RSF uses causal tracing and backtracking in a fine-tuned LLM to identify 'most probable' AD speech markers and generates synthetic transcripts that improve AD classification by about 3.5% accuracy.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Reverse-Speech-Finder: A Neural Network Backtracking Architecture for Generating Alzheimer's Disease Speech Samples and Improving Diagnosis Performance
RSF uses causal tracing and backtracking in a fine-tuned LLM to identify 'most probable' AD speech markers and generates synthetic transcripts that improve AD classification by about 3.5% accuracy.