Search

Code executability plays a vital role in software debugging and testing (e.g., detecting runtime exceptions or assertion violations). However, code execution, especially partial or arbitrary code execution, is a non-trivial task due to missing definitions and complex third-party dependencies. To make partial code (such as code snippets posted on the web or code fragments deep inside complex software projects) executable, the existing study has proposed a machine learning model to predict the undefined element types and inject the pre-defined dummy values into execution. However, the performance of their tool is limited due to its simply designed dummy values and the inability to continue learning. In this paper, we design and implement a novel framework, named SelfPiCo (Self Guided Partial Code Executor), to dynamically guide partial code execution by incorporating the open-source LLM (i.e., Code Llama) within an interactive loop. Particularly, SelfPiCo leverages few-shot in-context learning and chain-of-thought reasoning to elicit human knowledge and logical reasoning based on fine-tuning the Code Llama model. SelfPiCo continuously learns from code execution results and refines its predictions step after step. Our evaluations demonstrate that SelfPiCo can execute 72.7% and 83.3% of all lines in the open-source code and Stack Overflow snippets, outperforming the most recent state-of-the-art Lexecutor by 37.9% and 33.5%, respectively. Moreover, SelfPiCo successfully detected 18 and 33 runtime type error issues by executing the partial code from eight GitHub software projects and 43 Stack Overflow posts, demonstrating the practical usage and potential application of our framework in practice., Comment: Accepted by ISSTA'24

The increasing complexity and high costs associated with modern processor design have led to a surge in demand for processor design automation. Instruction-tuned large language models (LLMs) have demonstrated remarkable performance in automatically generating code for general-purpose programming languages like Python. However, these methods fail on hardware description languages (HDLs) like Verilog due to the scarcity of high-quality instruction tuning data, as even advanced LLMs like GPT-3.5 exhibit limited performance on Verilog generation. Regarding this issue, we observe that (1) Verilog code collected from the real world has higher quality than those generated by LLMs. (2) LLMs like GPT-3.5 excel in summarizing Verilog code rather than generating it. Based on these observations, this paper introduces CodeV, a series of open-source instruction-tuned Verilog generation LLMs. Instead of generating descriptions first and then getting the corresponding code from advanced LLMs, we prompt the LLM with Verilog code and let the LLM generate the corresponding natural language description by multi-level summarization. Experimental results show that CodeV relatively surpasses the previous open-source SOTA by 14.4% (BetterV in VerilogEval) and 11.3% (RTLCoder in RTLLM) respectively, and also relatively outperforms previous commercial SOTA GPT-4 by 22.1% in VerilogEval., Comment: 16 pages, 8 figures, conference

In this work, we propose a new approach called ``stationary reduction method based on nonisospectral deformation of orthogonal polynomials" for deriving discrete Painlev\'{e}-type (d-P-type) equations. We successfully apply this approach to (bi)orthogonal polynomials satisfying ordinary orthogonality, $(1,m)$-biorthogonality, generalized Laurent biorthogonality, Cauchy biorthogonality and partial-skew orthogonality. As a result, several seemingly novel classes of high order d-P-type equations, along with their particular solutions and respective Lax pairs, are derived. Notably, the d-P-type equation related to the Cauchy biorthogonality can be viewed as a stationary reduction of a nonisospectral generalization involving the first two flows of the Toda hierarchy of CKP type. Additionally, the d-P-type equation related to the partial-skew orthogonality is associated with the nonisospectral Toda hierarchy of BKP type, and this equation is found to admit a solution expressed in terms of Pfaffians., Comment: 88 pages; Several typos are corrected

Heterogeneous collaborative computing with NPU and CPU has received widespread attention due to its substantial performance benefits. To ensure data confidentiality and integrity during computing, Trusted Execution Environments (TEE) is considered a promising solution because of its comparatively lower overhead. However, existing heterogeneous TEE designs are inefficient for collaborative computing due to fine and different memory granularities between CPU and NPU. 1) The cacheline granularity of CPU TEE intensifies memory pressure due to its extra memory access, and 2) the cacheline granularity MAC of NPU escalates the pressure on the limited memory storage. 3) Data transfer across heterogeneous enclaves relies on the transit of non-secure regions, resulting in cumbersome re-encryption and scheduling. To address these issues, we propose TensorTEE, a unified tensor-granularity heterogeneous TEE for efficient secure collaborative tensor computing. First, we virtually support tensor granularity in CPU TEE to eliminate the off-chip metadata access by detecting and maintaining tensor structures on-chip. Second, we propose tensor-granularity MAC management with predictive execution to avoid computational stalls while eliminating off-chip MAC storage and access. Moreover, based on the unified granularity, we enable direct data transfer without re-encryption and scheduling dilemmas. Our evaluation is built on enhanced Gem5 and a cycle-accurate NPU simulator. The results show that TensorTEE improves the performance of Large Language Model (LLM) training workloads by 4.0x compared to existing work and incurs only 2.1% overhead compared to non-secure training, offering a practical security assurance for LLM training., Comment: Accepted by ASPLOS 2024

Recent advancements in open-source code large language models (LLMs) have demonstrated remarkable coding abilities by fine-tuning on the data generated from powerful closed-source LLMs such as GPT-3.5 and GPT-4 for instruction tuning. This paper explores how to further improve an instruction-tuned code LLM by generating data from itself rather than querying closed-source LLMs. Our key observation is the misalignment between the translation of formal and informal languages: translating formal language (i.e., code) to informal language (i.e., natural language) is more straightforward than the reverse. Based on this observation, we propose INVERSE-INSTRUCT, which summarizes instructions from code snippets instead of the reverse. Specifically, given an instruction tuning corpus for code and the resulting instruction-tuned code LLM, we ask the code LLM to generate additional high-quality instructions for the original corpus through code summarization and self-evaluation. Then, we fine-tune the base LLM on the combination of the original corpus and the self-generated one, which yields a stronger instruction-tuned LLM. We present a series of code LLMs named InverseCoder, which surpasses the performance of the original code LLMs on a wide range of benchmarks, including Python text-to-code generation, multilingual coding, and data-science code generation.

Large language models (LLMs) achieve promising results in code generation based on a given natural language description. They have been integrated into open-source projects and commercial products to facilitate daily coding activities. The natural language description in the prompt is crucial for LLMs to comprehend users' requirements. Prior studies uncover that LLMs are sensitive to the changes in the prompts, including slight changes that look inconspicuous. However, the natural language descriptions often vary in real-world scenarios (e.g., different formats, grammar, and wording). Prior studies on the robustness of LLMs are often based on random perturbations and such perturbations may not actually happen. In this paper, we conduct a comprehensive study to investigate how are code LLMs robust to variations of natural language description in real-world scenarios. We summarize 18 categories of perturbations of natural language and 3 combinations of co-occurred categories based on our literature review and an online survey with practitioners. We propose an automated framework, NLPerturbator, which can perform perturbations of each category given a set of prompts. Through a series of experiments on code generation using six code LLMs, we find that the perturbed prompts can decrease the performance of code generation by a considerable margin (e.g., up to 21.2%, and 4.8% to 6.1% on average). Our study highlights the importance of enhancing the robustness of LLMs to real-world variations in the prompts, as well as the essentiality of attentively constructing the prompts.

The increasing use of Large Language Models (LLMs) in software development has garnered significant attention from researchers assessing the quality of the code they generate. However, much of the research focuses on controlled datasets such as HumanEval, which fail to adequately represent how developers actually utilize LLMs' code generation capabilities or clarify the characteristics of LLM-generated code in real-world development scenarios. To bridge this gap, our study investigates the characteristics of LLM-generated code and its corresponding projects hosted on GitHub. Our findings reveal several key insights: (1) ChatGPT and Copilot are the most frequently utilized for generating code on GitHub. In contrast, there is very little code generated by other LLMs on GitHub. (2) Projects containing ChatGPT/Copilot-generated code are often small and less known, led by individuals or small teams. Despite this, most projects are continuously evolving and improving. (3) ChatGPT/Copilot is mainly utilized for generating Python, Java, and TypeScript scripts for data processing and transformation. C/C++ and JavaScript code generation focuses on algorithm and data structure implementation and user interface code. Most ChatGPT/Copilot-generated code snippets are relatively short and exhibit low complexity. (4) Compared to human-written code, ChatGPT/Copilot-generated code exists in a small proportion of projects and generally undergoes fewer modifications. Additionally, modifications due to bugs are even fewer, ranging from just 3% to 8% across different languages. (5) Most comments on ChatGPT/Copilot-generated code lack detailed information, often only stating the code's origin without mentioning prompts, human modifications, or testing status. Based on these findings, we discuss the implications for researchers and practitioners.

With the widespread application of Large Language Models (LLMs), it has become a significant concern to ensure their safety and prevent harmful responses. While current safe-alignment methods based on instruction fine-tuning and Reinforcement Learning from Human Feedback (RLHF) can effectively reduce harmful responses from LLMs, they often require high-quality datasets and heavy computational overhead during model training. Another way to align language models is to modify the logit of tokens in model outputs without heavy training. Recent studies have shown that contrastive decoding can enhance the performance of language models by reducing the likelihood of confused tokens. However, these methods require the manual selection of contrastive models or instruction templates. To this end, we propose Adversarial Contrastive Decoding (ACD), an optimization-based framework to generate two opposite system prompts for prompt-based contrastive decoding. ACD only needs to apply a lightweight prompt tuning on a rather small anchor dataset (< 3 min for each model) without training the target model. Experiments conducted on extensive models and benchmarks demonstrate that the proposed method achieves much better safety performance than previous model training-free decoding methods without sacrificing its original generation ability.

Software vulnerabilities pose significant risks to the security and integrity of software systems. Prior studies have proposed various approaches to vulnerability detection using deep learning or pre-trained models. However, there is still a lack of detailed explanations for understanding vulnerabilities beyond merely detecting their occurrence, which fails to truly help software developers understand and remediate the issues. Recently, large language models (LLMs) have demonstrated remarkable capabilities in comprehending complex contexts and generating content, presenting new opportunities for both detecting and explaining software vulnerabilities. In this paper, we conduct a comprehensive study to investigate the capabilities of LLMs in both detecting and explaining vulnerabilities, and we propose LLMVulExp, a framework that utilizes LLMs for these tasks. Under specialized fine-tuning for vulnerability explanation, our LLMVulExp not only detects the types of vulnerabilities in the code but also analyzes the code context to generate the cause, location, and repair suggestions for these vulnerabilities. These detailed explanations are crucial for helping developers quickly analyze and locate vulnerability issues, providing essential guidance and reference for effective remediation. We find that LLMVulExp can effectively enable the LLMs to perform vulnerability detection (e.g., achieving over a 90\% F1 score on the SeVC dataset) and provide detailed explanations. We also explore the potential of using advanced strategies such as Chain-of-Thought (CoT) to guide the LLMs in concentrating on vulnerability-prone code, achieving promising results.

Large language models (LLMs) have demonstrated remarkable capabilities in code generation tasks. However, repository-level code generation presents unique challenges, particularly due to the need to utilize information spread across multiple files within a repository. Existing retrieval-based approaches sometimes fall short as they are limited in obtaining a broader and deeper repository context. In this paper, we present CatCoder, a novel code generation framework designed for statically typed programming languages. CatCoder enhances repository-level code generation by integrating relevant code and type context. Specifically, it leverages static analyzers to extract type dependencies and merges this information with retrieved code to create comprehensive prompts for LLMs. To evaluate the effectiveness of CatCoder, we adapt and construct benchmarks that include 199 Java tasks and 90 Rust tasks. The results show that CatCoder outperforms the RepoCoder baseline by up to 17.35%, in terms of pass@k score. Furthermore, the generalizability of CatCoder is assessed using various LLMs, including both code-specialized models and general-purpose models. Our findings indicate consistent performance improvements across all models, which underlines the practicality of CatCoder.

Overfitting in RL has become one of the main obstacles to applications in reinforcement learning(RL). Existing methods do not provide explicit semantic constrain for the feature extractor, hindering the agent from learning a unified cross-domain representation and resulting in performance degradation on unseen domains. Besides, abundant data from multiple domains are needed. To address these issues, in this work, we propose prompt-based visual alignment (PVA), a robust framework to mitigate the detrimental domain bias in the image for zero-shot policy transfer. Inspired that Visual-Language Model (VLM) can serve as a bridge to connect both text space and image space, we leverage the semantic information contained in a text sequence as an explicit constraint to train a visual aligner. Thus, the visual aligner can map images from multiple domains to a unified domain and achieve good generalization performance. To better depict semantic information, prompt tuning is applied to learn a sequence of learnable tokens. With explicit constraints of semantic information, PVA can learn unified cross-domain representation under limited access to cross-domain data and achieves great zero-shot generalization ability in unseen domains. We verify PVA on a vision-based autonomous driving task with CARLA simulator. Experiments show that the agent generalizes well on unseen domains under limited access to multi-domain data., Comment: This paper has been accepted by ICML2024

In this study, we successfully developed two-dimensional paper-based analytical devices using a hybrid technique of injection molding and embossing. This innovative approach involves passive or active delivery of molten wax onto a glass substrate through a sealed chip, facilitating wax stamp creation.

Post-training quantization (PTQ) serves as a potent technique to accelerate the inference of large language models (LLMs). Nonetheless, existing works still necessitate a considerable number of floating-point (FP) operations during inference, including additional quantization and de-quantization, as well as non-linear operators such as RMSNorm and Softmax. This limitation hinders the deployment of LLMs on the edge and cloud devices. In this paper, we identify the primary obstacle to integer-only quantization for LLMs lies in the large fluctuation of activations across channels and tokens in both linear and non-linear operations. To address this issue, we propose I-LLM, a novel integer-only fully-quantized PTQ framework tailored for LLMs. Specifically, (1) we develop Fully-Smooth Block-Reconstruction (FSBR) to aggressively smooth inter-channel variations of all activations and weights. (2) to alleviate degradation caused by inter-token variations, we introduce a novel approach called Dynamic Integer-only MatMul (DI-MatMul). This method enables dynamic quantization in full-integer matrix multiplication by dynamically quantizing the input and outputs with integer-only operations. (3) we design DI-ClippedSoftmax, DI-Exp, and DI-Normalization, which utilize bit shift to execute non-linear operators efficiently while maintaining accuracy. The experiment shows that our I-LLM achieves comparable accuracy to the FP baseline and outperforms non-integer quantization methods. For example, I-LLM can operate at W4A4 with negligible loss of accuracy. To our knowledge, we are the first to bridge the gap between integer-only quantization and LLMs. We've published our code on anonymous.4open.science, aiming to contribute to the advancement of this field.

Real-time and high-performance 3D object detection plays a critical role in autonomous driving and robotics. Recent pillar-based 3D object detectors have gained significant attention due to their compact representation and low computational overhead, making them suitable for onboard deployment and quantization. However, existing pillar-based detectors still suffer from information loss along height dimension and large numerical distribution difference during pillar feature encoding (PFE), which severely limits their performance and quantization potential. To address above issue, we first unveil the importance of different input information during PFE and identify the height dimension as a key factor in enhancing 3D detection performance. Motivated by this observation, we propose a height-aware pillar feature encoder named PillarHist. Specifically, PillarHist statistics the discrete distribution of points at different heights within one pillar. This simple yet effective design greatly preserves the information along the height dimension while significantly reducing the computation overhead of the PFE. Meanwhile, PillarHist also constrains the arithmetic distribution of PFE input to a stable range, making it quantization-friendly. Notably, PillarHist operates exclusively within the PFE stage to enhance performance, enabling seamless integration into existing pillar-based methods without introducing complex operations. Extensive experiments show the effectiveness of PillarHist in terms of both efficiency and performance., Comment: 17 pages, 3 figures

Building open agents has always been the ultimate goal in AI research, and creative agents are the more enticing. Existing LLM agents excel at long-horizon tasks with well-defined goals (e.g., `mine diamonds' in Minecraft). However, they encounter difficulties on creative tasks with open goals and abstract criteria due to the inability to bridge the gap between them, thus lacking feedback for self-improvement in solving the task. In this work, we introduce autonomous embodied verification techniques for agents to fill the gap, laying the groundwork for creative tasks. Specifically, we propose the Luban agent target creative building tasks in Minecraft, which equips with two-level autonomous embodied verification inspired by human design practices: (1) visual verification of 3D structural speculates, which comes from agent synthesized CAD modeling programs; (2) pragmatic verification of the creation by generating and verifying environment-relevant functionality programs based on the abstract criteria. Extensive multi-dimensional human studies and Elo ratings show that the Luban completes diverse creative building tasks in our proposed benchmark and outperforms other baselines ($33\%$ to $100\%$) in both visualization and pragmatism. Additional demos on the real-world robotic arm show the creation potential of the Luban in the physical world.

With the increasing utilization of large language models such as ChatGPT during software development, it has become crucial to verify the quality of code content it generates. Recent studies proposed utilizing ChatGPT as both a developer and tester for multi-agent collaborative software development. The multi-agent collaboration empowers ChatGPT to produce test reports for its generated code, enabling it to self-verify the code content and fix bugs based on these reports. However, these studies did not assess the effectiveness of the generated test reports in validating the code. Therefore, we conduct a comprehensive empirical investigation to evaluate ChatGPT's self-verification capability in code generation, code completion, and program repair. We request ChatGPT to (1) generate correct code and then self-verify its correctness; (2) complete code without vulnerabilities and then self-verify for the presence of vulnerabilities; and (3) repair buggy code and then self-verify whether the bugs are resolved. Our findings on two code generation datasets, one code completion dataset, and two program repair datasets reveal the following observations: (1) ChatGPT often erroneously predicts its generated incorrect code as correct. (2) The self-contradictory hallucinations in ChatGPT's behavior arise. (3) The self-verification capability of ChatGPT can be enhanced by asking the guiding question, which queries whether ChatGPT agrees with assertions about incorrectly generated or repaired code and vulnerabilities in completed code. (4) Using test reports generated by ChatGPT can identify more vulnerabilities in completed code, but the explanations for incorrectly generated code and failed repairs are mostly inaccurate in the test reports. Based on these findings, we provide implications for further research or development using ChatGPT.

TODO comments are widely used by developers to remind themselves or others about incomplete tasks. In other words, TODO comments are usually associated with temporary or suboptimal solutions. In practice, all the equivalent suboptimal implementations should be updated (e.g., adding TODOs) simultaneously. However, due to various reasons (e.g., time constraints or carelessness), developers may forget or even are unaware of adding TODO comments to all necessary places, which results in the TODO-missed methods. These "hidden" suboptimal implementations in TODO-missed methods may hurt the software quality and maintainability in the long-term. Therefore, in this paper, we propose the novel task of TODO-missed methods detection and patching, and develop a novel model, namely TDPatcher (TODO-comment Patcher), to automatically patch TODO comments to the TODO-missed methods in software projects. Our model has two main stages: offline learning and online inference. During the offline learning stage, TDPatcher employs GraphCodeBERT and contrastive learning for encoding the TODO comment (natural language) and its suboptimal implementation (code fragment) into vector representations. For the online inference stage, we can identify the TODO-missed methods and further determine their patching position by leveraging the offline trained model. We built our dataset by collecting TODO-introduced methods from the top-10,000 Python GitHub repositories and evaluated TDPatcher on them. Extensive experimental results show the promising performance of our model over a set of benchmarks. We further conduct an in-the-wild evaluation which successfully detects 26 \textit{\major{TODO-missed} methods} from 50 GitHub repositories.

With the rapid development of deep learning, the implementation of intricate algorithms and substantial data processing have become standard elements of deep learning projects. As a result, the code has become progressively complex as the software evolves, which is difficult to maintain and understand. Existing studies have investigated the impact of refactoring on software quality within traditional software. However, the insight of code refactoring in the context of deep learning is still unclear. This study endeavors to fill this knowledge gap by empirically examining the current state of code refactoring in deep learning realm, and practitioners' views on refactoring. We first manually analyzed the commit history of five popular and well-maintained deep learning projects (e.g., PyTorch). We mined 4,921 refactoring practices in historical commits and measured how different types and elements of refactoring operations are distributed and found that refactoring operation types' distribution in deep learning projects is different from it in traditional Java software. We then surveyed 159 practitioners about their views of code refactoring in deep learning projects and their expectations of current refactoring tools. The result of the survey showed that refactoring research and the development of related tools in the field of deep learning are crucial for improving project maintainability and code quality, and that current refactoring tools do not adequately meet the needs of practitioners. Lastly, we provided our perspective on the future advancement of refactoring tools and offered suggestions for developers' development practices., Comment: 24 pages, 18 figures

Inspired by the great potential of Large Language Models (LLMs) for solving complex coding tasks, in this paper, we propose a novel approach, named Code2API, to automatically perform APIzation for Stack Overflow code snippets. Code2API does not require additional model training or any manual crafting rules and can be easily deployed on personal computers without relying on other external tools. Specifically, Code2API guides the LLMs through well-designed prompts to generate well-formed APIs for given code snippets. To elicit knowledge and logical reasoning from LLMs, we used chain-of-thought (CoT) reasoning and few-shot in-context learning, which can help the LLMs fully understand the APIzation task and solve it step by step in a manner similar to a developer. Our evaluations show that Code2API achieves a remarkable accuracy in identifying method parameters (65%) and return statements (66%) equivalent to human-generated ones, surpassing the current state-of-the-art approach, APIzator, by 15.0% and 16.5% respectively. Moreover, compared with APIzator, our user study demonstrates that Code2API exhibits superior performance in generating meaningful method names, even surpassing the human-level performance, and developers are more willing to use APIs generated by our approach, highlighting the applicability of our tool in practice. Finally, we successfully extend our framework to the Python dataset, achieving a comparable performance with Java, which verifies the generalizability of our tool.

The test failure causes analysis is critical since it determines the subsequent way of handling different types of bugs, which is the prerequisite to get the bugs properly analyzed and fixed. After a test case fails, software testers have to inspect the test execution logs line by line to identify its root cause. However, manual root cause determination is often tedious and time-consuming, which can cost 30-40% of the time needed to fix a problem. Therefore, there is a need for automatically predicting the test failure causes to lighten the burden of software testers. In this paper, we present a simple but hard-to-beat approach, named NCChecker to automatically identify the failure causes for failed test logs. Our approach can help developers efficiently identify the test failure causes, and flag the most probable log lines of indicating the root causes for investigation. Our approach has three main stages: log abstraction, lookup table construction, and failure causes prediction. We first perform log abstraction to parse the unstructured log messages into structured log events. NCChecker then automatically maintains and updates a lookup table via employing our heuristic rules, which record the matching score between different log events and test failure causes. When it comes to the failure cause prediction stage, for a newly generated failed test log, NCChecker can easily infer its failed reason by checking out the associated log events' scores from the lookup table. We have developed a prototype and evaluated our tool on a real-world industrial dataset with more than 10K test logs. The extensive experiments show the promising performance of our model over a set of benchmarks. Moreover, our approach is highly efficient and memory-saving, and can successfully handle the data imbalance problem.

Forking is a typical way of code reuse, which provides a simple way for developers to create a variant software (denoted as hard fork) by copying and modifying an existing codebase. Despite of the benefits, forking also leads to duplicate efforts in software maintenance. Developers need to port patches across the hard forks to address similar bugs or implement similar features. Due to the divergence between the source project and the hard fork, patch porting is complicated, which requires an adaption regarding different implementations of the same functionality. In this work, we take the first step to automate patch porting for hard forks under a zero-shot setting. We first conduct an empirical study of the patches ported from Vim to Neovim over the last ten years to investigate the necessities of patch porting and the potential flaws in the current practice. We then propose a large language model (LLM) based approach (namely PPatHF) to automatically port patches for hard forks on a function-wise basis. Specifically, PPatHF is composed of a reduction module and a porting module. Given the pre- and post-patch versions of a function from the reference project and the corresponding function from the target project, the reduction module first slims the input functions by removing code snippets less relevant to the patch. Then, the porting module leverages a LLM to apply the patch to the function from the target project. We evaluate PPatHF on 310 Neovim patches ported from Vim. The experimental results show that PPatHF outperforms the baselines significantly. Specifically, PPatHF can correctly port 131 (42.3%) patches and automate 57% of the manual edits required for the developer to port the patch., Comment: Accepted by ISSTA 2024

Large language models for code (i.e., code LLMs) have shown strong code understanding and generation capabilities. To evaluate the capabilities of code LLMs in various aspects, many benchmarks have been proposed (e.g., HumanEval and ClassEval). Code reasoning is one of the most essential abilities of code LLMs, but existing benchmarks for code reasoning are not sufficient. Typically, they focus on predicting the input and output of a program, ignoring the evaluation of the intermediate behavior during program execution, as well as the logical consistency (e.g., the model should not give the correct output if the prediction of execution path is wrong) when performing the reasoning. To address these problems, in this paper, we propose a framework, namely REval, for evaluating code reasoning abilities and consistency of code LLMs with program execution. We utilize existing code benchmarks and adapt them to new benchmarks within our framework. A large-scale empirical study is conducted and most LLMs show unsatisfactory performance on both Runtime Behavior Reasoning (i.e., an average accuracy of 44.4%) and Incremental Consistency Evaluation (i.e., an average IC score of 10.3). Evaluation results of current code LLMs reflect the urgent need for the community to strengthen the code reasoning capability of code LLMs. Our code, data, and \newname leaderboard are available at https://r-eval.github.io., Comment: Accepted by ICSE 2025 and this is our preprint version. Our REval leaderboard is available at https://r-eval.github.io

In the field of natural language processing, the rapid development of large language model (LLM) has attracted more and more attention. LLMs have shown a high level of creativity in various tasks, but the methods for assessing such creativity are inadequate. The assessment of LLM creativity needs to consider differences from humans, requiring multi-dimensional measurement while balancing accuracy and efficiency. This paper aims to establish an efficient framework for assessing the level of creativity in LLMs. By adapting the modified Torrance Tests of Creative Thinking, the research evaluates the creative performance of various LLMs across 7 tasks, emphasizing 4 criteria including Fluency, Flexibility, Originality, and Elaboration. In this context, we develop a comprehensive dataset of 700 questions for testing and an LLM-based evaluation method. In addition, this study presents a novel analysis of LLMs' responses to diverse prompts and role-play situations. We found that the creativity of LLMs primarily falls short in originality, while excelling in elaboration. Besides, the use of prompts and the role-play settings of the model significantly influence creativity. Additionally, the experimental results also indicate that collaboration among multiple LLMs can enhance originality. Notably, our findings reveal a consensus between human evaluations and LLMs regarding the personality traits that influence creativity. The findings underscore the significant impact of LLM design on creativity and bridges artificial intelligence and human creativity, offering insights into LLMs' creativity and potential applications.

Pre-trained code models have recently achieved substantial improvements in many code intelligence tasks. These models are first pre-trained on large-scale unlabeled datasets in a task-agnostic manner using self-supervised learning, and then fine-tuned on labeled datasets in downstream tasks. However, the labeled datasets are usually limited in size (i.e., human intensive efforts), which may hinder the performance of pre-trained code models in specific tasks. To mitigate this, one possible solution is to leverage the large-scale unlabeled data in the tuning stage by pseudo-labeling. However, directly employing the pseudo-labeled data can bring a large amount of noise, i.e., incorrect labels, leading to suboptimal performance. How to effectively leverage the noisy pseudo-labeled data is a challenging yet under-explored problem.In this paper, we propose a novel approach named HINT to improve pre-trained code models with large-scale unlabeled datasets by better utilizing the pseudo-labeled data. HINT includes two main modules: HybrId pseudo-labeled data selection and Noise-tolerant Training. In the hybrid pseudo-data selection module, considering the robustness issue, apart from directly measuring the quality of pseudo labels through training loss, we further propose to employ a retrieval-based method to filter low-quality pseudo-labeled data. The noise-tolerant training module aims to further mitigate the influence of errors in pseudo labels by training the model with a noise-tolerant loss function and by regularizing the consistency of model predictions.The experimental results show that HINT can better leverage those unlabeled data in a task-specific way and provide complementary benefits for pre-trained models, e.g., improving the best baseline model by 15.33%, 16.50%, and 8.98% on code summarization, defect detection, and assertion generation, respectively., Comment: Accepted by ICSE 2024

The number of vulnerabilities reported in open source software has increased substantially in recent years. Security patches provide the necessary measures to protect software from attacks and vulnerabilities. In practice, it is difficult to identify whether patches have been integrated into software, especially if we only have binary files. Therefore, the ability to test whether a patch is applied to the target binary, a.k.a. patch presence test, is crucial for practitioners. However, it is challenging to obtain accurate semantic information from patches, which could lead to incorrect results. In this paper, we propose a new patch presence test framework named PPT4J ($\textbf{P}$atch $\textbf{P}$resence $\textbf{T}$est $\textbf{for}$ $\textbf{J}$ava Binaries). PPT4J is designed for open-source Java libraries. It takes Java binaries (i.e. bytecode files) as input, extracts semantic information from patches, and uses feature-based techniques to identify patch lines in the binaries. To evaluate the effectiveness of our proposed approach PPT4J, we construct a dataset with binaries that include 110 vulnerabilities. The results show that PPT4J achieves an F1 score of 98.5% with reasonable efficiency, improving the baseline by 14.2%. Furthermore, we conduct an in-the-wild evaluation of PPT4J on JetBrains IntelliJ IDEA. The results suggest that a third-party library included in the software is not patched for two CVEs, and we have reported this potential security problem to the vendor., Comment: 12 pages

Although neural networks have made remarkable advancements in various applications, they require substantial computational and memory resources. Network quantization is a powerful technique to compress neural networks, allowing for more efficient and scalable AI deployments. Recently, Re-parameterization has emerged as a promising technique to enhance model performance while simultaneously alleviating the computational burden in various computer vision tasks. However, the accuracy drops significantly when applying quantization on the re-parameterized networks. We identify that the primary challenge arises from the large variation in weight distribution across the original branches. To address this issue, we propose a coarse & fine weight splitting (CFWS) method to reduce quantization error of weight, and develop an improved KL metric to determine optimal quantization scales for activation. To the best of our knowledge, our approach is the first work that enables post-training quantization applicable on re-parameterized networks. For example, the quantized RepVGG-A1 model exhibits a mere 0.3% accuracy loss. The code is in https://github.com/NeonHo/Coarse-Fine-Weight-Split.git, Comment: 23 pages

In software development, developers extensively utilize third-party libraries to avoid implementing existing functionalities. When a new third-party library vulnerability is disclosed, project maintainers need to determine whether their projects are affected by the vulnerability, which requires developers to invest substantial effort in assessment. However, existing tools face a series of issues: static analysis tools produce false alarms, dynamic analysis tools require existing tests and test generation tools have low success rates when facing complex vulnerabilities. Vulnerability exploits, as code snippets provided for reproducing vulnerabilities after disclosure, contain a wealth of vulnerability-related information. This study proposes a new method based on vulnerability exploits, called VESTA (Vulnerability Exploit-based Software Testing Auto-Generator), which provides vulnerability exploit tests as the basis for developers to decide whether to update dependencies. VESTA extends the search-based test generation methods by adding a migration step, ensuring the similarity between the generated test and the vulnerability exploit, which increases the likelihood of detecting potential library vulnerabilities in a project. We perform experiments on 30 vulnerabilities disclosed in the past five years, involving 60 vulnerability-project pairs, and compare the experimental results with the baseline method, TRANSFER. The success rate of VESTA is 71.7\% which is a 53.4\% improvement over TRANSFER in the effectiveness of verifying exploitable vulnerabilities.

During software development, vulnerabilities have posed a significant threat to users. Patches are the most effective way to combat vulnerabilities. In a large-scale software system, testing the presence of a security patch in every affected binary is crucial to ensure system security. Identifying whether a binary has been patched for a known vulnerability is challenging, as there may only be small differences between patched and vulnerable versions. Existing approaches mainly focus on detecting patches that are compiled in the same compiler options. However, it is common for developers to compile programs with very different compiler options in different situations, which causes inaccuracy for existing methods. In this paper, we propose a new approach named PS3, referring to precise patch presence test based on semantic-level symbolic signature. PS3 exploits symbolic emulation to extract signatures that are stable under different compiler options. Then PS3 can precisely test the presence of the patch by comparing the signatures between the reference and the target at semantic level. To evaluate the effectiveness of our approach, we constructed a dataset consisting of 3,631 (CVE, binary) pairs of 62 recent CVEs in four C/C++ projects. The experimental results show that PS3 achieves scores of 0.82, 0.97, and 0.89 in terms of precision, recall, and F1 score, respectively. PS3 outperforms the state-of-the-art baselines by improving 33% in terms of F1 score and remains stable in different compiler options., Comment: Accepted to ICSE 2024; Camera-ready version

Stable Diffusion has established itself as a foundation model in generative AI artistic applications, receiving widespread research and application. Some recent fine-tuning methods have made it feasible for individuals to implant personalized concepts onto the basic Stable Diffusion model with minimal computational costs on small datasets. However, these innovations have also given rise to issues like facial privacy forgery and artistic copyright infringement. In recent studies, researchers have explored the addition of imperceptible adversarial perturbations to images to prevent potential unauthorized exploitation and infringements when personal data is used for fine-tuning Stable Diffusion. Although these studies have demonstrated the ability to protect images, it is essential to consider that these methods may not be entirely applicable in real-world scenarios. In this paper, we systematically evaluate the use of perturbations to protect images within a practical threat model. The results suggest that these approaches may not be sufficient to safeguard image privacy and copyright effectively. Furthermore, we introduce a purification method capable of removing protected perturbations while preserving the original image structure to the greatest extent possible. Experiments reveal that Stable Diffusion can effectively learn from purified images over all protective methods.

Offline meta-reinforcement learning (OMRL) utilizes pre-collected offline datasets to enhance the agent's generalization ability on unseen tasks. However, the context shift problem arises due to the distribution discrepancy between the contexts used for training (from the behavior policy) and testing (from the exploration policy). The context shift problem leads to incorrect task inference and further deteriorates the generalization ability of the meta-policy. Existing OMRL methods either overlook this problem or attempt to mitigate it with additional information. In this paper, we propose a novel approach called Context Shift Reduction for OMRL (CSRO) to address the context shift problem with only offline datasets. The key insight of CSRO is to minimize the influence of policy in context during both the meta-training and meta-test phases. During meta-training, we design a max-min mutual information representation learning mechanism to diminish the impact of the behavior policy on task representation. In the meta-test phase, we introduce the non-prior context collection strategy to reduce the effect of the exploration policy. Experimental results demonstrate that CSRO significantly reduces the context shift and improves the generalization ability, surpassing previous methods across various challenging domains.

In the field of multi-task reinforcement learning, the modular principle, which involves specializing functionalities into different modules and combining them appropriately, has been widely adopted as a promising approach to prevent the negative transfer problem that performance degradation due to conflicts between tasks. However, most of the existing multi-task RL methods only combine shared modules at the task level, ignoring that there may be conflicts within the task. In addition, these methods do not take into account that without constraints, some modules may learn similar functions, resulting in restricting the model's expressiveness and generalization capability of modular methods. In this paper, we propose the Contrastive Modules with Temporal Attention(CMTA) method to address these limitations. CMTA constrains the modules to be different from each other by contrastive learning and combining shared modules at a finer granularity than the task level with temporal attention, alleviating the negative transfer within the task and improving the generalization ability and the performance for multi-task RL. We conducted the experiment on Meta-World, a multi-task RL benchmark containing various robotics manipulation tasks. Experimental results show that CMTA outperforms learning each task individually for the first time and achieves substantial performance improvements over the baselines., Comment: This paper has been accepted at NeurIPS 2023 as a poster

Deep reinforcement learning (DRL) has led to a wide range of advances in sequential decision-making tasks. However, the complexity of neural network policies makes it difficult to understand and deploy with limited computational resources. Currently, employing compact symbolic expressions as symbolic policies is a promising strategy to obtain simple and interpretable policies. Previous symbolic policy methods usually involve complex training processes and pre-trained neural network policies, which are inefficient and limit the application of symbolic policies. In this paper, we propose an efficient gradient-based learning method named Efficient Symbolic Policy Learning (ESPL) that learns the symbolic policy from scratch in an end-to-end way. We introduce a symbolic network as the search space and employ a path selector to find the compact symbolic policy. By doing so we represent the policy with a differentiable symbolic expression and train it in an off-policy manner which further improves the efficiency. In addition, in contrast with previous symbolic policies which only work in single-task RL because of complexity, we expand ESPL on meta-RL to generate symbolic policies for unseen tasks. Experimentally, we show that our approach generates symbolic policies with higher performance and greatly improves data efficiency for single-task RL. In meta-RL, we demonstrate that compared with neural network policies the proposed symbolic policy achieves higher performance and efficiency and shows the potential to be interpretable., Comment: Accepted by NeurIPS2023

Static analysis tools have gained popularity among developers for finding potential bugs, but their widespread adoption is hindered by the accomnpanying high false alarm rates (up to 90%). To address this challenge, previous studies proposed the concept of actionable warnings, and apply machine-learning methods to distinguish actionable warnings from false alarms. Despite these efforts, our preliminary study suggests that the current methods used to collect actionable warnings are rather shaky and unreliable, resulting in a large proportion of invalid actionable warnings. In this work, we mined 68,274 reversions from Top-500 Github C repositories to create a substantia actionable warning dataset and assigned weak labels to each warning's likelihood of being a real bug. To automatically identify actionable warnings and recommend those with a high probability of being real bugs (AWHB), we propose a two-stage framework called ACWRecommender. In the first stage, our tool use a pre-trained model, i.e., UniXcoder, to identify actionable warnings from a huge number of SA tool's reported warnings. In the second stage, we rerank valid actionable warnings to the top by using weakly supervised learning. Experimental results showed that our tool outperformed several baselines for actionable warning detection (in terms of F1-score) and performed better for AWHB recommendation (in terms of nDCG and MRR). Additionaly, we also performed an in-the-wild evaluation, we manually validated 24 warnings out of 2,197 reported warnings on 10 randomly selected projects, 22 of which were confirmed by developers as real bugs, demonstrating the practical usage of our tool., Comment: accepted by ASE2023 Industry Track

Large language models (LLMs) show their powerful automatic reasoning and planning capability with a wealth of semantic knowledge about the human world. However, the grounding problem still hinders the applications of LLMs in the real-world environment. Existing studies try to fine-tune the LLM or utilize pre-defined behavior APIs to bridge the LLMs and the environment, which not only costs huge human efforts to customize for every single task but also weakens the generality strengths of LLMs. To autonomously ground the LLM onto the environment, we proposed the Self-Driven Grounding (SDG) framework to automatically and progressively ground the LLM with self-driven skill learning. SDG first employs the LLM to propose the hypothesis of sub-goals to achieve tasks and then verify the feasibility of the hypothesis via interacting with the underlying environment. Once verified, SDG can then learn generalized skills with the guidance of these successfully grounded subgoals. These skills can be further utilized to accomplish more complex tasks which fail to pass the verification phase. Verified in the famous instruction following task set-BabyAI, SDG achieves comparable performance in the most challenging tasks compared with imitation learning methods that cost millions of demonstrations, proving the effectiveness of learned skills and showing the feasibility and efficiency of our framework.

Existing studies show that code summaries help developers understand and maintain source code. Unfortunately, these summaries are often missing or outdated in software projects. Code summarization aims to generate natural language descriptions automatically for source code. Code summaries are highly structured and have repetitive patterns. Besides the patternized words, a code summary also contains important keywords, which are the key to reflecting the functionality of the code. However, the state-of-the-art approaches perform poorly on predicting the keywords, which leads to the generated summaries suffering a loss in informativeness. To alleviate this problem, this paper proposes a novel retrieve-and-edit approach named EditSum for code summarization. Specifically, EditSum first retrieves a similar code snippet from a pre-defined corpus and treats its summary as a prototype summary to learn the pattern. Then, EditSum edits the prototype automatically to combine the pattern in the prototype with the semantic information of input code. Our motivation is that the retrieved prototype provides a good start-point for post-generation because the summaries of similar code snippets often have the same pattern. The post-editing process further reuses the patternized words in the prototype and generates keywords based on the semantic information of input code. We conduct experiments on a large-scale Java corpus and experimental results demonstrate that EditSum outperforms the state-of-the-art approaches by a substantial margin. The human evaluation also proves the summaries generated by EditSum are more informative and useful. We also verify that EditSum performs well on predicting the patternized words and keywords., Comment: Accepted by the 36th IEEE/ACM International Conference on Automated Software Engineering (ASE 2021)

Developers write logging statements to generate logs that provide run-time information for various tasks. The readability of log messages in the logging statements (i.e., the descriptive text) is rather crucial to the value of the generated logs. Immature log messages may slow down or even obstruct the process of log analysis. Despite the importance of log messages, there is still a lack of standards on what constitutes good readability in log messages and how to write them. In this paper, we conduct a series of interviews with 17 industrial practitioners to investigate their expectations on the readability of log messages. Through the interviews, we derive three aspects related to the readability of log messages, including Structure, Information, and Wording, along with several specific practices to improve each aspect. We validate our findings through a series of online questionnaire surveys and receive positive feedback from the participants. We then manually investigate the readability of log messages in large-scale open source systems and find that a large portion (38.1%) of the log messages have inadequate readability. Motivated by such observation, we further explore the potential of automatically classifying the readability of log messages using deep learning and machine learning models. We find that both deep learning and machine learning models can effectively classify the readability of log messages with a balanced accuracy above 80.0% on average. Our study provides comprehensive guidelines for composing log messages to further improve practitioners' logging practices., Comment: Accepted as a research paper at the 38th IEEE/ACM International Conference on Automated Software Engineering (ASE 2023)

Design activity -- constructing an artifact description satisfying given goals and constraints -- distinguishes humanity from other animals and traditional machines, and endowing machines with design abilities at the human level or beyond has been a long-term pursuit. Though machines have already demonstrated their abilities in designing new materials, proteins, and computer programs with advanced artificial intelligence (AI) techniques, the search space for designing such objects is relatively small, and thus, "Can machines design like humans?" remains an open question. To explore the boundary of machine design, here we present a new AI approach to automatically design a central processing unit (CPU), the brain of a computer, and one of the world's most intricate devices humanity have ever designed. This approach generates the circuit logic, which is represented by a graph structure called Binary Speculation Diagram (BSD), of the CPU design from only external input-output observations instead of formal program code. During the generation of BSD, Monte Carlo-based expansion and the distance of Boolean functions are used to guarantee accuracy and efficiency, respectively. By efficiently exploring a search space of unprecedented size 10^{10^{540}}, which is the largest one of all machine-designed objects to our best knowledge, and thus pushing the limits of machine design, our approach generates an industrial-scale RISC-V CPU within only 5 hours. The taped-out CPU successfully runs the Linux operating system and performs comparably against the human-designed Intel 80486SX CPU. In addition to learning the world's first CPU only from input-output observations, which may reform the semiconductor industry by significantly reducing the design cycle, our approach even autonomously discovers human knowledge of the von Neumann architecture., Comment: 28 pages

Domain adaptation in reinforcement learning (RL) mainly deals with the changes of observation when transferring the policy to a new environment. Many traditional approaches of domain adaptation in RL manage to learn a mapping function between the source and target domain in explicit or implicit ways. However, they typically require access to abundant data from the target domain. Besides, they often rely on visual clues to learn the mapping function and may fail when the source domain looks quite different from the target domain. To address these problems, we propose a novel framework Online Prototype Alignment (OPA) to learn the mapping function based on the functional similarity of elements and is able to achieve the few-shot policy transfer within only several episodes. The key insight of OPA is to introduce an exploration mechanism that can interact with the unseen elements of the target domain in an efficient and purposeful manner, and then connect them with the seen elements in the source domain according to their functionalities (instead of visual clues). Experimental results show that when the target domain looks visually different from the source domain, OPA can achieve better transfer performance even with much fewer samples from the target domain, outperforming prior methods., Comment: This paper has been accepted at ICML2023

Unlearning techniques are proposed to prevent third parties from exploiting unauthorized data, which generate unlearnable samples by adding imperceptible perturbations to data for public publishing. These unlearnable samples effectively misguide model training to learn perturbation features but ignore image semantic features. We make the in-depth analysis and observe that models can learn both image features and perturbation features of unlearnable samples at an early stage, but rapidly go to the overfitting stage since the shallow layers tend to overfit on perturbation features and make models fall into overfitting quickly. Based on the observations, we propose Progressive Staged Training to effectively prevent models from overfitting in learning perturbation features. We evaluated our method on multiple model architectures over diverse datasets, e.g., CIFAR-10, CIFAR-100, and ImageNet-mini. Our method circumvents the unlearnability of all state-of-the-art methods in the literature and provides a reliable baseline for further evaluation of unlearnable techniques.

Diffusion models have demonstrated remarkable performance in image generation tasks, paving the way for powerful AIGC applications. However, these widely-used generative models can also raise security and privacy concerns, such as copyright infringement, and sensitive data leakage. To tackle these issues, we propose a method, Unlearnable Diffusion Perturbation, to safeguard images from unauthorized exploitation. Our approach involves designing an algorithm to generate sample-wise perturbation noise for each image to be protected. This imperceptible protective noise makes the data almost unlearnable for diffusion models, i.e., diffusion models trained or fine-tuned on the protected data cannot generate high-quality and diverse images related to the protected training data. Theoretically, we frame this as a max-min optimization problem and introduce EUDP, a noise scheduler-based method to enhance the effectiveness of the protective noise. We evaluate our methods on both Denoising Diffusion Probabilistic Model and Latent Diffusion Models, demonstrating that training diffusion models on the protected data lead to a significant reduction in the quality of the generated images. Especially, the experimental results on Stable Diffusion demonstrate that our method effectively safeguards images from being used to train Diffusion Models in various tasks, such as training specific objects and styles. This achievement holds significant importance in real-world scenarios, as it contributes to the protection of privacy and copyright against AI-generated content.

The advents of Large Language Models (LLMs) have shown promise in augmenting programming using natural interactions. However, while LLMs are proficient in compiling common usage patterns into a programming language, e.g., Python, it remains a challenge how to edit and debug an LLM-generated program. We introduce ANPL, a programming system that allows users to decompose user-specific tasks. In an ANPL program, a user can directly manipulate sketch, which specifies the data flow of the generated program. The user annotates the modules, or hole with natural language descriptions offloading the expensive task of generating functionalities to the LLM. Given an ANPL program, the ANPL compiler generates a cohesive Python program that implements the functionalities in hole, while respecting the dataflows specified in sketch. We deploy ANPL on the Abstraction and Reasoning Corpus (ARC), a set of unique tasks that are challenging for state-of-the-art AI systems, showing it outperforms baseline programming systems that (a) without the ability to decompose tasks interactively and (b) without the guarantee that the modules can be correctly composed together. We obtain a dataset consisting of 300/400 ARC tasks that were successfully decomposed and grounded in Python, providing valuable insights into how humans decompose programmatic tasks. See the dataset at https://iprc-dip.github.io/DARC.

The range of the U bosonic coupling constants in neutron star matter is a very interesting but still unsolved problem which has multifaceted influences in nuclear physics, particle physics, astrophysics and cosmology. The combination of the theoretical numerical simulation and the recent observations provides a very good opportunity to solve this problem. In the present work, the range of the U bosonic coupling constants is inferred based on the three relations of the mass-radius, mass-frequency and mass-tidal deformability in neutron star containing hyperons using the GM1, TM1 and NL3 parameter sets under the two flavor symmetries of the SU(6) and SU(3) in the framework of the relativistic mean field theory. Combined with observations from PSRs J1614-2230, J0348+0432, J2215-5135, J0952-0607, J0740+6620, J0030-0451, J1748-2446ad, XTE J1739-285, GW170817 and GW190814 events, our numerical results show that the U bosonic coupling constants may tend to be within the range from 0 to 20 GeV$^{-2}$ in neutron star containing hyperons. Moreover, the numerical results of the three relations obtained by the SU(3) symmetry are better in accordance with observation data than those obtained by the SU(6) symmetry. The results will help us to improve the strict constraints of the equation of state for neutron stars containing hyperons., Comment: 14pages,4figures

Despite the broad application of deep reinforcement learning (RL), transferring and adapting the policy to unseen but similar environments is still a significant challenge. Recently, the language-conditioned policy is proposed to facilitate policy transfer through learning the joint representation of observation and text that catches the compact and invariant information across environments. Existing studies of language-conditioned RL methods often learn the joint representation as a simple latent layer for the given instances (episode-specific observation and text), which inevitably includes noisy or irrelevant information and cause spurious correlations that are dependent on instances, thus hurting generalization performance and training efficiency. To address this issue, we propose a conceptual reinforcement learning (CRL) framework to learn the concept-like joint representation for language-conditioned policy. The key insight is that concepts are compact and invariant representations in human cognition through extracting similarities from numerous instances in real-world. In CRL, we propose a multi-level attention encoder and two mutual information constraints for learning compact and invariant concepts. Verified in two challenging environments, RTFM and Messenger, CRL significantly improves the training efficiency (up to 70%) and generalization ability (up to 30%) to the new environment dynamics., Comment: Accepted by AAAI 2023