Search

This work investigates the application of Machine Unlearning (MU) for mitigating the impact of trojans embedded in conventional large language models of natural language (Text-LLMs) and large language models of code (Code-LLMs) We propose a novel unlearning approach, LYA, that leverages both gradient ascent and elastic weight consolidation, a Fisher Information Matrix (FIM) based regularization technique, to unlearn trojans from poisoned models. We compare the effectiveness of LYA against conventional techniques like fine-tuning, retraining, and vanilla gradient ascent. The subject models we investigate are BERT and CodeBERT, for sentiment analysis and code defect detection tasks, respectively. Our findings demonstrate that the combination of gradient ascent and FIM-based regularization, as done in LYA, outperforms existing methods in removing the trojan's influence from the poisoned model, while preserving its original functionality. To the best of our knowledge, this is the first work that compares and contrasts MU of trojans in LLMs, in the NL and Coding domain.

Unit testing is crucial in software engineering for ensuring quality. However, it's not widely used in parallel and high-performance computing software, particularly scientific applications, due to their smaller, diverse user base and complex logic. These factors make unit testing challenging and expensive, as it requires specialized knowledge and existing automated tools are often ineffective. To address this, we propose an automated method for generating unit tests for such software, considering their unique features like complex logic and parallel processing. Recently, large language models (LLMs) have shown promise in coding and testing. We explored the capabilities of Davinci (text-davinci-002) and ChatGPT (gpt-3.5-turbo) in creating unit tests for C++ parallel programs. Our results show that LLMs can generate mostly correct and comprehensive unit tests, although they have some limitations, such as repetitive assertions and blank test cases.

Large language models (LLMs) have revolutionized software development practices, yet concerns about their safety have arisen, particularly regarding hidden backdoors, aka trojans. Backdoor attacks involve the insertion of triggers into training data, allowing attackers to manipulate the behavior of the model maliciously. In this paper, we focus on analyzing the model parameters to detect potential backdoor signals in code models. Specifically, we examine attention weights and biases, and context embeddings of the clean and poisoned CodeBERT and CodeT5 models. Our results suggest noticeable patterns in context embeddings of poisoned samples for both the poisoned models; however, attention weights and biases do not show any significant differences. This work contributes to ongoing efforts in white-box detection of backdoor signals in LLMs of code through the analysis of parameters and embeddings., Comment: This work has been accepted at the 1st ACM International Conference on AI-powered Software (AIware), co-located with the ACM International Conference on the Foundations of Software Engineering (FSE) 2024, Porto de Galinhas, Brazil. arXiv admin note: substantial text overlap with arXiv:2402.12936

Large language models (LLMs) have provided a lot of exciting new capabilities in software development. However, the opaque nature of these models makes them difficult to reason about and inspect. Their opacity gives rise to potential security risks, as adversaries can train and deploy compromised models to disrupt the software development process in the victims' organization. This work presents an overview of the current state-of-the-art trojan attacks on large language models of code, with a focus on triggers -- the main design point of trojans -- with the aid of a novel unifying trigger taxonomy framework. We also aim to provide a uniform definition of the fundamental concepts in the area of trojans in Code LLMs. Finally, we draw implications of findings on how code models learn on trigger design., Comment: arXiv admin note: substantial text overlap with arXiv:2305.03803

Trojan signatures, as described by Fields et al. (2021), are noticeable differences in the distribution of the trojaned class parameters (weights) and the non-trojaned class parameters of the trojaned model, that can be used to detect the trojaned model. Fields et al. (2021) found trojan signatures in computer vision classification tasks with image models, such as, Resnet, WideResnet, Densenet, and VGG. In this paper, we investigate such signatures in the classifier layer parameters of large language models of source code. Our results suggest that trojan signatures could not generalize to LLMs of code. We found that trojaned code models are stubborn, even when the models were poisoned under more explicit settings (finetuned with pre-trained weights frozen). We analyzed nine trojaned models for two binary classification tasks: clone and defect detection. To the best of our knowledge, this is the first work to examine weight-based trojan signature revelation techniques for large-language models of code and furthermore to demonstrate that detecting trojans only from the weights in such models is a hard problem., Comment: This work has been accepted at the International Conference on Learning Representations 2024 Workshop on Secure and Trustworthy Large Language Models, SeT LLM @ ICLR 2024 (Vienna, Austria)

Large language models (LLMs) have revolutionized software development practices, yet concerns about their safety have arisen, particularly regarding hidden backdoors, aka trojans. Backdoor attacks involve the insertion of triggers into training data, allowing attackers to manipulate the behavior of the model maliciously. In this paper, we focus on analyzing the model parameters to detect potential backdoor signals in code models. Specifically, we examine attention weights and biases, activation values, and context embeddings of the clean and poisoned CodeBERT models. Our results suggest noticeable patterns in activation values and context embeddings of poisoned samples for the poisoned CodeBERT model; however, attention weights and biases do not show any significant differences. This work contributes to ongoing efforts in white-box detection of backdoor signals in LLMs of code through the analysis of parameters and activations.

Large language models (LLMs) are becoming an integrated part of software development. These models are trained on large datasets for code, where it is hard to verify each data point. Therefore, a potential attack surface can be to inject poisonous data into the training data to make models vulnerable, aka trojaned. It can pose a significant threat by hiding manipulative behaviors inside models, leading to compromising the integrity of the models in downstream tasks. In this paper, we propose an occlusion-based human-in-the-loop technique, OSeql, to distinguish trojan-triggering inputs of code. The technique is based on the observation that trojaned neural models of code rely heavily on the triggering part of input; hence, its removal would change the confidence of the models in their prediction substantially. Our results suggest that OSeql can detect the triggering inputs with almost 100% recall. We discuss the problem of false positives and how to address them. These results provide a baseline for future studies in this field.

With the rapid growth of research in trojaning deep neural models of source code, we observe that there is a need of developing a benchmark trojaned models for testing various trojan detection and unlearning techniques. In this work, we aim to provide the scientific community with diverse trojaned code models, that cover a variety of state-of-the-art architectures, on which they can examine such techniques. We thus present TrojanedCM, a publicly available repository of clean and poisoned models of source code. We provide poisoned models for two code classification tasks (defect detection and clone detection) and a code generation task (text-to-code generation). We finetuned popular pretrained code models such as CodeBERT, PLBART, CodeT5, CodeT5+, on poisoned datasets that we generated from benchmark datasets (Devign, BigCloneBench, CONCODE) for the above mentioned tasks. The repository also provides full access to the architecture and parameters of the models, allowing practitioners to investigate different white-box analysis techniques. In addition to the poisoned models, we also provide a poisoning framework using which practitioners can deploy various poisoning strategies for the different tasks and models of source code. All the material are accessible via this link: https://github.com/UH-SERG/TrojanedCM.

In this work, we study literature in Explainable AI and Safe AI to understand poisoning of neural models of code. In order to do so, we first establish a novel taxonomy for Trojan AI for code, and present a new aspect-based classification of triggers in neural models of code. Next, we highlight recent works that help us deepen our conception of how these models understand software code. Then we pick some of the recent, state-of-art poisoning strategies that can be used to manipulate such models. The insights we draw can potentially help to foster future research in the area of Trojan AI for code.

Although deep neural models substantially reduce the overhead of feature engineering, the features readily available in the inputs might significantly impact training cost and the performance of the models. In this paper, we explore the impact of an unsuperivsed feature enrichment approach based on variable roles on the performance of neural models of code. The notion of variable roles (as introduced in the works of Sajaniemi et al. [Refs. 1,2]) has been found to help students' abilities in programming. In this paper, we investigate if this notion would improve the performance of neural models of code. To the best of our knowledge, this is the first work to investigate how Sajaniemi et al.'s concept of variable roles can affect neural models of code. In particular, we enrich a source code dataset by adding the role of individual variables in the dataset programs, and thereby conduct a study on the impact of variable role enrichment in training the Code2Seq model. In addition, we shed light on some challenges and opportunities in feature enrichment for neural code intelligence models., Comment: Accepted in the 1st International Workshop on Interpretability and Robustness in Neural Software Engineering (InteNSE'23), Co-located with ICSE

Finding important features that contribute to the prediction of neural models is an active area of research in explainable AI. Neural models are opaque and finding such features sheds light on a better understanding of their predictions. In contrast, in this work, we present an inverse perspective of distractor features: features that cast doubt about the prediction by affecting the model's confidence in its prediction. Understanding distractors provide a complementary view of the features' relevance in the predictions of neural models. In this paper, we apply a reduction-based technique to find distractors and provide our preliminary results of their impacts and types. Our experiments across various tasks, models, and datasets of code reveal that the removal of tokens can have a significant impact on the confidence of models in their predictions and the categories of tokens can also play a vital role in the model's confidence. Our study aims to enhance the transparency of models by emphasizing those tokens that significantly influence the confidence of the models., Comment: The 1st International Workshop on Interpretability and Robustness in Neural Software Engineering, Co-located with ICSE (InteNSE'23)

Neural code intelligence (CI) models are opaque black-boxes and offer little insight on the features they use in making predictions. This opacity may lead to distrust in their prediction and hamper their wider adoption in safety-critical applications. Recently, input program reduction techniques have been proposed to identify key features in the input programs to improve the transparency of CI models. However, this approach is syntax-unaware and does not consider the grammar of the programming language. In this paper, we apply a syntax-guided program reduction technique that considers the grammar of the input programs during reduction. Our experiments on multiple models across different types of input programs show that the syntax-guided program reduction technique is faster and provides smaller sets of key tokens in reduced programs. We also show that the key tokens could be used in generating adversarial examples for up to 65% of the input programs., Comment: The 6th ACM SIGPLAN International Symposium on Machine Programming (MAPS'22); Related to arXiv:2202.06474

With the wide spread use of AI-driven systems in the edge (a.k.a edge intelligence systems), such as autonomous driving vehicles, wearable biotech devices, intelligent manufacturing, etc., such systems are becoming very critical for our day-to-day lives. A challenge in designing edge intelligence systems is that we have to deal with a large number of constraints in two design spaces that form the basis of such systems: the edge design space and the deep learning design space. Thus in this work, a new systematic, extendable, manual approach, READLE, is proposed for creating representations of specifications in edge intelligent systems, capturing constraints in the edge system design space (e.g. timing constraints and other performance constraints) and constraints in the deep learning space (e.g. model training duration, required level of accuracy) in a coherent fashion. In particular, READLE leverages benefits of real-time logic and binary decision diagrams to generate unified specifications. Several insights learned in building READLE are also discussed, which should help in future research in the domain of formal specifications for edge intelligent systems.

Software fuzzing mutates bytes in the test seeds to explore different behaviors of the program under test. Initial seeds can have great impact on the performance of a fuzzing campaign. Mutating a lot of uninteresting bytes in a large seed wastes the fuzzing resources. In this paper, we present the preliminary results of our approach that aims to improve the performance of fuzzers through identifying and removing uninteresting bytes in the seeds. In particular, we present DIAR, a technique that reduces the size of the seeds based on their coverage. Our preliminary results suggest fuzzing campaigns that start with reduced seeds, find new paths faster, and can produce higher coverage overall.

Software fuzzing is a strong testing technique that has become the de facto approach for automated software testing and software vulnerability detection in the industry. The random nature of fuzzing makes monitoring and understanding the behavior of fuzzers difficult. In this paper, we report the development of Fuzzer Mutation Visualizer (FMViz), a tool that focuses on visualizing byte-level mutations in fuzzers. In particular, FMViz extends American Fuzzy Lop (AFL) to visualize the generated test inputs and highlight changes between consecutively generated seeds as a fuzzing campaign progresses. The overarching goal of our tool is to help developers and students comprehend the inner-workings of the AFL fuzzer better. In this paper, we present the architecture of FMViz, discuss a sample case study of it, and outline the future work. FMViz is open-source and publicly available at https://github.com/AftabHussain/afl-test-viz.

Deep Neural Networks (DNNs) are increasingly being used in software engineering and code intelligence tasks. These are powerful tools that are capable of learning highly generalizable patterns from large datasets through millions of parameters. At the same time, their large capacity can render them prone to memorizing data points. Recent work suggests that the memorization risk manifests especially strongly when the training dataset is noisy, involving many ambiguous or questionable samples, and memorization is the only recourse. The goal of this paper is to evaluate and compare the extent of memorization and generalization in neural code intelligence models. It aims to provide insights on how memorization may impact the learning behavior of neural models in code intelligence systems. To observe the extent of memorization in models, we add random noise to the original training dataset and use various metrics to quantify the impact of noise on various aspects of training and testing. We evaluate several state-of-the-art neural code intelligence models and benchmarks based on Java, Python, and Ruby codebases. Our results highlight important risks: millions of trainable parameters allow the neural networks to memorize anything, including noisy data, and provide a false sense of generalization. We observed all models manifest some forms of memorization. This can be potentially troublesome in most code intelligence tasks where they rely on rather noise-prone and repetitive data sources, such as code from GitHub. To the best of our knowledge, we provide the first study to quantify memorization effects in the domain of software engineering and code intelligence systems. This work raises awareness and provides new insights into important issues of training neural models in code intelligence systems that are usually overlooked by software engineering researchers., Comment: Information and Software Technology, IST Journal 2022, Elsevier

In this work we present an account of the status of requirements engineering in the gaming industry. Recent papers in the area were surveyed. Characterizations of the gaming industry were deliberated upon by portraying its relations with the market industry. Some research directions in the area of requirements engineering in the gaming industry were also mentioned.

Enriched by natural language texts, Stack Overflow code snippets are an invaluable code-centric knowledge base of small units of source code. Besides being useful for software developers, these annotated snippets can potentially serve as the basis for automated tools that provide working code solutions to specific natural language queries. With the goal of developing automated tools with the Stack Overflow snippets and surrounding text, this paper investigates the following questions: (1) How usable are the Stack Overflow code snippets? and (2) When using text search engines for matching on the natural language questions and answers around the snippets, what percentage of the top results contain usable code snippets? A total of 3M code snippets are analyzed across four languages: C\#, Java, JavaScript, and Python. Python and JavaScript proved to be the languages for which the most code snippets are usable. Conversely, Java and C\# proved to be the languages with the lowest usability rate. Further qualitative analysis on usable Python snippets shows the characteristics of the answers that solve the original question. Finally, we use Google search to investigate the alignment of usability and the natural language annotations around code snippets, and explore how to make snippets in Stack Overflow an adequate base for future automatic program generation., Comment: 13th IEEE/ACM International Conference on Mining Software Repositories, 11 pages