Learning Perceptually-Aligned Representations via Adversarial Robustness

By Logan Engstrom, Andrew Ilyas, Shibani Santurkar, Dimitris Tsipras, Brandon Tran, Aleksander Madry. 2019

63 citations recorded by the source corpus.

Many applications of machine learning require models that are human-aligned, i.e., that make decisions based on human-meaningful information about the input. We identify the pervasive brittleness of deep networks' learned representations as a fundamental barrier to attaining this goal. We then re-cast robust optimization as a tool for enforcing human priors on the features learned by deep neural networks. The resulting robust feature representations turn out to be significantly more aligned with human perception. We leverage these representations to perform input interpolation, feature manipulation, and sensitivity mapping, without any post-processing or human intervention after model training. Our code and models for reproducing these results is available at https://git.io/robust-reps.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Representation & Self-Supervised Learning, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Representation & Self-Supervised Learning

In Safety, Robustness & Interpretability

Citations in the corpus

Shortcut learning in deep neural networks (2020, background): “f adversarial robustness tightly coupled to causality research in machine learning [128]. Adversarially robust models are somewhat more aligned with humans and show promising generalisation abilities [129,130]. While adversarial attacks test model performance on model-dependent worst-case noise, a r”
Robustness properties of Facebook's ResNeXt WSL models (2019, methodology): “Engstrom et al. (2019) recently showed that models trained with robust optimization learn fundamentally different features from models trained in the standard way (through minimization of training loss).”
Out-of-Distribution Generalization via Risk Extrapolation (REx) (2020, background): “…has become a major research topic, with many different threads of research including Bayesian deep learning (Neal, 1996; Gal, 2016), adversarial (Engstrom et al., 2019; Jacobsen et al., 2018) and non-adversarial (Hendrycks & Dietterich, 2019; Yin et al., 2019) robustness, causality (Arjovsky et…”
Better Set Representations For Relational Reasoning (2020, background): “An alternative view of the SRN is as a generative model akin to representation inversion [5] or energy-based models [14].”
Understanding invariance via feedforward inversion of discriminatively trained classifiers (2021, methodology): “Engstrom et al. (2019) argued that training a model for adversarial robustness (Madry et al., 2017) provides a useful prior to learn meaningful high-level visual representations.”