Transformer attention implements Pavlovian conditioning in linear form, yielding a capacity bound of O(sqrt(d_k)) for worst-case error-free retrieval and insights on depth-width trade-offs.
Isaacson and Massimo Scanziani
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Understanding Transformers through the Lens of Pavlovian Conditioning
Transformer attention implements Pavlovian conditioning in linear form, yielding a capacity bound of O(sqrt(d_k)) for worst-case error-free retrieval and insights on depth-width trade-offs.