CodeGraphGPT: An LLM-Powered Fuzz Driver Generation System Enhanced by Code Knowledge Graphs for Improved Software Testing
Integrating Large Language Models (LLMs) with code knowledge graphs significantly enhances automated fuzz driver generation, leading to improved code coverage and more effective vulnerability detection in software testing.
Solar: A Framework for Evaluating and Mitigating Social Bias in LLM-Generated Code
Large Language Models (LLMs) are prone to embedding social biases in generated code, necessitating a dedicated framework called Solar for evaluation and mitigation of these biases to ensure fairness.
AI 통합 시스템의 감사 가능성 평가: 프레임워크 및 학습 분석 사례 연구
AI 기반 시스템, 특히 학습 분석(LA) 분야에서 감사 가능성을 높이려면 검증 가능한 주장, 이를 뒷받침하는 증거, 감사자의 접근성을 포괄하는 프레임워크가 필수적입니다.
VALTEST: Using Token Probabilities to Validate Test Cases Generated by Large Language Models
VALTEST leverages token probabilities from Large Language Models (LLMs) to automatically validate the correctness of generated test cases, even when the source code is unavailable, significantly improving the reliability of LLM-based software testing.
Importance-Driven Model-Based Testing for Deep Reinforcement Learning Safety
This paper introduces a novel model-based testing framework for Deep Reinforcement Learning (DRL) policies that prioritizes testing in states where decisions have the most significant impact on safety, thereby enabling efficient and rigorous safety verification.
Improving Automated Unit Test Generation Using Large Language Models and Template-Based Repair: Introducing TestART
TestART, a novel approach combining large language models (LLMs) with template-based repair, significantly improves the quality and effectiveness of automated unit test generation for Java code.
SYNTER: An LLM-Based Approach to Automatically Repairing Obsolete Test Cases Using Static Analysis and Neural Reranking
SYNTER is a novel approach that leverages the power of Large Language Models (LLMs) in combination with static analysis and neural reranking techniques to automatically repair obsolete test cases caused by syntactic breaking changes in evolving software.
LLM4FPM: Using Precise and Complete Code Context to Improve Automatic False Positive Mitigation in Static Application Security Testing
Precise and complete code context is crucial for Large Language Models (LLMs) to effectively mitigate false positives in Static Application Security Testing (SAST) tools, and the LLM4FPM framework, incorporating eCPG-Slicer and FARF algorithm, demonstrates significant improvements in accuracy and efficiency in identifying false positives.
MdEval: A Massively Multilingual Benchmark for Evaluating Code Debugging in Large Language Models
This paper introduces MdEval, a new benchmark for evaluating the code debugging capabilities of large language models across 18 programming languages, addressing the limitations of existing benchmarks that primarily focus on Python.
Fuzzing Processing Pipelines for Zero-Knowledge Circuits: A Novel Approach to Detecting Logic Bugs
This paper introduces a novel fuzzing technique using metamorphic testing to effectively detect logic bugs in zero-knowledge circuit processing pipelines, as demonstrated by the open-source tool Circuzz.
© 2024 by Linnk AI