Embedding Space Collapse
A phenomenon in which the representation vectors a model produces for different inputs become highly similar, reducing the discriminability between vectors and impairing the performance of similarity-based clustering, retrieval and matching tasks.
Timeline (last 90 days)
Several widely used multilingual models exhibited embedding space collapse, i.e., representations of different inputs became highly similar, making them unsuitable for similarity-based clustering tasks.
Entity clustering in coreference resolution relies on similarity-based clustering ability, so embedding collapse directly harms model performance.
All Facts (2)
Several widely used multilingual models exhibited embedding space collapse, i.e., representations of different inputs became highly similar, making them unsuitable for similarity-based clustering tasks.
50%UnverifiedEntity clustering in coreference resolution relies on similarity-based clustering ability, so embedding collapse directly harms model performance.
50%