Yuling Shi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Yuling Shi has a total of 62 co-authors. Here are the top co-authors by number of shared publications:
- Xiaodong Gu
Shanghai Jiao Tong University, Shanghai, China
9 shared publications - Chengcheng Wan
East China Normal University, Shanghai, China
4 shared publications - Beijun Shen
Shanghai Jiao Tong University, Shanghai, China
3 shared publications - Chaoxiang Xie
Hohai University, Nanjing, China
2 shared publications - Hongyu Zhang
Chongqing University
2 shared publications - Lei Ma
University of Tokyo, Tokyo, Japan
2 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 14350-14359
HyperBalance: Hypergraph-based Topology Optimization for Reliable Multi-Agent Communication
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 14350-14359
Open AccessHyperBalance: Hypergraph-based Topology Optimization for Reliable Multi-Agent Communication
- Open Access
- 7 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 411-423
LLM Agents Can See Code Repositories
- 7 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 411-423
Open AccessLLM Agents Can See Code Repositories
- Open Access
- 1 October 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Software Engineering
- Vol. 3 (ISSTA) , 1740-1764
Seeing Is Coding: On the Effectiveness of Vision Language Models in Code Understanding
Large Language Models (LLMs) have achieved remarkable success in source code understanding, yet as software systems grow in scale, computational efficiency has become a critical bottleneck. Currently, these models rely on a text-based paradigm that treats source code as a linear sequence of tokens, which leads to a linear increase in context length and associated computational costs. The rapid advancement of Multimodal LLMs (MLLMs) introduces an opportunity to optimize efficiency by representing source code as rendered images. Unlike text, which is difficult to compress without losing semantic meaning, the image modality is inherently suitable for compression. By adjusting resolution, images can be scaled to a fraction of their original token cost while remaining recognizable to vision-capable models. To explore the feasibility of this approach, we conduct the first systematic study on the effectiveness of MLLMs for code understanding. Our experiments reveal that: (1) MLLMs can effectively understand code with substantial token reduction, achieving up to 8× compression; (2) MLLMs can effectively leverage visual cues such as syntax highlighting, improving code completion performance under 4× compression; and (3) Code-understanding tasks like clone detection exhibit exceptional resilience to visual compression, with some compression ratios even slightly outperforming raw text inputs. Our findings highlight both the potential and current limitations of MLLMs in code understanding, which points out a shift toward image-modality code representation as a pathway to more efficient inference.
- 1 October 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Software Engineering
- Vol. 3 (ISSTA) , 1740-1764
Open AccessSeeing Is Coding: On the Effectiveness of Vision Language Models in Code Understanding
Large Language Models (LLMs) have achieved remarkable success in source code understanding, yet as software systems grow in scale, computational efficiency has become a critical bottleneck. Currently, these models rely on a text-based paradigm that treats source code as a linear sequence of tokens, which leads to a linear increase in context length and associated computational costs. The rapid advancement of Multimodal LLMs (MLLMs) introduces an opportunity to optimize efficiency by representing source code as rendered images. Unlike text, which is difficult to compress without losing semantic meaning, the image modality is inherently suitable for compression. By adjusting resolution, images can be scaled to a fraction of their original token cost while remaining recognizable to vision-capable models. To explore the feasibility of this approach, we conduct the first systematic study on the effectiveness of MLLMs for code understanding. Our experiments reveal that: (1) MLLMs can effectively understand code with substantial token reduction, achieving up to 8× compression; (2) MLLMs can effectively leverage visual cues such as syntax highlighting, improving code completion performance under 4× compression; and (3) Code-understanding tasks like clone detection exhibit exceptional resilience to visual compression, with some compression ratios even slightly outperforming raw text inputs. Our findings highlight both the potential and current limitations of MLLMs in code understanding, which points out a shift toward image-modality code representation as a pathway to more efficient inference.
- Open Access
- 1 October 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Software Engineering
- Vol. 3 (ISSTA) , 710-733
Test vs Mutant: Adversarial LLM Agents for Robust Unit Test Generation
Software testing is a critical, yet resource-intensive phase of the software development lifecycle. Search-based approaches typically achieve high coverage but produce tests with low readability, whereas large language model (LLM)-based methods generate more human-readable tests but often suffer from low coverage and compilability. While the majority of research efforts have focused on improving test coverage and readability, comparatively less attention has been paid to enhancing the robustness of bug detection. To address this gap, we propose AdverTest, a novel adversarial framework for LLM-powered test case generation that pairs a test case generation agent T with a mutant generation agent M: M persistently creates mutants "hacking" the blind spots of T's current test suite, whileT iteratively refines its tests to "kill" the challenging mutants, with the interaction guided by both coverage and mutation scores. Experimental results on Defects4J show that our approach improves fault detection rates by 8.56% over the best existing LLM-based methods and by 50.20% over EvoSuite, while remaining competitive on line and branch coverage.
- 1 October 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Software Engineering
- Vol. 3 (ISSTA) , 710-733
Open AccessTest vs Mutant: Adversarial LLM Agents for Robust Unit Test Generation
Software testing is a critical, yet resource-intensive phase of the software development lifecycle. Search-based approaches typically achieve high coverage but produce tests with low readability, whereas large language model (LLM)-based methods generate more human-readable tests but often suffer from low coverage and compilability. While the majority of research efforts have focused on improving test coverage and readability, comparatively less attention has been paid to enhancing the robustness of bug detection. To address this gap, we propose AdverTest, a novel adversarial framework for LLM-powered test case generation that pairs a test case generation agent T with a mutant generation agent M: M persistently creates mutants "hacking" the blind spots of T's current test suite, whileT iteratively refines its tests to "kill" the challenging mutants, with the interaction guided by both coverage and mutation scores. Experimental results on Defects4J show that our approach improves fault detection rates by 8.56% over the best existing LLM-based methods and by 50.20% over EvoSuite, while remaining competitive on line and branch coverage.
- Open Access
- 1 October 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Software Engineering
- Vol. 3 (ISSTA) , 221-245
Datura: Progressive Red Teaming Testing for Tool Invocation Chain in LLM Agents
LLM agents that invoke external tools face critical safety vulnerabilities when malicious manipulations exploit their implicit trust in tool outputs and metadata. However, identifying these vulnerabilities through testing is challenging due to the need to bypass safety guardrails with semantically legitimate inputs, the difficulty in characterizing successful attacks amid implicit tool trust, and the requirement to maintain logical consistency across fragile state-dependent execution chains. In this paper, we first conduct an empirical study to investigate how external tools influence agent reasoning. Guided by the findings, we propose Datura, an automated red teaming testing framework that exposes safety vulnerabilities through chained tool manipulation. Through a five-stage workflow, Datura dynamically generates test cases where each individual step appears legitimate yet collectively leads to harmful outcomes. We evaluate Datura across five LLMs and 740 safety-critical tasks under five defense settings, including real-world safety mechanisms. Under Model Alignment, Datura achieves 94.86--99.59% attack success rate (ASR), outperforming the strongest baseline by up to 25.27 percentage points. Under Prompt Refuge, Datura maintains 78.78--95.54% ASR, showing that progressive tool-chain manipulation remains effective even under prompt-level safeguards.
- 1 October 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Software Engineering
- Vol. 3 (ISSTA) , 221-245
Open AccessDatura: Progressive Red Teaming Testing for Tool Invocation Chain in LLM Agents
LLM agents that invoke external tools face critical safety vulnerabilities when malicious manipulations exploit their implicit trust in tool outputs and metadata. However, identifying these vulnerabilities through testing is challenging due to the need to bypass safety guardrails with semantically legitimate inputs, the difficulty in characterizing successful attacks amid implicit tool trust, and the requirement to maintain logical consistency across fragile state-dependent execution chains. In this paper, we first conduct an empirical study to investigate how external tools influence agent reasoning. Guided by the findings, we propose Datura, an automated red teaming testing framework that exposes safety vulnerabilities through chained tool manipulation. Through a five-stage workflow, Datura dynamically generates test cases where each individual step appears legitimate yet collectively leads to harmful outcomes. We evaluate Datura across five LLMs and 740 safety-critical tasks under five defense settings, including real-world safety mechanisms. Under Model Alignment, Datura achieves 94.86--99.59% attack success rate (ASR), outperforming the strongest baseline by up to 25.27 percentage points. Under Prompt Refuge, Datura maintains 78.78--95.54% ASR, showing that progressive tool-chain manipulation remains effective even under prompt-level safeguards.
- Open Access
- 19 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 2498-2508
DLLM-Searcher: Adapting Diffusion Language Model for Efficient Search Agents
- 19 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 2498-2508
Open AccessDLLM-Searcher: Adapting Diffusion Language Model for Efficient Search Agents
- Open Access
- 5 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 340-348
ClassEval-Pro: A Cross-Domain Benchmark for Class-Level Code Generation
- 5 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 340-348
Open AccessClassEval-Pro: A Cross-Domain Benchmark for Class-Level Code Generation
- Open Access
- 30 June 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Software Engineering
- Vol. 3 (FSE) , 1469-1491
In Line with Context: Repository-Level Code Generation via Context Inlining
Repository-level code generation has attracted growing attention in recent years. Unlike function-level code generation, it requires the model to understand the entire repository and reason over complex dependencies across functions, classes, and modules. However, existing approaches such as retrieval-augmented generation (RAG) or context-based function selection often fall short; they primarily rely on surface-level similarity and struggle to capture the rich dependencies that govern repository-level semantics. In this paper, we introduce InlineCoder, a novel framework for repository-level code generation. InlineCoder enhances the understanding of repository context by inlining the unfinished function into its call graph, thereby reframing the challenge of repository understanding into a simpler function-level coding task. Given a function signature, InlineCoder first generates a draft completion (termed an "anchor"), which approximates downstream dependencies and enables perplexity-based confidence estimation. This anchor drives a bidirectional inlining process: (1) Upstream Inlining, which embeds the anchor into its callers to capture diverse usage scenarios; and (2) Downstream Retrieval, which integrates the anchor's callees into the prompt to provide precise dependency context. The enriched context, combining draft completion with upstream and downstream perspectives, equips the LLM with a comprehensive repository view. Extensive experiments on the DevEval and RepoExec benchmarks demonstrate that InlineCoder substantially outperforms a wide range of state-of-the-art baselines, achieving average relative gains of 29.73% in EM, 20.82% in ES, and 49.34% in BLEU on RepoExec compared to the strongest baseline. These results highlight its effectiveness in repository context understanding as well as its generalization across domains.
- 30 June 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Software Engineering
- Vol. 3 (FSE) , 1469-1491
Open AccessIn Line with Context: Repository-Level Code Generation via Context Inlining
Repository-level code generation has attracted growing attention in recent years. Unlike function-level code generation, it requires the model to understand the entire repository and reason over complex dependencies across functions, classes, and modules. However, existing approaches such as retrieval-augmented generation (RAG) or context-based function selection often fall short; they primarily rely on surface-level similarity and struggle to capture the rich dependencies that govern repository-level semantics. In this paper, we introduce InlineCoder, a novel framework for repository-level code generation. InlineCoder enhances the understanding of repository context by inlining the unfinished function into its call graph, thereby reframing the challenge of repository understanding into a simpler function-level coding task. Given a function signature, InlineCoder first generates a draft completion (termed an "anchor"), which approximates downstream dependencies and enables perplexity-based confidence estimation. This anchor drives a bidirectional inlining process: (1) Upstream Inlining, which embeds the anchor into its callers to capture diverse usage scenarios; and (2) Downstream Retrieval, which integrates the anchor's callees into the prompt to provide precise dependency context. The enriched context, combining draft completion with upstream and downstream perspectives, equips the LLM with a comprehensive repository view. Extensive experiments on the DevEval and RepoExec benchmarks demonstrate that InlineCoder substantially outperforms a wide range of state-of-the-art baselines, achieving average relative gains of 29.73% in EM, 20.82% in ES, and 49.34% in BLEU on RepoExec compared to the strongest baseline. These results highlight its effectiveness in repository context understanding as well as its generalization across domains.
- Open Access
- 28 May 2026
- Published by Association for Computing Machinery (ACM)
- p. 626-635
Reasoning in Trees: Improving Retrieval-Augmented Generation for Multi-Hop Question Answering
- 28 May 2026
- Published by Association for Computing Machinery (ACM)
- p. 626-635
Open AccessReasoning in Trees: Improving Retrieval-Augmented Generation for Multi-Hop Question Answering
- Open Access
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 3014-3027
Comfrey: Mitigating Integration Failures in LLM-enabled Software at Run-Time
Due to the unrestricted outputs of LLMs and strict requirements of software components, integration failures are widespread in software that incorporates LLM agents and retrieval-augmented generation (RAG). Even seemingly correct LLM/RAG responses can trigger software misbehaviors if they violate these requirements. In this paper, we conduct an empirical study to understand integration failures in real-world LLM-enabled applications. Guided by this study, we present Comfrey [104], a runtime framework that adapts the LLM agent and RAG responses to meet software requirements, serving as a middle layer between AI and software components. It automatically detects and resolves potential integration failures through a three-stage workflow, ensuring component compatibility. Our evaluation with a variety of open-source applications demonstrates that Comfrey detects 75.1% and prevents 63.3% of potential integration failures with 8.4% overhead, significantly outperforming the baselines.
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 3014-3027
Open AccessComfrey: Mitigating Integration Failures in LLM-enabled Software at Run-Time
Due to the unrestricted outputs of LLMs and strict requirements of software components, integration failures are widespread in software that incorporates LLM agents and retrieval-augmented generation (RAG). Even seemingly correct LLM/RAG responses can trigger software misbehaviors if they violate these requirements. In this paper, we conduct an empirical study to understand integration failures in real-world LLM-enabled applications. Guided by this study, we present Comfrey [104], a runtime framework that adapts the LLM agent and RAG responses to meet software requirements, serving as a middle layer between AI and software components. It automatically detects and resolves potential integration failures through a three-stage workflow, ensuring component compatibility. Our evaluation with a variety of open-source applications demonstrates that Comfrey detects 75.1% and prevents 63.3% of potential integration failures with 8.4% overhead, significantly outperforming the baselines.