Latent reasoning through eight bottleneck tokens gives multimodal retrieval performance close to explicit chain-of-thought models at a fraction of the inference cost.
arXiv preprint arXiv:2602.10229 , year=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
LaME: Learning to Think in Latent Space for Multimodal Embedding via Information Bottleneck
Latent reasoning through eight bottleneck tokens gives multimodal retrieval performance close to explicit chain-of-thought models at a fraction of the inference cost.