IFIP Open Access Digital Library
Not a member yet
22614 research outputs found
Sort by
The Impact of Augmentation Techniques on Icon Detection Using Machine Learning Techniques
No full textPart 2: Machine LearningInternational audienceThis article examines the use of image augmentation techniques to improve icon detection in mobile interfaces, a critical task due to the small size of graphical user interface (GUI) elements and the insufficiency of comprehensive datasets. It evaluates whether diversifying the dataset or using specific augmentation methods alone can enhance detection performance. The study specifically compares the performance of two models, Faster R-CNN and YOLOv8, in detecting these elements, highlighting the challenges and potential solutions in automating complex process interactions through improved icon recognition. The findings from our computational experiments reveal that the application of classical image augmentation methods to enhance dataset diversity significantly improves the performance of these models. Remarkably, such augmentation techniques are capable of yielding results that are comparable to, or even exceed, those obtained from training the models on considerably larger datasets. It is particularly noteworthy that models demonstrate superior performance when they are initially provided with a substantial volume of annotations, surpassing the outcomes associated with models trained on extensive data collections. Among the various augmentation techniques evaluated, image rotation emerged as the most effective in enhancing the performance of both models. Nonetheless, it was observed that the Faster R-CNN model consistently outperformed the YOLOv8 model across all experiments
Large Scale Heap Dump Embedding for Machine Learning: Predicting OpenSSH Key Locations
No full textInternational audienceWith the evolving landscape of cybersecurity, forensic analysis has become increasingly pivotal, especially with the integration of machine learning (ML) techniques. However, the use of ML in cybersecurity, and especially in heap dump analysis is still in its infancy, both due to the lack of quality datasets and the difficulty of processing large-scale heap dumps collections. Another complex task lies in the transition from raw byte heap dump data into dense vector representations, or embeddings, that can be used with ML models. This paper addresses these challenges by introducing a novel methodology and the Mem2Graph tool for processing large-scale heap dumps collections. This method has been introduced while developing a novel approach to enhance the detection of session keys in OpenSSH heap dumps. Such a novel approach has significantly advanced the state of the art in predicting the location of keys in OpenSSH heap dumps. Importantly, it paves the way for automated ML applications that leverage the structure and embeddings from reconstructed memory graphs, opening new frontiers in both cybersecurity and data science
Hiding Your Awful Online Choices Made More Efficient and Secure: A New Privacy-Aware Recommender System
No full textInternational audienceRecommender systems are an integral part of online platforms that recommend new content to users with similar interests. However, they demand a considerable amount of user activity data where, if the data is not adequately protected, constitute a critical threat to the user privacy. Privacy-aware recommender systems enable protection of such sensitive user data while still maintaining a similar recommendation accuracy compared to the traditional non-private recommender systems. However, at present, the current privacy-aware recommender systems suffer from a significant trade-off between privacy and computational efficiency. For instance, it is well known that architectures that rely purely on cryptographic primitives offer the most robust privacy guarantees, however, they suffer from substantial computational and network overhead. Thus, it is crucial to improve this trade-off for better performance. This paper presents a novel privacy-aware recommender system that combines privacy-aware machine learning algorithms for practical scalability and efficiency with cryptographic primitives like Homomorphic Encryption and Multi-Party Computation - without assumptions like trusted-party or secure hardware - for solid privacy guarantees. Experiments on standard benchmark datasets show that our approach results in time and memory gains by three orders of magnitude compared to using cryptographic primitives in a standalone for constructing a privacy-aware recommender system. Furthermore, for the first time our method makes it feasible to compute private recommendations for datasets containing 100 million entries, even on memory-constrained low-power SOC (System on Chip) devices
“Alexa, How Do You Protect My Privacy?” A Quantitative Study of User Preferences and Requirements About Smart Speaker Privacy Settings
No full textInternational audienceVoice assistants are becoming increasingly popular. While users may benefit from the convenience of voice interactions, the use of voice assistants raises privacy issues. To address them, existing voice assistants propose some privacy settings. However, we are lacking knowledge about (1) which privacy settings are important to users, (2) what are their preferences about their application, and (3) what are their requirements beyond existing privacy settings. Gaining such knowledge is important to understand why users may not use these settings and to identify which settings should be introduced to allow users to better protect their privacy. To this end, we have conducted a quantitative online study with 1,103 German smart speaker owners. In addition to partly replicating findings obtained with different samples, the results show that the currently available privacy settings do not fully cover user requirements and indicate a general desire for more transparency and control over the collected data. Our results hence serve as basis for designing future privacy-preserving solutions
Exploring a Low-Cost Hardware Reverse Engineering Approach: A Use Case Experiment
No full textInternational audienceThis research delves into Hardware Reverse Engineering, specifically focusing on hardware manipulation. It leverages the Supply Chain attack vector and conducts a practical experiment to manipulate sensor trustworthiness. Results demonstrate that identifying weaknesses and exploiting hardware vulnerabilities can be achieved without costly reverse-engineering platforms. The study further benefits novices venturing into Hardware Reverse Engineering research by advocating a low-cost approach using open-access knowledge
Machine Learning Models for Predicting Celiac Disease Based on Non-invasive Clinical Symptoms
No full textPart 1: Biomedical/ClassificationInternational audienceCeliac Disease is an autoimmune disease with a prevalence between 1–1.8% worldwide. As it is believed to be highly underdiagnosed, new approaches and protocols are being explored for this purpose. In this paper, we aim to use artificial intelligence models to support the medical diagnosis of Celiac Disease from clinical symptoms and biomarkers. Through our experiments, we identified accurate models able to predict Celiac Disease diagnosis from noninvasive clinical indicators with an accuracy of 0.97, and the specific type of disease from 6 different classes with an accuracy of 0.92. Thus, even with the limitations of the present study, we conclude that machine learning can help support the diagnosis of Celiac Disease and provide an accurate identification of its type. We encourage researchers to collect and share data on Celiac Disease as it is necessary for improving the intelligent models and their robustness
IRfold: An RNA Secondary Structure Prediction Approach
No full textPart 1: Biomedical/ClassificationInternational audienceRibonucleic acid (RNA) sequences can be viewed as an ordered list of symbols representing the component nucleobases of the RNA sequence also known as its primary structure. The prediction of an RNA’s secondary structure from its primary structure involves predicting which of its bases are most likely to pair. Computing the likelihood of each pair to obtain the set of pairings with the highest cumulative probability is computationally intractable. We propose a new approach, IRfold, which considers possible base pairings across secondary structures known as inverted repeats (IRs) and composes a secondary structure prediction. Our approach identifies the set of minimal thermodynamic free energy IRs that satisfies empirically determined thermodynamic and steric constraints. The proposed method is implemented as a constraint programming problem, which is benchmarked against state-of-the-art secondary structure prediction approaches on the bpRNA-1m dataset. Our results yield promising initial outcomes, and we discuss potential avenues for further investigation
Information Ethics in Knowledge Sharing, Transfer, and Translation
No full textPart 3: General TrackInternational audienceTechnology development practices in many low-income countries result in less-than-optimal outcomes for their intended beneficiaries. Information and Communication Technologies for Development projects aim to mitigate such inefficiencies and inequities. One means is through knowledge sharing, and its encompassing functions: knowledge transfer and knowledge translation. The practices may have unintended detrimental consequences, which we scrutinize through the lens of information ethics. We present examples from an Emergency Medicine project in the Democratic Republic of Congo to explore the ethical imperatives associated with enabling access to information and the need for knowledge translation. Findings from three focus group discussions are presented that draw attention to moral obligations of agents engaging in knowledge sharing, and the importance of addressing information ethics in related initiatives
Perceptions on the Adoption of Blockchain for FinTech Applications in the Banking Industry in Developing Countries
No full textPart 3: General TrackInternational audienceThis study explores the adoption of blockchain for FinTech applications in the banking industry in developing countries to understand the perceptions and identify use cases and factors affecting the adoption. Thematic analysis was performed on data collected through semi-structured interviews across four banking-related firms. Results suggest a lack of understanding and FinTech perceptions negatively associated with Bitcoin. It is proposed that through further education and implementation of working production-grade use cases, the perception will improve. Several use cases were identified, with Real-Time Gross Settlement as the most prevalent. Based on the Technology-Organisation-Environment (TOE) model particularised, 13 factors were found to affect the adoption positively or negatively. Theoretical insights gained about TOE suggest the issue of similar factors designated using different terminologies across different related studies
Table Orientation Classification Model Based on BERT and TCSMN
No full textPart 1: Machine LearningInternational audienceTables are commonly used for structuring and consolidating knowledge, significantly enhancing the efficiency for human readers to acquire relevant information. However, due to their diverse structures and open domains, employing computational methods for their automatic analysis remains a substantial challenge. Among these challenges, accurately classifying the forms of tables is fundamental for achieving deep comprehension and analysis, forming the basis for understanding, retrieving, and extracting knowledge within tables. Common table formats include row tables, column tables, and matrix tables, where data is arranged in rows, columns, and combinations of rows and columns, respectively. This paper introduces a novel approach for table classification based on the neural network model, TableTC. TableTC initially utilizes fine-tuning of the BERT pre-trained model to comprehend table content. Additionally, it proposes an improved Temporal Convolutional Network (TCN) named Temporal Convolutional Sparse Multilayer Perceptron Network (TCSMN). This network captures sequential structural features of cells and their surrounding neighbors, enhancing the ability to extract semantic features and positions. Finally, it employs an attention mechanism to further augment the capability of extracting row-column positions and semantic features. The evaluation of our proposed method is conducted using table data from scientific literature found in the PubMed Central website. Experimental results demonstrate that TableTC achieves a 2.7% improvement in table classification accuracy, as measured by the F1 score, compared to previous state-of-the-art methods on this dataset