A network intrusion detection (NID) system plays a critical role in cybersecurity. However, the existing machine learning-based NID research has a vital issue that their experimental settings do not reflect real-world situations where unknown attacks are constantly emerging. In particular, their train and test sets are from a single data set, which inevitably overestimates the detection power since all test attack types are known in training, and test cases will have similar characteristics to the training data. This paper introduces a new strategy to constitute test data with updated traffic with attack types not included in training data. In the proposed setting, the prediction accuracy of the existing detectors is dropped by about 20% compared to what has been reported. Also, in- depth analysis of detection performance by attack types has revealed that the existing models have strength at certain attack types but struggle to detect the other attack types such as DoS, DDoS, web attack, and port scan. To overcome the issues, we propose a new neural detector, called MHSA, based on a multi-head self-attention mechanism whose architecture suits better to capture scattered pieces of evidence in network traffic. Our model improved the overall detection performance by 29% in false positive rate at the true positive rate of 0.9 and by 9% in AUC over the current state-of-the-art models, successfully detecting the attacks that are not well captured before. Furthermore, we show that our proposed MHSA model even outperforms the best ensemble detector constructed by joining the state-of-the-art classifiers.

The rapid evolution of software development, propelled by competitive demands and the continuous integration of new features, frequently leads to inadvertent security oversights. Traditional security practices, often reactive in nature, primarily focus on identifying known vulnerabilities, creating a significant shortfall in detecting emergent, zero-day threats. This paper introduces CODE-SMASH, a novel deep learning-based source code vulnerability detector that utilizes a Siamese neural network with a hierarchical architecture integrating BiGRU and attention mechanisms. Our experiments using real-world datasets, specifically the Chromium and Debian datasets, demonstrate CODE-SMASH’s superiority over existing methods. It achieves significant improvements in detection performance across all key metrics, including accuracy, precision, recall, and F1-score, with average improvements of approximately 8.3%, 11.6%, 27.75%, and 17.7%, respectively, compared to the best-performing existing methods in our experiments. Moreover, CODE-SMASH shows its superior capability in handling complex and lengthy code sequences, with performance improvements for long-length code (60 to 80 lines) in F1 scores of 4.53 percentage points on the Chromium dataset and 5.62 percentage points on the Debian dataset compared to the second-best model’s performance. We believe our research makes a significant contribution to the field of automated vulnerability detection by providing a high-precision solution to the growing challenges in software security. Furthermore, based on our findings, we anticipate that future research could enhance CODE-SMASH by expanding its generalizability to various programming languages and reducing computational demands to improve efficiency.

The growing complexity and volume of modern software have led to an increase in source code vulnerabilities, posing significant security risks. In response, deep learning-based automated source code vulnerability detection methods, particularly those utilizing source code similarity analysis, have recently emerged as promising solutions. However, existing similarity-based source code vulnerability detection methods frequently fail to fully utilize information from the hierarchical structure of source code and are often computationally expensive, limiting their practicality in real-world scenarios. In this paper, we introduce XTransformer, a novel deep learning-based source code vulnerability detector tailored for comparing target source code against archived vulnerable codes across various levels of the source code’s hierarchical structure by leveraging extra cross-attention imposed on the transformer architecture. Additionally, we propose a specialized training strategy based on supervised contrastive learning to improve XTransformer’s ability to effectively learn and differentiate between vulnerable and non-vulnerable source codes. Comprehensive experiments demonstrate that XTransformer outperforms current state-of-the-art methods across different datasets and code lengths while significantly reducing the inference time compared to other similarity-based methods that utilize hierarchical information from source code.

Model-stealing attacks are emerging as a severe threat to AI-based services because an adversary can create models that duplicate the functionality of the black-box AI models inside the services with regular query-based access. To avoid detection or query costs, the model-stealing adversary must consider minimizing the number of queries to obtain an accurate clone model. To achieve this goal, we propose SwiftThief, a novel model-stealing framework that utilizes both queried and unqueried data to reduce query complexity. In particular, SwiftThief uses contrastive learning, a recent technique for representation learning. We formulate a new objective function for model stealing consisting of self-supervised (for abundant unqueried inputs from public datasets) and soft-supervised (for queried inputs) contrastive losses, jointly optimized with an output matching loss (for queried inputs). In addition, we suggest a new sampling strategy to prioritize rarely queried classes to improve attack performance. Our experiments proved that SwiftThief could significantly enhance the efficiency of model-stealing attacks compared to the existing methods, achieving similar attack performance using only half of the query budgets of the competing approaches. Also, SwiftThief showed high competence even when a defense was activated for the victims.

Universal application of AI has increased the need to explain why an AI model makes a specific decision in a human-understandable form. Among many related works, the class activation map (CAM)-based methods have been successful recently, creating input attribution based on the weighted sum of activation maps in convolutional neural networks. However, existing methods use channel-wise importance weights with specific architectural assumptions, relying on arbitrarily chosen attribution threshold values in their quality assessment: we think these can degrade the quality of attribution. In this paper, we propose Libra-CAM, a new CAM-style attribution method based on the best linear approximation of the layer (as a function) between the penultimate activation and the target-class score output. From the approximation, we derive the base formula of Libra-CAM, which is applied with multiple reference activations from a pre-built library. We construct Libra-CAM by averaging these base attribution maps, taking a threshold calibration procedure to optimize its attribution quality. Our experiments show that Libra-CAM can be computed in a reasonable time and is superior to the existing attribution methods in quantitative and qualitative attribution quality evaluations.

Model induction is one of the most popular methods to extract information to better understand AI’s decisions by estimating the contribution of input features for a class of interest. However, we found a potential issue: most model induction methods, especially those that compute class activation maps, rely on arbitrary thresholding to mute some of their computed attribution scores, which can cause the severe quality degradation of model induction. Therefore, we propose a new threshold fine-tuning (TFT) procedure to enhance the quality of input attribution based on model induction. Our TFT replaces arbitrary thresholding with an iterative procedure to find the optimal cut-off threshold value of input attribution scores using a new quality metric. Furthermore, to remove the burden of computing optimal threshold values on a per-input basis, we suggest an activation fine-tuning (AFT) framework using a tuner network attached to the original convolutional neural network (CNN), retraining the tuner-attached network with auxiliary data produced by TFT. The purpose of the tuner network is to make the activations of the original CNN less noisy and thus better suited for computing input attribution scores based on class activation maps from the activations. In our experiments, we show that the per-input optimal thresholding of attribution scores using TFT can significantly improve the quality of input attribution, and CNNs fine-tuned with our AFT can be used to produce improved input attribution matching the quality of TFT-tuned input attribution without requiring costly per-input threshold optimization.