[2P3-J-2] Machine learning: advances in neural networks

[2P3-J-2-04] Efficient learning in machine language instruction sequence classification by transfer learning

〇Yuhei Otsubo^1,2, Akira Otsuka², Yoshihiro Iwata¹, Mamoru Mimura³, Takeshi Sakaki⁴ (1. National Police Agency, 2. Institute of Information Security, 3. National Deffense Academy, 4. University of Tokyo)

Keywords:Transfer Learning, CNN, Machine language instruction sequence classification

In this paper, we propose a learning method for a machine language instruction sequence which is difficult to prepare a large-scale data set, e.g., packer classification.
We used the learning result of the compiler classification which is easy to prepare a large-scale data set.
Our examination result shows that it is possible to classify with accuracy of 92% even if only 10 samples of each class are gathered in 30 class classification.