Red Teaming Deep Neural Networks with Feature Synthesis Tools

TL;DR

This paper introduces a benchmark for interpretability tools to detect trojans in neural networks by implanting known trojans and assessing whether tools can identify them, revealing current limitations and proposing new methods.

Contribution

It proposes trojan discovery as a new evaluation benchmark, assesses existing interpretability tools, and introduces improved feature-synthesis methods for bug detection.

Findings

State-of-the-art tools often fail to detect implanted trojans.

Benchmark reveals significant challenges in current interpretability methods.

New feature-synthesis variants improve trojan detection performance.

Abstract

Interpretable AI tools are often motivated by the goal of understanding model behavior in out-of-distribution (OOD) contexts. Despite the attention this area of study receives, there are comparatively few cases where these tools have identified previously unknown bugs in models. We argue that this is due, in part, to a common feature of many interpretability methods: they analyze model behavior by using a particular dataset. This only allows for the study of the model in the context of features that the user can sample in advance. To address this, a growing body of research involves interpreting models using \emph{feature synthesis} methods that do not depend on a dataset. In this paper, we benchmark the usefulness of interpretability tools on debugging tasks. Our key insight is that we can implant human-interpretable trojans into models and then evaluate these tools based on whether they can help humans discover them. This is analogous to finding OOD bugs, except the ground truth is known, allowing us to know when an interpretation is correct. We make four contributions. (1) We propose trojan discovery as an evaluation task for interpretability tools and introduce a benchmark with 12 trojans of 3 different types. (2) We demonstrate the difficulty of this benchmark with a preliminary evaluation of 16 state-of-the-art feature attribution/saliency tools. Even under ideal conditions, given direct access to data with the trojan trigger, these methods still often fail to identify bugs. (3) We evaluate 7 feature-synthesis methods on our benchmark. (4) We introduce and evaluate 2 new variants of the best-performing method from the previous evaluation. A website for this paper and its code is at https://benchmarking-interpretability.csail.mit.edu/