Robustness

MoE-BiEntIRL: A novel explainable inverse reinforcement learning method enhances GNN robustness against diverse social media attacks by reconstructing attacker policies and generating more robust trai…

GEBM: a novel graph-based energy model for robust GNN uncertainty estimation.

Elliptical Attention enhances transformers by using a Mahalanobis distance metric, stretching the feature space to focus on contextually relevant information, thus improving robustness and reducing re…

ECLipsE: A novel compositional approach drastically accelerates Lipschitz constant estimation for deep neural networks, achieving speedups of thousands of times compared to the state-of-the-art while …

Training robust cooperative AI agents requires diverse and specialized training partners, but existing methods often produce overfit partners. This paper proposes novel methods using reinforcement and…

Diffusion models are certifiably robust classifiers due to their inherent O(1) Lipschitzness, a property further enhanced by generalizing to noisy data, achieving over 80% certified robustness on CIFA…

DiffHammer unveils weaknesses in diffusion-based adversarial defenses by introducing a novel attack bypassing existing evaluation limitations, leading to more robust security solutions.

Deep learning models’ robustness can be efficiently evaluated using a novel method, margin consistency, which leverages the correlation between input and logit margins for faster, accurate vulnerabili…

Boost AI model robustness against adversarial attacks by creatively mixing training sample’s frequency amplitude with distractor images, focusing model learning on phase patterns, thus enhancing accur…

This paper presents novel algorithms for linear bandits that are robust to corrupted rewards, achieving minimax optimality and optimal scaling for gap-dependent misspecification, extending to reinforc…

Constrained Adaptive Attack (CAA) significantly improves adversarial attacks on deep learning models for tabular data by combining gradient and search-based methods, achieving up to 96.1% accuracy dro…

This paper presents efficient algorithms for Bayesian persuasion under approximate best response, offering polynomial-time solutions for specific cases and a quasi-polynomial-time approximation scheme…

Accelerate DEQ certification up to 7x with Serialized Random Smoothing (SRS), achieving certified robustness on large-scale datasets without sacrificing accuracy.

CausalDiff leverages causal inference and diffusion models to create a robust AI defense against unseen adversarial attacks, significantly outperforming state-of-the-art methods.

Improved adversarially robust streaming algorithms for L_p estimation are presented, surpassing previous state-of-the-art space bounds and disproving the existence of inherent barriers.

Adversarially Robust Decision Transformer (ARDT) enhances offline RL robustness against powerful adversaries by conditioning policies on minimax returns, achieving superior worst-case performance.

Certifying neural network robustness across diverse domains, this paper introduces knowledge continuity—a novel framework ensuring model stability independent of input type, norms, and distribution.

This paper introduces a novel framework for evaluating the robustness of machine learning models when the true data distribution is only partially known. It defines a new risk measure (‘identifiable r…