Original title: Long-Range Indirect Control-Flow Prediction in Stripped Binaries via Dual Virtual Hubs and Multi-Task Graph Learning
Kun Liu, Zhengming Ding, Chenke Luo, Tianyi Xu +3 · September 2026
When security experts inspect compiled computer programs stripped of text clues, they must map indirect jumps. These are paths where code leaps to a target decided while the software runs. Current tools struggle when the clues and targets are far apart. To solve this, researchers built ICFlowNet. It uses two central routing hubs, called a Global Code Hub and a Global Data Hub. These hubs act like central airports, forming quick shortcuts that connect distant clues in code and data. At the same time, the system uses multi-task graph learning to tackle four jump types together instead of apart. The team discovered that their hubs boost long-range accuracy by up to 9.13 F1 points, joint learning adds up to 5.81 points, and their tool beats older methods by over 13 points.
Imagine you're trying to reconstruct a city's traffic flow from surveillance cameras, but most cameras are offline and you only have snippets. Normally, you'd look at each camera separately, but here, you need to connect distant snippets to see full routes. In this paper, the virtual hubs act like relay stations that link far-apart camera feeds, and the multi-task learning helps predict not just car movements but also buses and bikes all at once.
Security analysts need to understand compiled programs without source code labels. Better prediction of code jumps helps find vulnerabilities and secure systems, especially for stripped binaries.
The abstract doesn't explicitly list limitations, but the dataset is specific to x86-64 binaries and only 1,351 of 15,901 have dynamic ground truth for validation.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Long-Range Indirect Control-Flow Prediction in Stripped Binaries via Dual Virtual Hubs and Multi-Task Graph Learning — Kun Liu, Zhengming Ding, Chenke Luo et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.