1 paper · 1 filter
Timur Nabiev, Evgeny Frolov
Sequential self-attention models usually rely on additive positional embeddings, which inject positional information into item representations at the input. In the absence of posit…