UDIB, an upper-bounded Deterministic Information Bottleneck clustering algorithm, is applied to LLM prompt and response embeddings to produce topic spaces that sharpen semantic divergence metrics for hallucination analysis.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Topic Identification in LLM Input-Output Pairs through the Lens of Information Bottleneck
UDIB, an upper-bounded Deterministic Information Bottleneck clustering algorithm, is applied to LLM prompt and response embeddings to produce topic spaces that sharpen semantic divergence metrics for hallucination analysis.