance. Two mainstream approaches in biometric template protection—cancelable biometrics and biometric cryptosystems—have been widely adopted; however, the protected templates produced by these methods still contain some of the original biometric data, which can lead to privacy leakage. To address these challenges, we propose a novel framework named BioDeepHash that integrates deep hashing with cryptographic hash functions. In our approach, a deep hashing model generates consistent templates for similar biometric data from the same user, thereby eliminating intra-class variations. An application-specific XOR string is then used to achieve revocability, and finally these consistent templates are processed by cryptographic hash functions to produce protected templates that meet strict security standards. Our experimental results show that, compared with existing methods, BioDeepHash increases the average Genuine Acceptance Rate by 10.12% on the iris dataset and by 3.12% on the facial dataset, while achieving an extremely low False Acceptance Rate—0% for the iris dataset and only 0.0002% for the facial dataset.

The Vehicle Routing Problem with Time Windows under Uncertain Travel Times (VRPTW-UT) is a challenging and practically significant combinatorial optimization problem. Although evolutionary algorithms (EAs) have shown potential in solving VRPTW-UT, they often struggle to balance robustness and convergence. Conventional EA approaches evaluate solutions across multiple disturbance scenes and discard those that become infeasible under any scenario. This often leads to the premature elimination of solutions that are only infeasible in a limited number of scenes, hindering the ability to effectively explore the trade-off between robustness and convergence. To address this issue, this paper proposes a Game-Theoretic Evolutionary Algorithm (GTEA) that models the search process as a game between two adversarial components: a perturbation generation part that constructs high-impact uncertainty scenes, and a robustness enhancement part that improves solutions under those critical conditions. This antagonistic process forces the population to evolve toward solutions that possess both high robustness and convergence, so that GTEA can efficiently produce solutions with high robustness and convergence. Extensive experiments on four benchmark datasets demonstrate that GTEA outperforms five state-of-the-art algorithms designed for VRPTW-UT, achieving superior convergence and robustness.

The capacitated electric vehicle routing problem (CEVRP) has drawn much attention from researchers in the recent decade against the background of the rising electric transportation industry. Existing studies have found the CEVRP more difficult to address than the typical CVRP since the CEVRP needs to simultaneously optimize routing plans and charging decisions at a high computational budget. Given a certain routing plan, it takes much computational cost to exhaustively or approximately achieve the accurate optimal charging decision under the routing plan. This paper proposes that it is unnecessary to search for the accurate optimal charging decisions for potentially low-quality routing plans found during the CEVRP optimization, and instead obtaining acceptable charging decisions significantly reduces the computational cost and helps maintain fast convergence. A multifidelity-based ant colony optimization (MFACO) algorithm is then proposed to flexibly search charging decisions based on the potential quality of routing plans so that CEVRPs can be addressed at high efficiency. The proposed MFACO employs a low-fidelity search strategy to obtain coarse charging decisions for potentially low-quality routing plans, whereas for potentially high-quality routing plans MFACO employs three high-fidelity search strategies to local search in different regions of decision space and provide accurate optimal charging decisions. Experimental results demonstrate that the proposed multifidelity method enhances the efficiency of ACO in solving CEVRPs and the proposed MFACO significantly outperforms four state-of-the-art algorithms for CEVRPs, providing a competitive performance in terms of both solution quality and computational cost.

Instruction-tuned Code Large Language Models (Code LLMs) have excelled in diverse code-related tasks, such as program synthesis, automatic program repair, and code explanation. To collect training datasets for instruction-tuning, a popular method involves having models autonomously generate instructions and corresponding responses. However, the direct generation of responses does not ensure functional correctness, a crucial requirement for generating responses to code instructions. To overcome this, we present Verification-Based Self-Play (VERSE), aiming to enhance model proficiency in generating correct responses. VERSE establishes a robust verification framework that covers various code instructions. Employing VERSE, Code LLMs engage in self-play to generate instructions and corresponding verifications. They evaluate execution results and self-consistency as verification outcomes, using them as scores to rank generated data for self-training. Experiments show that VERSE improves multiple base Code LLMs (average 7.6%) across various languages and tasks on many benchmarks, affirming its effectiveness.

The Capacitated Vehicle Routing Problems with Three-Dimensional Loading Constraints (3L-CVRPs) present significantly greater complexities when compared to the classical Capacitated Vehicle Routing Optimization Problems (CVRPs). This heightened complexity stems from the necessity to simultaneously address two intertwined objectives: optimizing vehicle routing and cargo loading configurations to maximize loading efficiency, all while adhering to stringent vehicle dimensional constraints (length, width, and height). Both of these subproblems are recognized within the computational complexity class of NP-hard problems. Despite the proven efficacy of evolutionary algorithms in addressing 3L-CVRPs, prevalent methodologies typically adopt a sequential approach, prioritizing route op-timization over loading optimization, thereby overlooking the intrinsic interconnectedness between these two facets. This paper proposes a novel dual-population evolutionary algorithm in response to this limitation. One population is dedicated to refining vehicle routing solutions, whereas the other focuses on enhancing loading efficiency. These two populations undergo a co-evolutionary process to identify optimal solutions for 3L-CVRPs. Empirical evaluations across four diverse datasets reveal that the proposed algorithm surpasses three contemporary algorithms, achieving superior loading efficiency and reduced transportation costs.

Large language models have been successfully applied to programming assistance tasks, such as code completion, code insertion, and instructional code editing. However, these applications remain insufficiently automated and struggle to effectively integrate various types of information during the programming process, including coding history, code context, and user instructions. In this work, we propose a new framework that comprehensively integrates these information sources, collect data to train our models and evaluate their performance. Firstly, to thoroughly evaluate how well models align with different types of information and the quality of their outputs, we introduce a new benchmark, APEval (Assist Programming Eval), to comprehensively assess the performance of models in programming assistance tasks. Then, for data collection, we develop a data generation pipeline, Programming-Instruct, which synthesizes training data from diverse sources, such as GitHub and online judge platforms. This pipeline can automatically generate various types of messages throughout the programming process. Finally, using this pipeline, we generate 219K samples, fine-tune multiple models, and develop the CursorCore series. We show that CursorCore outperforms other models of comparable size. This framework unifies applications such as inline chat and automated editing, contributes to the advancement of coding assistants.

The Dynamic Vehicle Routing Problem with Time Windows (DVRPTW) is a widespread real-world challenge, and numerous algorithms have been proposed to address it. However, in the context of an emerging logistics paradigm, namely the instant delivery, the performance of existing algorithms tailored for DVRPTW degrades significantly, as instant delivery allows only very limited computational time for solving DVRPTW instances. Owing to the periodic nature of customer orders, this paper proposes a data-driven evolutionary algorithm (DDEA) for solving DVRPTW under limited computation time. In the offline phase, a set of generalized solutions is derived from historical data via a dedicated evolutionary algorithm. These solutions are then directly employed in the online phase to construct high-quality solutions for new problem instances. By leveraging these precomputed generalized solutions, DDEA effectively operates within tight time constraints. Extensive experiments using synthetic and real-world data demonstrate that DDEA outperforms five state-of-the-art algorithms designed for DVRPTW under limited computation time, particularly under extremely short time constraints.

The Multi-depot Vehicle Routing Problem with Uncertain Demand (MD-VRPUD) can be modeled as a bi-level optimization problem (BLOP), because it requires both assigning customers to different depots and determining the routes for servicing customers, where the optimization of these two parts is coupled with each other. Although the bi-level evolutionary algorithm is a fitting approach for tackling the MD-VRPUD, its nested structure often leads to computational inefficiency. To this end, this paper tailors a surrogate-assisted bi-level evolutionary algorithm (SABLEA) to achieve highly efficient nested algorithms tailored for solving the MD-VRPUD. To deal with the combinatorial property of MD-VRPUD, two groups of continuous features are first extracted to help the surrogate model to effectively distinguish the superiority and inferiority of schemes. Then, a management strategy is designed to adaptively build the surrogate models in different subspaces so as to alleviate the performance bottleneck faced by the model. Finally, an adaptive computing resource allocation strategy is integrated into the lower-level optimization, to allocate more resources to promising customer assignment schemes, enabling the discovery of better routes and improving the overall accuracy of models. The comprehensive experimental results demonstrate the effectiveness of the SABLEA in handling MD-VRPUD, outperforming four existing algorithms in terms of both computational efficiency and solution quality.

Change-proneness prediction aims to identify source files that deserve developers' attention for early quality improvement, ultimately reducing future maintenance costs. Existing studies have utilized diverse change features and machine learning techniques to construct effective prediction models. However, these approaches do not leverage the grouping of change features for the learning of feature representations; and the roles of dependency files in influencing the change-proneness of the target file are not adequately exploited. To this end, in this paper, we propose a novel approach called DeepCP, which utilizes a deep learning model to predict the change-proneness of source files. The key rationale of DeepCP is that the grouping of change features should be considered when constructing deep learning models, given that feeding the sequence of features directly into models has a detrimental impact on the learning of feature representations. To achieve this, DeepCP utilizes different subnetworks to learn different categories of features and merges the resulting representations with a multi-head attention network to generate the source file representation. Another key rationale of DeepCP is that the features of dependency files should be exploited because the change-proneness of dependency files influences that of the target file due to ripple effects. To accomplish this, DeepCP first learns the representations of the target file and its structurally dependent files and then integrates the resulting representations with an attention network. Our evaluation results demonstrate that DeepCP outperforms the state-of-the-art approach in predicting change-proneness.

The purpose of this study was to develop an evolutionary algorithm (EA) with bilevel surrogate modeling, called BL-SAEA, for tackling bilevel optimization problems (BLOPs), in which an upper level problem is to be solved subject to the optimality of a corresponding lower level problem. The motivation of this article is that the extensive lower level optimization required by each upper level solution consumes too many function evaluations, leading to poor optimization performance of EAs. To this end, during the upper level optimization, the BL-SAEA builds an upper level surrogate model to select several promising upper level solutions for the lower level optimization. Because only a small number of upper level solutions require the lower level optimization, the number of function evaluations can be considerably reduced. During the lower level optimization, the BL-SAEA constructs multiple lower level surrogate models to initialize the population of the lower level optimization, thus further decreasing the number of function evaluations. Experimental results on two widely used benchmarks and two real-world BLOPs demonstrate the superiority of our proposed algorithm over six state-of-the-art algorithms in terms of effectiveness and efficiency.

Multiobjective optimization problems (MOPs) with expensive constraints pose stiff challenges to existing surrogate-assisted evolutionary algorithms (SAEAs) in a very limited computational cost, due to the fact that the number of expensive constraints for an MOP is often large. For existing SAEAs, they always approximate constraint functions in a single granularity, namely, approximating the constraint violation (CV, coarse-grained) or each constraint (fine-grained). However, the landscape of CV is often too complex to be accurately approximated by a surrogate model. Although the modeling of each constraint function may be simpler than that of CV, approximating all the constraint functions independently may result in tremendous cumulative errors and high computational costs. To address this issue, in this article, we develop a multigranularity surrogate modeling framework for evolutionary algorithms (EAs), where the approximation granularity of constraint surrogates is adaptively determined by the position of the population in the fitness landscape. Moreover, a dedicated model management strategy is also developed to reduce the impact resulting from the errors introduced by constraint surrogates and prevent the population from trapping into local optima. To evaluate the performance of the proposed framework, an implementation called K-MGSAEA is proposed, and the experimental results on a large number of test problems show that the proposed framework is superior to seven state-of-the-art competitors.

Sparse large-scale multi-objective optimization problems (LSMOPs) widely exist in real-world applications. Due to the characteristics of high-dimensional search space and sparse Pareto optimal solutions, most existing multi-objective evolutionary algorithms encounter difficulties in solving this kind of optimization problems. In this paper, a sparsity-guided elitism co-evolutionary framework, namely SGECF, is proposed. At each generation, SGECF first performs k-means clustering on non-dominated solutions to calculate the present best sparsity. Then, SGECF divides the population into a winner subpopulation and two loser subpopulations based on the Pareto dominance relationship and the present best sparsity. In the process of offspring reproduction, the winner subpopulation generates offspring solutions using conventional genetic operators, and the two loser subpopulations reproduce offspring solutions under the guidance of the present best sparsity and decision variable importance. In the experiments, the performance of SGECF is investigated on eight benchmark problems and a real-world application. Experimental results show that SGECF is superior over four state-of-the-art algorithms.

User modeling techniques profile users' latent characteristics (e.g., preference) from their observed behaviors, and play a crucial role in decision-making. Unfortunately, traditional user models may unconsciously capture biases related to sensitive attributes (e.g., gender) from behavior data, even when this sensitive information is not explicitly provided. This can lead to unfair issues and discrimination against certain groups based on these sensitive attributes. Recent studies have been proposed to improve fairness by explicitly decorrelating user modeling results and sensitive attributes. However, most existing approaches assume that fully sensitive attribute labels are available in the training set, which is unrealistic due to collection limitations like privacy concerns, and hence bear the limitation of performance. In this paper, we focus on a practical situation with limited sensitive data and propose a novel FairLISA framework, which can efficiently utilize data with known and unknown sensitive attributes to facilitate fair model training. We first propose a novel theoretical perspective to build the relationship between data with both known and unknown sensitive attributes with the fairness objective. Then, based on this, we provide a general adversarial framework to effectively leverage the whole user data for fair user modeling. We conduct experiments on representative user modeling tasks including recommender system and cognitive diagnosis. The results demonstrate that our FairLISA can effectively improve fairness while retaining high accuracy in scenarios with different ratios of missing sensitive attributes.

The generation of SAT instances is an important issue in computer science, and it is useful for researchers to verify the effectiveness of SAT solvers. Addressing this issue could inspire researchers to propose new search strategies. SAT problems exist in various real-world applications, some of which have more than one solution. However, although several algorithms for generating random SAT instances have been proposed, few can be used to generate hard instances that have multiple predefined solutions. In this paper, we propose the KHidden-M algorithm to generate SAT instances with multiple predefined solutions that could be hard to solve by the local search strategy when the number of predefined solutions is small enough and the Hamming distance between them is not less than half of the solution length. Specifically, first, we generate an SAT instance that is satisfied by all of the predefined solutions. Next, if the generated SAT instance does not satisfy the hardness condition, then a strategy will be conducted to adjust clauses through multiple iterations to improve the hardness of the whole instance. We propose three strategies to generate the SAT instance in the first part. The first strategy is called the random strategy, which randomly generates clauses that are satisfied by all of the predefined solutions. The other two strategies are called the estimating strategy and greedy strategy, and using them, we attempt to generate an instance that directly satisfies or is closer to the hardness condition for the local search strategy. We employ two SAT solvers (i.e., WalkSAT and Kissat) to investigate the hardness of the SAT instances generated by our algorithm in the experiments. The experimental results show the effectiveness of the random, estimating and greedy strategies. Compared to the state-of-the-art algorithm for generating SAT instances with predefined solutions, namely, M-hidden, our algorithm could be more effective in generating hard SAT instances.

Evolutionary algorithms and mathematical programming methods are currently the most popular optimizers for solving continuous optimization problems. Owing to the population based search strategies, evolutionary algorithms can find a set of promising solutions without using any problem-specific information. By contrast, with the assistance of gradient and other information of the functions, mathematical programming methods can quickly converge to a single optimum. While these two types of optimizers have their own advantages and disadvantages, the performance comparison between them is rarely touched. It is known that gradient descent methods generally converge faster than evolutionary algorithms, but when can evolutionary algorithms outperform gradient descent methods? How is the scalability of them? To answer these questions, this paper first gives a review of popular evolutionary algorithms and mathematical programming methods, then conducts several experiments to compare their performance from various aspects, and finally draws some conclusions.

Identifying functional modules in protein-protein interaction (PPI) networks elucidates cellular organization and mechanism. Various methods have been proposed to identify the functional modules in PPI networks, but most of these methods do not consider the noisy links in PPI networks. They achieve a competitive performance on the PPI networks without noisy links, but the performance of these methods considerably deteriorates in the noisy PPI networks. Furthermore, the noisy links are inevitable in the PPI networks. In this paper, we propose a novel link-driven label propagation algorithm (LLPA) to identify functional modules in PPI networks. The LLPA first find link clusters in PPI networks, and then the functional modules are identified from the link clusters. Two strategies aimed to ensure the robustness of LLPA are proposed. One strategy involves the proposed LLPA updating the link labels in accordance with the designed weight of the link, which can reduce the incidence of noisy links. The other strategy involves the filtration of some noisy labels from the link clusters to further reduce the influence of noisy links. The performance evaluation on three real PPI networks shows that LLPA outperforms other eight state-of-the-art detection algorithms in terms of accuracy and robustness.

The algorithm recommendation is attracting increasing attention in solving real-world capacitated vehicle routing problems (CVRPs), due to the fact that existing meta-heuristic algorithms often show different performances on different CVRPs. To effectively perform algorithm recommendation for CVRPs, it becomes vital to extract suitable features to characterize the CVRPs accurately. To this end, in this article three groups of penetrating features are proposed to capture the characteristics of CVRPs. The first group consists of some basic features of CVRPs, where several features are suggested to capture the distribution of customer demand, the relationship between customer demand and vehicle capacity, besides some common attributes widely used in CVRPs. The second group is composed of the features extracted from some CVRP solutions generated by local search, where in addition to the feasible and better solutions, the worse solutions and the distribution of travel cost are also used to measure the sensitivity of CVRPs to local search operations. The third group is made up of image features obtained by depicting CVRP instances through images, which is first introduced by us to enhance the generalization of algorithm recommendation. Furthermore, based on the three groups of features, an algorithm recommendation method called ARM-I is built on the basis of a KNN classifier to recommend suitable algorithm for CVRPs. Experimental results on several selected benchmarks demonstrate the effectiveness of the designed features. More interestingly, the proposed ARM-I shows high generalization on real-world instances.

Dynamic social networking has always been the focus of the social network research, and a large number of effective community detection algorithms have been developed. However, as the real social networks are difficult to access, it is also difficult to evaluate the effectiveness of the community detection algorithms on dynamic social networks. Existing dynamic social network generators only focus on edge or node changes, whereas the quality of the community structure (modularity) is not considered. We propose a time-evolving social network generator based on modularity (TESNG-M). In TESNG-M, according to the community partition of the original network, the evolutionary behavior is simulated by adding or deleting nodes and flipping edges so that a static social network with a specified modularity will be generated. By repeating the static generation process, we obtain a dynamic social network with a specified partition and modularity at each time step. Thus, the network generated by TESNG-M can effectively simulate a real dynamic social network and be used for community detection. Furthermore, the specified modularity of static synthetic networks and the dynamic modularity of dynamic synthetic networks could be regarded as the performance baseline of community detection algorithms in static and dynamic social networks, respectively.

Secure password storage is a vital aspect in systems based on password authentication, which is still the most widely used authentication technique, despite some security flaws. In this paper, we propose a password authentication framework that is designed for secure password storage and could be easily integrated into existing authentication systems. In our framework, first, the received plain password from a client is hashed through a cryptographic hash function (e.g., SHA-256). Then, the hashed password is converted into a negative password. Finally, the negative password is encrypted into an encrypted negative password (ENP) using a symmetric-key algorithm (e.g., AES), and multi-iteration encryption could be employed to further improve security. The cryptographic hash function and symmetric encryption make it difficult to crack passwords from ENPs. Moreover, there are lots of corresponding ENPs for a given plain password, which makes precomputation attacks (e.g., lookup table attack and rainbow table attack) infeasible. The algorithm complexity analyses and comparisons show that the ENP could resist lookup table attack and provide stronger password protection under dictionary attack. It is worth mentioning that the ENP does not introduce extra elements (e.g., salt); besides this, the ENP could still resist precomputation attacks. Most importantly, the ENP is the first password protection scheme that combines the cryptographic hash function, the negative password, and the symmetric-key algorithm, without the need for additional information except the plain password.

The negative survey could preserve the privacy of individuals when collecting sensitive information, which has received wide attention. The existing reconstruction methods of negative surveys are usually designed for the single question negative surveys. Although the reconstruction methods of single question negative surveys could also be applied to multiquestion negative surveys after minor modifications, the consistency between the joint distribution and the marginal distributions in the multiquestion reconstructed results has never been considered. In this paper, the concept of consistency in the multiquestion reconstructed results is defined. Then the consistency of the reconstructed results obtained by two typical reconstruction methods, i.e., NStoPS and NStoPS-I, is analyzed. Furthermore, a novel reconstruction method for multiquestion negative survey is proposed, which could return consistent and reasonable multiquestion reconstructed results. Finally, the simulated experimental results on collecting healthcare data also illustrate that the performance of our method is excellent in terms of accuracy.

Most of the community detection methods require the global information of the original network to be available, however, it is often expensive (even no way) to obtain the global information of the network in many real-world networks. So, the local community detection, only based on the local information, becomes especially important. The local community is the community in the network to which a given starting node belongs. Some local community detection methods have been proposed. However, these methods did not consider the characteristics of the local community during the local community formation. In this paper, we analyze the formation of the local community and propose two local community detection algorithms based on the dynamic membership function. Each of the algorithms is divided into three stages: 1) the initial stage, 2) the middle stage, and 3) the closing stage. At the initial stage, we design a dynamical membership function to detect local community and nodes with the greatest neighborhood intersect rate could be added to the local community. At the middle stage, we design another dynamical membership function, and the goal of this stage is to make the connection of the node in the local community closest. At the closing stage, the third dynamical membership function is provided, and the local community is further improved by collecting some nodes that should not be omitted. We test our algorithms on several synthetic datasets and real datasets; the results show that the local communities detected by our method are closer to the real local communities.

Currently, many intelligent transportation systems (ITSs) require aggregation of location information. Although users enjoy the convenience of ITSs, privacy concerns have resulted in user's caution in offering location information. Certain studies have been performed to preserve user location privacy. However, most studies that focus on preserving location privacy require a trusted third party or do not consider the movements of users. In this paper, we propose a method based on negative surveys that can be used to estimate the number of people in geographic locations. This method, which can preserve user privacy regardless of user movements, adopts a simple negative survey algorithm for user devices and an estimation algorithm for the server and might be suitable for low-power mobile devices. The experimental results demonstrate that our method is capable of locating the people's gathering places with fine control granularity, which renders it a promising application.