malimg dataset-基于图像分析的恶意软件检测和分类数据集

我们考虑了基于图像分析的恶意软件检测和分类问题。我们将可执行文件转换为图像，并使用深度学习（DL）模型应用图像识别。为了训练这些模型，我们基于已经在大量图像数据集上进行了预先训练的现有DL模型，采用转移学习。我们使用这种技术进行了各种实验，并将其性能与一种非常简单的机器学习技术（即k最近邻（k-NN））的性能进行了比较。对于我们的k-NN实验，我们使用直接从可执行文件中提取的特征，而不是图像分析。尽管我们的基于图像的DL技术在实验中表现良好，但令人惊讶的是，它的效果优于k-NN。我们证明DL模型在模拟零日实验中的性能优于k-NN，就可以更好地概括数据。

We consider the problem of malware detection and classification based on image analysis. We convert executable files into images and apply deep learning (DL) models for image recognition. To train these models, we adopt transfer learning using existing DL models that have been pre-trained on large-scale image datasets. We carried out a variety of experiments using this technique and compared its performance against that of a straightforward machine learning technique, namely k-nearest neighbor (k-NN). For our k-NN experiments, we utilized features directly extracted from executable files rather than employing image analysis. Although our image-based DL technique achieved favorable performance in the experiments, it surprisingly outperformed k-NN. We demonstrate that DL models outperform k-NN in simulated zero-day experiments, as they can better generalize to unseen data.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于恶意软件检测和分类的图像数据集，通过将可执行文件转换为图像并应用深度学习技术进行分析。数据集大小为25.3M，采用了转移学习方法，并与k-NN方法进行了性能比较。