J-Miner: Recovering the Decision Logic of Fine-Tuned LLM Classifiers as Compact Rules

Name: J-Miner: Recovering the Decision Logic of Fine-Tuned LLM Classifiers as Compact Rules
Uploaded: 2026-10-03T23:58:52.458339+00:00
Duration: 368 s
Description: J-Miner extracts hidden decision logic from fine-tuned large language models by mining internal signals to create compact, executable Boolean rules. It recovers these rules through three stages—Recover, Reveal, and Transfer—allowing knowledge to be inspected and transferred to a much smaller model without needing the original complex classifier.

J-Miner extracts hidden decision logic from fine-tuned large language models by mining internal signals to create compact, executable Boolean rules. It recovers these rules through three stages—Recover, Reveal, and Transfer—allowing knowledge to be inspected and transferred to a much smaller model without needing the original complex classifier.

Open the full episode