Wieland Brendel

Affiliations

University of Tübingen

papers

2,592

total citations

papers (17)

On Adaptive Attacks to Adversarial Example Defenses

NEURIPS 2020arXiv

917

citations

Improving robustness against common corruptions by covariate shift adaptation

NEURIPS 2020arXiv

593

citations

Self-Supervised Learning with Data Augmentations Provably Isolates Content from Style

NEURIPS 2021arXiv

363

citations

Does CLIP’s generalization performance mainly stem from high train-test similarity?

ICLR 2024arXiv

citations

How Well do Feature Visualizations Support Causal Understanding of CNN Activations?

NEURIPS 2021arXiv

citations

Don't trust your eyes: on the (un)reliability of feature visualizations

ICML 2024arXiv

citations

Embrace the Gap: VAEs Perform Independent Mechanism Analysis

NEURIPS 2022arXiv

citations

Scale Alone Does not Improve Mechanistic Interpretability in Vision Models

NEURIPS 2023arXiv

citations

Position: Understanding LLMs Requires More Than Statistical Generalization

ICML 2024arXiv

citations

Increasing Confidence in Adversarial Robustness Evaluations

NEURIPS 2022arXiv

citations

Identifiable Exchangeable Mechanisms for Causal Structure and Representation Learning

ICLR 2025arXiv

citations

LLMs on the Line: Data Determines Loss-to-Loss Scaling Laws

ICML 2025arXiv

citations

VGGSounder: Audio-Visual Evaluations for Foundation Models

ICCV 2025arXiv

citations

Wieland Brendel

Affiliations

papers (17)

On Adaptive Attacks to Adversarial Example Defenses

Improving robustness against common corruptions by covariate shift adaptation

Self-Supervised Learning with Data Augmentations Provably Isolates Content from Style

Partial success in closing the gap between human and machine vision

Fast Minimum-norm Adversarial Attacks through Adaptive Norm Constraints

A Simple Way to Make Neural Networks Robust Against Diverse Image Corruptions

Compositional Generalization from First Principles

Does CLIP’s generalization performance mainly stem from high train-test similarity?

How Well do Feature Visualizations Support Causal Understanding of CNN Activations?

Don't trust your eyes: on the (un)reliability of feature visualizations

Embrace the Gap: VAEs Perform Independent Mechanism Analysis

Scale Alone Does not Improve Mechanistic Interpretability in Vision Models

Position: Understanding LLMs Requires More Than Statistical Generalization

Increasing Confidence in Adversarial Robustness Evaluations

Identifiable Exchangeable Mechanisms for Causal Structure and Representation Learning

LLMs on the Line: Data Determines Loss-to-Loss Scaling Laws

VGGSounder: Audio-Visual Evaluations for Foundation Models

papers (17)

On Adaptive Attacks to Adversarial Example Defenses

Improving robustness against common corruptions by covariate shift adaptation

Self-Supervised Learning with Data Augmentations Provably Isolates Content from Style

Partial success in closing the gap between human and machine vision

Fast Minimum-norm Adversarial Attacks through Adaptive Norm Constraints

A Simple Way to Make Neural Networks Robust Against Diverse Image Corruptions

Compositional Generalization from First Principles

Does CLIP’s generalization performance mainly stem from high train-test similarity?

How Well do Feature Visualizations Support Causal Understanding of CNN Activations?

Don't trust your eyes: on the (un)reliability of feature visualizations

Embrace the Gap: VAEs Perform Independent Mechanism Analysis

Scale Alone Does not Improve Mechanistic Interpretability in Vision Models

Position: Understanding LLMs Requires More Than Statistical Generalization

Increasing Confidence in Adversarial Robustness Evaluations

Identifiable Exchangeable Mechanisms for Causal Structure and Representation Learning

LLMs on the Line: Data Determines Loss-to-Loss Scaling Laws

VGGSounder: Audio-Visual Evaluations for Foundation Models