Title: Compiler Provenance Recovery for Multi-CPU Architectures Using a Centrifuge Mechanism

Abstract: Bit-stream recognition (BSR) has many applications, such as forensic investigations, detection of copyright infringement, and malware analysis. We propose the first BSR that takes a bare input bit-stream and outputs a class label without any preprocessing. To achieve our goal, we propose a centrifuge mechanism, where the upstream layers (sub-net) capture global features and tell the downstream layers (main-net) to switch the focus, even if a part of the input bit-stream has the same value. We applied the centrifuge mechanism to compiler provenance recovery, a type of BSR, and achieved excellent classification. Additionally, downstream transfer learning (DTL), one of the learning methods we propose for the centrifuge mechanism, pre-trains the main-net using the sub-net's ground truth instead of the sub-net's output. We found that sub-predictions made by DTL tend to be highly accurate when the sub-label classification contributes to the essence of the main prediction.