This paper proposes MalVIS, an android malware detection framework based on pyramid vision transformers V2(PVTv2). MalVIS leverages the hierarchical self-attention mechanism of (PVTv2) to enhance detection of android based malware binaries. The pyramid structure of MalVIS enables it to efficiently capture the fine-grained and high-level features. It employs Spatial Reduction Attention (SRA) to reduce computational complexity by decreasing the number of tokens at each stage, which makes it suitable for deployment on resource-constrained environments. Additionally, MalVIS benefits from convolutional feed-forward networks (FFNs) to improve feature representation for effective malware classification. Experimental analysis on the benchmark MALNET-IMAGE dataset shows that MalVIS outperforms many state-of-the-art android malware detection frameworks, such as ViT-B Sherlock, ResNet, DenseNet, and MobileNetV2. It achieves an F1-score of 0.965 (binary classification) and 0.724 (multi-class classification) on the MALNET-IMAGE dataset.

This paper presents VULDetect: a lightweight and highly accurate ensemble based software system vulnerability detection framework trained on a large C/C++ file corpus comprising thousands of vulnerable and non-vulnerable source code snippets. VULDetect uses a deep learning based ensemble model comprising multiple classifiers to perform a comprehensive vulnerability assessment of the software programs. Experimental results on a proprietary dataset and a benchmark DIVERSEVUL dataset show that VULDetect is capable of detecting wide range of software system vulnerabilities with high accuracy and low false positive rate. In addition, VULDetect is also highly compact and lightweight in comparison to other Large Language Model (LLM) based frameworks proposed in the literature, which makes it highly suitable for deployment on computation constrained edge devices.

Abstract Word embedding techniques have been proposed in the literature to analyze and determine the sentiments expressed in various textual documents such as social media posts, online product reviews, and so forth. However, it is difficult to capture the entire gamut of intricate inter‐dependencies among words in the textual documents using a specific word embedding technique. In this article, we aim to address this issue by proposing a computation‐efficient stacking ensemble based sentiment analysis framework using multiple word embeddings. The proposed framework uses a combination of three distinct word embeddings generated by three different state of the art word embedding techniques, namely, Word2Vec, GloVe, and BERT for performing the sentiment analysis task. It uses an explicitly trained Word2Vec model to generate the first set of 200‐dimensional word embedding. Similarly, pre‐trained GloVe and BERT models are used to generate the other two sets of 200‐dimensional and a 768‐dimensional word embeddings, respectively. These three distinct word embedding sets are then used to train a heterogeneous stacking ensemble based classifier model comprising LSTM, GRU, and Bi‐GRU based base‐level classifiers, and a LSTM based meta‐level classifier. Experimental results on four different datasets, namely, Sentiment140, IMDB Review, Twitter conversation thread, and Twitter Emotion show that the proposed framework achieves high performance with low false positive rate. The proposed framework is also shown to outperform other sentiment analysis frameworks proposed in the literature.

Abstract Signature‐based intrusion detection systems (IDSs) are employed to monitor computer networks for signs of network intrusions. However, they produce a large number of false positive alarms when operated with default settings without considering the underlying network environment. Inundation of false alarms is the Achilles heel of IDS technology, which could render the IDS ineffective in detecting network attacks. Several false alarm minimization approaches have been proposed in the literature. However, there are many drawbacks associated with these works, namely, modification of well‐established attack signatures; heavy dependence on the attack signatures' reference numbers, which might not always be available; and non‐consideration of the underlying network context information. In this paper, we propose an efficient game theory‐based false alarm minimization scheme for signature‐based IDS. The proposed scheme uses a game theory‐based correlation engine to correlate IDS alarms with network vulnerabilities to minimize the overall false positive alarm rate of the IDS. Experimental results and comparison analysis of the proposed false alarm minimization framework with other frameworks on the benchmark DARPA intrusion detection evaluation dataset and an in‐house IIT Guwahati Lab dataset show that the proposed scheme achieves the highest accuracy among all the frameworks under consideration without degrading the overall detection rate of the IDS. Copyright © 2016 John Wiley & Sons, Ltd.