Construct · 1 fields · 1 citations

rank collapse in self-attention: a standalone construct

One name. One construct. A landmark on the map.

● anchor · a standalone construct Nearest construct by meaning·not indexed

A single standalone construct; nearby constructs are located against it in the Collider. Tap a field for its definition, unit, and measures.

Machine learning

the result that stacked pure attention layers drive token representations toward a rank-one matrix doubly exponentially with depth unless skip connections intervene.

In machine learning, rank collapse in self-attention refers to the result that stacked pure attention layers drive token representations toward a rank-one matrix doubly exponentially with depth unless skip connections intervene. This meaning comes from Dong (2021) in arXiv.[1]

Unit
Recovery means
Measured by

Run it through the Collider, or take the citations with you.

See how rank collapse in self-attention sits against neighbouring constructs in the Collider, or export every reference on this page in one click.

Open in Collider Export BibTeX Report this entry

Nearby constructs

attention5 fields · 10 citationsjingle · one name, five constructs collapse4 fields · 5 citationsjingle · one word, four constructs depth3 fields · 6 citationsjingle · one word, three constructs

References

Dong, Y., Cordonnier, J., & Loukas, A. (2021). Attention is Not All You Need: Pure Attention Loses Rank Doubly Exponentially with Depth. arXiv. https://doi.org/10.48550/arxiv.2103.03404

Cited on this page?

Confirm your construct, or correct us.

If your work is cited behind one of these entries, you know better than we do whether we captured the meaning right. Confirm it, suggest an edit, or dispute it, and your name goes on the entry as an author who has checked it, linked to your ORCID.

Claim your construct