Crow: Eliminating Backdoors From Large Language Models Via Internal Consistency Regularization,
2025
Singapore Management University
Crow: Eliminating Backdoors From Large Language Models Via Internal Consistency Regularization, Nay Myat Min, Long H. Pham, Yige Li, Jun Sun
Research Collection School Of Computing and Information Systems
Large Language Models (LLMs) are vulnerable to backdoor attacks that manipulate outputs via hidden triggers. Existing defense methods—designed for vision/text classification tasks—fail for text generation. We propose Internal Consistency Regularization (CROW), a defense leveraging the observation that backdoored models exhibit unstable layer-wise hidden representations when triggered, while clean models show smooth transitions. CROW enforces consistency across layers via adversarial perturbations and regularization during finetuning, neutralizing backdoors without requiring clean reference models or trigger knowledge—only a small clean dataset. Experiments across Llama-2 (7B, 13B), CodeLlama (7B, 13B), and Mistral-7B demonstrate CROW’s effectiveness: it achieves significant reductions in attack success rates across …
An On-The-Fly Synthesis Framework For Ltl Over Finite Traces,
2025
Singapore Management University
An On-The-Fly Synthesis Framework For Ltl Over Finite Traces, Shengping Xiao, Yongkang Li, Shufang Zhu, Jun Sun, Jianwen Li, Geguang Pu, Moshe Vardi
Research Collection School Of Computing and Information Systems
We present an on-the-fly synthesis framework for Linear Temporal Logic over finite traces (LTLf) based on top-down deterministic automata construction. Existing approaches rely on constructing a complete Deterministic Finite Automaton (DFA) corresponding to the LTLf specification, a process with doubly exponential complexity relative to formula size in the worst case. In this case, the synthesis cannot be conducted until the entire DFA is constructed. This inefficiency is the main bottleneck of existing approaches. To address this challenge, we first present a method for converting LTLf into Transition-based DFA (TDFA) by directly leveraging LTLf semantics, incorporating intermediate results as direct components …
Repairing Adversarial Texts Through Perturbation,
2025
Singapore Management University
Repairing Adversarial Texts Through Perturbation, Guoliang Dong, Jingyi Wang, Jun Sun, Sudipta Chattopadhyay, Xinyu Wang, Ting Dai, Jie Shi, Jin Song Dong
Research Collection School Of Computing and Information Systems
It is known that neural networks are subject to attacks through adversarial perturbations. Worse yet, such attacks are impossible to eliminate, i.e., the adversarial perturbation is still possible after applying mitigation methods such as adversarial training. Multiple approaches have been developed to detect and reject such adversarial inputs. Rejecting suspicious inputs however may not be always feasible or ideal. First, normal inputs may be rejected due to false alarms generated by the detection algorithm. Second, denial-of-service attacks may be conducted by feeding such systems with adversarial inputs. To address this, in this work, we focus on the text domain and …
Unambiguous Granularity Distillation For Asymmetric Image Retrieval,
2025
Singapore Management University
Unambiguous Granularity Distillation For Asymmetric Image Retrieval, Hongrui Zhang, Yi Xie, Haoquan Zhang, Cheng Xu, Xuandi Luo, Donglei Chen, Xuemiao Xu, Huaidong Zhang, Pheng Ann Heng, Shengfeng He
Research Collection School Of Computing and Information Systems
Previous asymmetric image retrieval methods based on knowledge distillation have primarily focused on aligning the global features of two networks to transfer global semantic information from the gallery network to the query network. However, these methods often fail to effectively transfer local semantic information, limiting the fine-grained alignment of feature representation spaces between the two networks. To overcome this limitation, we propose a novel approach called Layered-Granularity Localized Distillation (GranDist). GranDist constructs layered feature representations that balance the richness of contextual information with the granularity of local features. As we progress through the layers, the contextual information becomes more detailed, …
Runtime Anomaly Detection For Drones: An Integrated Rule-Mining And Unsupervised Learning Approach,
2025
Singapore Management University
Runtime Anomaly Detection For Drones: An Integrated Rule-Mining And Unsupervised Learning Approach, Ivan Wei Han Tan, Wei Minn, Christopher M. Poskitt, Lwin Khin Shar, Lingxiao Jiang
Research Collection School Of Computing and Information Systems
Unmanned Aerial Vehicles (UAVs), commonly referred to as drones, have witnessed a remarkable surge in popularity due to their versatile applications. These cyber-physical systems depend on multiple sensor inputs, such as cameras, GPS receivers, accelerometers, and gyroscopes, with faults potentially leading to physical instability and serious safety concerns. To mitigate such risks, anomaly detection has emerged as a crucial safeguarding mechanism, capable of identifying the physical manifestations of emerging issues and allowing operators to take preemptive action at runtime. Recent anomaly detection methods based on LSTM neural networks have shown promising results, but three challenges persist: the need for models …
Exploring The Capabilities Of Llms For Code-Change-Related Tasks,
2025
Singapore Management University
Exploring The Capabilities Of Llms For Code-Change-Related Tasks, Lishui Fan, Jiakun Liu, Zhongxin Liu, David Lo, Xin Xia, Shanping Li
Research Collection School Of Computing and Information Systems
Developers deal with code-change-related tasks daily, e.g., reviewing code. Pre-trained code and code-change-oriented models have been adapted to help developers with such tasks. Recently, large language models (LLMs) have shown their effectiveness in code-related tasks. However, existing LLMs for code focus on general code syntax and semantics rather than the differences between two code versions. Thus, it is an open question how LLMs perform on code-change-related tasks.To answer this question, we conduct an empirical study using 1B parameters LLMs on three code-change-related tasks, i.e., code review generation, commit message generation, and just-in-time comment update, with in-context learning (ICL) and parameter-efficient …
Explaining Explanations: An Empirical Study Of Explanations In Code Reviews,
2025
Singapore Management University
Explaining Explanations: An Empirical Study Of Explanations In Code Reviews, Ratnadira Widyasari, Ting Zhang, Abir Bouraffa, Walid Maalej, David Lo
Research Collection School Of Computing and Information Systems
Code reviews are central for software quality assurance. Ideally, reviewers should explain their feedback to enable authors of code changes to understand the feedback and act accordingly. Different developers might need different explanations in different contexts. Therefore, assisting this process first requires understanding the types of explanations reviewers usually provide. The goal of this article is to study the types of explanations used in code reviews and explore the potential of Large Language Models (LLMs), specifically ChatGPT, in generating these specific types. We extracted 793 code review comments from Gerrit and manually labeled them based on whether they contained a …
How Are We Detecting Inconsistent Method Names? An Empirical Study From Code Review Perspective,
2025
Singapore Management University
How Are We Detecting Inconsistent Method Names? An Empirical Study From Code Review Perspective, Kisub Kim, Xin Zhou, Dongsun Kim, Julia Lawall, Kui Liu, Tegawendé F. Bissyandé, Jacques Klein, Jaekwon Lee, David Lo
Research Collection School Of Computing and Information Systems
Proper naming of methods can make program code easier to understand, and thus enhance software maintainability. Yet, developers may use inconsistent names due to poor communication or a lack of familiarity with conventions within the software development lifecycle. To address this issue, much research effort has been invested into building automatic tools that can check for method name inconsistency and recommend consistent names. However, existing datasets generally do not provide precise details about why a method name was deemed improper and required to be changed. Such information can give useful hints on how to improve the recommendation of adequate method …
Expanding Access To Cybergis-Compute Through Support For Heterogeneous Workflows,
2025
University of Illinois at Urbana-Champaign
Expanding Access To Cybergis-Compute Through Support For Heterogeneous Workflows, Alexander C. Michels, Ian Zhang, Anand Padmanabhan, John Speaks, Rebecca Vandewalle, Shaowen Wang
I-GUIDE Forum
CyberGIS-Compute is a geospatial middleware tool designed to lower technical barriers to High-Performance Computing (HPC) resources. It provides end-users with a Graphical User Interface (GUI) for submitting models to HPC and allows model developers to contribute their workflows by adding a manifest to their repositories. However, the simplification of the user interface and streamlining of model contribution have unintentionally limited the scope of models that could be run on CyberGIS-Compute. In this paper, we discuss recent developments to the CyberGIS-Compute project that are aimed at supporting a wider variety of workflows including performance enhancements, supporting additional configuration options for jobs, …
A Systematic Evaluation Of Threaded Internode Communication In Hpc,
2025
University of New Mexico - Main Campus
A Systematic Evaluation Of Threaded Internode Communication In Hpc, William Pepper Marts
Computer Science ETDs
High Performance Computing (HPC) applications increasingly rely on both process and thread-level parallelism to maximize performance across complex, multi-node systems. However, conventional bulk synchronous communication strategies often leave both compute and network resources underutilized due to synchronization delays. This dissertation systematically evaluates the potential of fine-grained, threaded inter-node communication as a strategy for reducing these inefficiencies. To this end, I design and develop two tools: the MiniMod modular application framework and the Configurable Messaging Benchmark (CMB), which together enable empirical, reproducible assessment of communication performance across varying application behaviors, threading models, and communication granularities. Through experiments across multiple systems and …
How’S It Growing? Tools For Observing Snow And Sea Ice In A Changing Arctic Ocean,
2025
Thayer School of Engineering at Dartmouth College
How’S It Growing? Tools For Observing Snow And Sea Ice In A Changing Arctic Ocean, Ian Alexander Raphael
Dartmouth College Ph.D Dissertations
September Arctic sea ice extent has diminished by roughly 50% in the 45 years since satellite observations began. The Arctic Ocean may experience ice-free summers within the next decade, with implications for habitat, resource extraction, geopolitics, and local and global climate change. To predict how Arctic sea ice will change in the future, we need to understand its behavior in the present. In situ sea ice mass balance measurements (snow accumulation, ice growth, snow and ice surface melt, and bottom melt) are essential for studying the processes driving rapid changes in the ice pack, and for validating remote sensing measurements …
Improving Culinary Informatics Through Meaningful Social Web Engineering,
2025
Multimedia University, Malaysia
Improving Culinary Informatics Through Meaningful Social Web Engineering, Suraya Nurain Kalid, Fahmi Mikail Fahrid, Sin-Ban Ho
Journal of Informatics and Web Engineering
This paper gives a good reflection about social web engineering aspirations to ardently see the truth to enlighten culinary concentration and wisdom. With the compassionate path to discover food and engage customers to share content with loving kindness spread to every corner. There are so many features realized, which include direct interfaces through intuitive interfaces to lead the daily food streamlining. May one engaged with constructive creation to grow compassions in every culinary living sharing. Contentment would develop in our heart involved in different phases put forward in this paper. The initial phase begins with honesty in planning and design. …
K-Means Clustering Optimization Of Toddler Malnutrition Status Using Elbow Method,
2025
Dian Nuswantoro University, Indonesia
K-Means Clustering Optimization Of Toddler Malnutrition Status Using Elbow Method, Femmi Widyawati, Ahmad Yahya Dawod, Heru Agus Santoso
Journal of Informatics and Web Engineering
The problem of nutritional status is still a major challenge in the health sector in developing countries, including Indonesia. Malnutrition in toddlers can have serious long-term impacts on children's growth and development, including increased risk of disease, impaired cognitive function, and low productivity in the future. To overcome this problem, an in-depth analysis is needed to determine the distribution of nutritional status of toddlers in one of the provincial capitals in Indonesia, which can be used as a basis for planning more effective interventions. This study uses the K-Means algorithm to classify areas based on the prevalence of malnutrition in …
A Fundamental Framework For Analysis Of Rainfall Prediction Features Significance,
2025
Multimedia University, Malaysia
A Fundamental Framework For Analysis Of Rainfall Prediction Features Significance, Ye Zhian Teoh, Yim Ling Loo
Journal of Informatics and Web Engineering
Rainfall prediction efforts had been prevalent ever since the impact of climate change on occurrences of natural disasters globally. Implementation of machine and deep learning techniques on features that contribute to rainfall occurrences were conducted with aims of seeking greater prediction accuracy for rainfall occurrences with a lack of study for significance of features in rainfall occurrence prediction. This study presents a framework of rainfall prediction features' significance analysis in the case study of Peninsular Malaysia rainfall occurrences. Features investigated in this study consist of temperature, humidity and wind speed. The designed framework for the investigation includes phases of data …
Enhancing Citrus Plant Health Through The Application Of Image Processing Techniques For Disease Detection,
2025
SRM Valliammai Engineering College, India
Enhancing Citrus Plant Health Through The Application Of Image Processing Techniques For Disease Detection, C Pabitha, B Vanathi, K Revathi, S. Benila
Journal of Informatics and Web Engineering
The foremost task in agriculture is the decisive identification of citrus plants and the timely identification of diseases in the plants with the aim of improving the quality of crops and the yield. In this work, a machine learning algorithm focuses on image processing of citrus to solve issues that are significant and cause concern in agriculture. This work focus on the machine learning models like VGG 19 and VGG 16. In addition, dataset curation, data augmentation and various other methods were employed. The dataset used in this research is a composed one which is recorded in a comprehensive manner …
A New Bw Index For Quantifying Scholars' Research Influence,
2025
Taiyuan University of Technology, China
A New Bw Index For Quantifying Scholars' Research Influence, Bilal Ahmed, Li Wang, Waqar Hussain, Saim Qureshi
Journal of Informatics and Web Engineering
The growing importance of measuring and evaluating academic performance in academic hiring, promotions, funding allocation, and resource distribution has fueled the demand for better metrics. Traditional ranking indicators, such as publication count and citation-based indices, often fail to capture for the interdisciplinary influence and qualitative dimensions of research impact. These shortcomings highlight the need for more comprehensive evaluation metrics. The current study introduces a novel BW Index, which integrates both quantitative and qualitative aspects of researcher contributions aiming to provide a more balanced and comprehensive evaluation of scholarly impact. For evaluating the effectiveness of proposed index, a comparative analysis was …
Facilitating Digital Experience Sharing Among Vehicles Through Utilisation Of Pre-Existing Communication Infrastructure,
2025
University of Science & Technology, Pakistan
Facilitating Digital Experience Sharing Among Vehicles Through Utilisation Of Pre-Existing Communication Infrastructure, Asad Hussain, Umar Farooq, Ihsan Rabbi
Journal of Informatics and Web Engineering
Vehicular communication applications are expanding quickly because of the approach of related technologies, such as vehicular cloud and the Internet of Vehicles (IoV). The combination of the Internet of Things (IoT) and smart transportation is the Internet of vehicles. Data related to infotainment, safety, and effectiveness with different vehicles and the Sustainable framework of vehicles can be exchanged. However, after the appearance of such empowering advancements, still a huge number of ideas need research. Data sharing (related trips and navigation) of new/old models and other new/old model vehicles with the owner's agreement should be taken care of. This paper proposes …
Sentiment Analysis Of The 2024 General Election Through Twitter Using Long-Short-Term Memory Algorithm,
2025
Tokyo Metropolitan University, Japan
Sentiment Analysis Of The 2024 General Election Through Twitter Using Long-Short-Term Memory Algorithm, Angga Wahyu W, Haidar Hilmy Andana, Junta Zeniarja, Aris Febriyanto
Journal of Informatics and Web Engineering
This study analyses sentiment related to the 2024 Indonesian Presidential Election using the Long Short-Term Memory (LSTM) algorithm. A total of 2,400 tweets in the Indonesian language were gathered, with approximately 400 tweets sampled per week. In the data preparation, lexicon-based sentiment tagging, oversampling for class balance, and the creation and training of an LSTM model are all included in the study approach. The built model consists of embedding layers, Conv1D, and two LSTM layers. The LSTM model was selected due to its ability to capture long-range contextual dependencies in sequential text data like tweets, facilitated by its gate mechanisms …
Sentiment Analysis Of The Israel-Palestine Conflict On X: Insights From The Indonesian Perspective Using A Long Short-Term Memory Algorithm,
2025
Maulana Azad University, India
Sentiment Analysis Of The Israel-Palestine Conflict On X: Insights From The Indonesian Perspective Using A Long Short-Term Memory Algorithm, Mohammad Taleb Noori, Muhammad Alif Rahman, Agus Purnomo, Aripin
Journal of Informatics and Web Engineering
The Israel-Palestine conflict which has persisted for decades drives mounting global interest that consequently influences public opinion worldwide. This article examines the sentiment analysis of X (Twitter) data pertaining to the conflict using the Long Short-Term Memory (LSTM) model. This study presents public reactions through an analysis of 1,700 tweets collected between May and July 2023 which encapsulate key recent developments. In this study, several steps were conducted, namely 1) crawling process to get raw data; 2) preprocessing: cleansing, case folding, tokenization, stop word removal, and stemming; 3) modelling and validation using the LSTM model; 4) model evaluation based on …
Aspects-Based Sentiment Analysis Of Extreme Weather On Twitter Using Long Short-Term Memory,
2025
Termez State University, Uzbekistan
Aspects-Based Sentiment Analysis Of Extreme Weather On Twitter Using Long Short-Term Memory, Tursun Abdurahmonov, Muhammad Nabil Toby Abiyyu, Dzikru Nur Khayat, M. Ary Heryanto
Journal of Informatics and Web Engineering
This study presents an aspect-based sentiment analysis of tweets related to extreme weather events in Indonesia, utilizing the Long Short-Term Memory (LSTM) model. The dataset was obtained through a Twitter crawling process, followed by a series of preprocessing steps including data cleaning, stop word removal, normalization, tokenization, and stemming. The three primary areas of emphasis in the study were kinds of bad weather forecasts, and the government or society reactions. Using a lexicon-based technique, sentiment labelling generated three groups: positive, neutral, and negative. A random oversampling method was employed to address the data imbalance. The model using the LSTM algorithm …
