Initializing a learnable attention mask with task structure, rather than the query-key projections, keeps the prior alive through training and changes Transformer extrapolation on Boolean and arithmetic tasks.
k-means mask transformer
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Mask-Based Priors Are More Persistent than Query-Key Initializations
Initializing a learnable attention mask with task structure, rather than the query-key projections, keeps the prior alive through training and changes Transformer extrapolation on Boolean and arithmetic tasks.