Image-Encoding-Based IPv6 Target Generation for Few-Seed Prefixes
A way to generate probe targets in the IPv6 address space when only a handful of known-active seed addresses exist under a routing prefix. Each 128-bit address is laid out as an 8x16 binary image, one row per hexadecimal group, so that the same bit position across groups shares a column and the model can read configuration structure spatially; seed images are clustered by variational-autoencoder latent features, stitched together to enlarge the training signal, and modelled autoregressively so that generated images decode back into candidate addresses, with probe results fed back as fine-tuning data to correct the seeds sampling bias.
Generation in Few-seed Scenarios Wenjian Zhang1, Guanglei Song2, Lin He1,2, Jinlei Lin1, Songyun
Brute-force scanning the 128-bit IPv6 address space is impossible, so measurement tools instead run a target generation algorithm: given a set of known-active seed addresses, it learns the addressing…