Problem
Research questions and friction points this paper is trying to address.
Interpretability
Turing Machines
Algorithmic Structure
Susceptibilities
Interpretability for Turing Machines
๐ค AI Summary
็ ็ฉถ้่ฟๅบ็จ็ฅ็ป็ฝ็ป็ๅฏ่งฃ้ๆงๆๆฏโโๆๆๆง๏ผๅๆๅพ็ตๆบ็ๅฑ้จๆๅคฑๆฏ่ง๏ผ่ฏๅซ็ฎๆณ็ปๆ๏ผๅนถ็จไธปๆๅๅๆๅ่็ฑปๆนๆณๆขๅค็ฎๆณ็นๅพใ
๐ Abstract
We show that susceptibilities, an interpretability technique developed for neural networks, can identify the presence of algorithmic structure in Turing machines by probing the local loss landscape of a learning problem for noisy Turing machines introduced by Murfet and Troiani (arXiv:2504.08075). We prove that symmetries and path separation in the algorithm implemented by a Turing machine induce permutation symmetries and low-rank blocks in its susceptibility matrix. We study this empirically on a set of deterministic finite automata (DFAs) and demonstrate that algorithmic features can be recovered by principal component analysis and clustering methods in susceptibility space.
Innovation
Methods, ideas, or system contributions that make the work stand out.
susceptibilities
Turing machines
local loss landscape
permutation symmetries
low-rank blocks
๐ Similar Papers
2024-07-02arXiv.orgCitations: 12
๐ผ Related Jobs
