Search for: All records

Identifying, localizing, and resolving bugs in software engineering is challenging and costly. Approaches to resolve software bugs range from Large Language Model (LLM) code analysis and repair, and automated code repair technology that aims to alleviate the technical burden of difficult to solve bugs. We propose RAGFix, which enhances LLM’s capabilities for bug localization and code repair using Retrieval Augmented Generation (RAG) based on dynamically collected Stack Overflow posts. These posts are searchable via a Question and Answer Knowledge Graph (KGQA). We evaluate our method on the HumanEvalFix benchmark for Python using relevant closed and open-source models. Our approach facilitates error resolution in Python coding problems by creating a searchable, embedded knowledge graph representation of bug and solution information from Stack Overflow, interlinking bugs, and solutions through semi-supervised graph construction methods. We use cosine similarity on embeddings based on LLM-synthesized summaries and algorithmic features describing the coding problem and potential solution to find relevant results that improve LLM in-context performance. Our results indicate that our system enhances small open-source models’ ability to effectively repair code, particularly where these models have less parametric knowledge about relevant coding problems and can leverage nonparametric knowledge to provide accurate, actionable fixes. 

Identifying, localizing, and resolving bugs in software engineering is challenging and costly. Approaches to resolve software bugs range from Large Language Model (LLM) code analysis and repair, and automated code repair technology that aims to alleviate the technical burden of difficult to solve bugs. We propose RAGFix, which enhances LLM’s capabilities for bug localization and code repair using Retrieval Augmented Generation (RAG) based on dynamically collected Stack Overflow posts. These posts are searchable via a Question and Answer Knowledge Graph (KGQA). We evaluate our method on the HumanEvalFix benchmark for Python using relevant closed and open-source models. Our approach facilitates error resolution in Python coding problems by creating a searchable, embedded knowledge graph representation of bug and solution information from Stack Overflow, interlinking bugs, and solutions through semi-supervised graph construction methods. We use cosine similarity on embeddings based on LLM-synthesized summaries and algorithmic features describing the coding problem and potential solution to find relevant results that improve LLM in-context performance. Our results indicate that our system enhances small open-source models’ ability to effectively repair code, particularly where these models have less parametric knowledge about relevant coding problems and can leverage nonparametric knowledge to provide accurate, actionable fixes. 

The graph distinguishability problem investigates whether a graph can be uniquely identified by the spectrum of its adjacency matrix, specifically determining if two graphs with the same spectrum are isomorphic. This issue is central to spectral graph theory and has significant implications for graph machine learning. In this paper, we explore the intricate connections between graph distinguishability and graph controllability–an essential concept in the control of networked systems. Focusing on oriented graphs and their skew-adjacency matrices, we establish controllability-based conditions that ensure their distinguishability. Notably, our conditions are less restrictive than existing methods, enabling a broader class of graphs to satisfy the distinguishability criteria. We illustrate the effectiveness of our results with several examples. Our findings highlight the applications of network control methods in tackling this crucial problem in algebraic graph theory, with implications for machine learning and network design. 

Deep neural networks (DNNs) are increasingly used in critical applications like autonomous vehicles and medical diagnosis, where accuracy and reliability are crucial. However, debugging DNNs is challenging and expensive, often leading to unpredictable behavior and performance issues. Identifying and diagnosing bugs in DNNs is difficult due to complex and obscure failure symptoms, which are data-driven and compute-intensive. To address this, we propose TransBug a framework that combines transformer models for feature extraction with deep learning models for classification to detect and diagnose bugs in DNNs. We employ a pre-trained transformer model, which has been trained in programming languages, to extract semantic features from both faulty and correct DNN models. We then use these extracted features in a separate deep-learning model to determine whether the code contains bugs. If a bug is detected, the model further classifies the type of bug. By leveraging the powerful feature extraction capabilities of transformers, we capture relevant characteristics from the code, which are then used by a deep learning model to identify and classify various types of bugs. This combination of transformer-based feature extraction and deep learning classification allows our method to accurately link bug symptoms to their causes, enabling developers to take targeted corrective actions. Empirical results show that the TransBug shows an accuracy of 81% for binary classification and 91% for classifying bug types. 

Individuals with specific language impairment (SLI) struggle with language acquisition despite average non-verbal intelligence and otherwise typical development. One SLI account focuses on grammar acquisition delay. The current study aimed to detect novel rare genetic variants associated with performance on a grammar assessment, the Test of Early Grammatical Impairment (TEGI), in English-speaking children. The TEGI was selected due to its sensitivity and specificity, consistently high heritability estimates, and its absence from all but one molecular genetic study. We performed whole exome sequencing (WES) in eight families with SLI (n = 74 total) and follow-up Sanger sequencing in additional unrelated probands (n = 146). We prioritized rare exonic variants shared by individuals with low TEGI performance (n = 34) from at least two families under two filtering workflows: (1) novel and (2) previously reported candidate genes. Candidate variants were observed on six new genes (PDHA2, PCDHB3, FURIN, NOL6, IQGAP3, and BAHCC1), and two genes previously reported for overall language ability (GLI3 and FLNB). We specifically suggest PCDHB3, a protocadherin gene, and NOL6 are critical for ribosome synthesis, as they are important targets of SLI investigation. The proposed SLI candidate genes associated with TEGI performance emphasize the utility of precise phenotyping and family-based genetic study.