1 citations · 1 across the 7 of their papers we have counts for
1 paper · 1 filter
Han Xiao
We frame embedding inversion as conditional masked diffusion, recovering all tokens in parallel through iterative denoising rather than sequential autoregressive generation. A mask…