Towards Meta-Models for Automated Interpretability

Lauro Langosco, Neel Alex, William Baker, David John Quarel, Herbie Bradley, David Krueger

Primary Area: visualization or interpretation of learned representations

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: interpretability, meta-models, automated interpretability, backdoors, trojans, backdoor detection, tracr, rasp

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: Meta-models are neural networks that take parameters of other neural nets as input; we use them to uncover properties of the input nets.

Abstract:

Mechanistic interpretability aims to open the black box of neural networks. Previous work has demonstrated that the mechanisms implemented by small neural networks can be fully reverse-engineered. Since these efforts rely on human labor that does not scale to models with billions of parameters, there is growing interest in automating interpretability methods. We propose to use \emph{meta-models}, neural networks that take another network's parameters as input, to scale interpretability efforts.
To this end, we present a scalable meta-model architecture and successfully apply it to a variety of problems, including mapping neural network parameters to human-legible code and detecting backdoors in networks. Our results aim to provide a proof-of-concept for automating mechanistic interpretability methods.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 7884