CLOOB: Modern Hopfield Networks with InfoLOOB Outperform CLIP

Andreas Fürst, Elisabeth Rumetshofer, Johannes Lehner, Viet Tran, Fei Tang, Hubert Ramsauer, David Kreil, Michael Kopp, Günter Klambauer, Angela Bitto-Nemling, Sepp Hochreiter

Neural Information Processing Systems 

Therefore, with high probability the term DE is small. Next, we show that DE is governed by the variance of sim( x, y) for unmatched pairs.

Similar Docs  Excel Report  more

TitleSimilaritySource
None found