作者位次 #2/8ACL 2026Poster· Natural Language Processing
Empirical Analysis of Decoding Biases in Masked Diffusion Models
Pengcheng Huang · Tianming Liu · Zhenghao Liu · Yukun Yan · Shuo Wang · Tong Xiao · Zulong Chen · Maosong Sun
摘要
Masked Diffusion Models (MDMs) have recently emerged as a promising non-autoregressive paradigm for sequence generation. However, their performance is highly sensitive to the choice of decoding strategy. In this work, we reveal that prevalent uncertainty-based decoding strategies induce two decod...