PIE prunes CLT features first via FAP and FAP-Synergy to match baseline circuit fidelity at lower feature budgets on IOI and Doc-String tasks, reducing interpretation costs.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Prune, Interpret, Evaluate: A Cross-Layer Transcoder-Native Framework for Efficient Circuit Discovery via Feature Attribution
PIE prunes CLT features first via FAP and FAP-Synergy to match baseline circuit fidelity at lower feature budgets on IOI and Doc-String tasks, reducing interpretation costs.