Unverified50% confidenceFactExact time
Samplers in discrete diffusion (including the remask sampler and the uniform sampler) write to multiple token positions simultaneously in a single step, with each position sampled independently from its own marginal distribution.
1
Sources
50%
Confidence
Long-term
Relevance
10/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe study set up five experimental conditions, varying three variables—evidence masking, ownership marking, and neutral-fill replacement—across six initialization clusters, each containing two data orderings.71% similarUnverifiedThe root cause of the issue in this case was data leakage in the data split, where random assignment placed nearly duplicate form templates on both the training set and regression test set sides simultaneously69% similarUnverifiedWriting only one position or conditionally independent positions per step can theoretically match the distribution exactly, while writing more dependent positions causes errors to accumulate.69% similarUnverifiedSample-level leakage refers to the situation where, when multiple augmented samples originate from the same original sample, if the split is done randomly rather than by original individual, information about the same entity appears in both the training set and the validation set.68% similarUnverified分层抽样按各类别在总体中的比例进行抽样，sklearn的StratifiedShuffleSplit在此基础上增加随机打乱与多次重采样能力，特别适合类别不平衡场景67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/968468API
curl https://kongchang.com/api/v1/knowledge/claims/968468MCP
get_claim(id=968468)