Support is All You Need for Certified VAE Training

2025/04/16 by Changming Xu, Xu, Changming, Debangshu Banerjee +5
Computer Science · #Adversarial Robustness in Machine Learning #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Wireless Signal Modulation Classification

paper · pdf · doi:10.48550/arxiv.2504.11831

openalex publication_date 2025/04/16 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Variational Autoencoders (VAEs) have become increasingly popular and deployed in safety-critical applications. In such applications, we want to give certified probabilistic guarantees on performance under adversarial attacks. We propose a novel method, CIVET, for certified training of VAEs. CIVET depends on the key insight that we can bound worst-case VAE error by bounding the error on carefully chosen support sets at the latent layer. We show this point mathematically and present a novel training algorithm utilizing this insight. We show in an extensive evaluation across different datasets (in both the wireless and vision application areas), architectures, and perturbation magnitudes that our method outperforms SOTA methods achieving good standard performance with strong robustness guarantees.

Citations

• Relational DNN Verification With Cross Executional Bound Refinement
• Cross-Input Certified Training for Universal Perturbations
• Incremental Randomized Smoothing Certification
• Expressive Losses for Verified Robustness via Convex Combinations
• TAPS: Connecting Certified and Adversarial Training
• Interpreting Robustness Proofs of Deep Neural Networks
• Certified Training: Small Boxes are All You Need
• Provable Defense Against Geometric Transformations
• Robust Universal Adversarial Perturbations
• Fast Certified Robust Training with Short Warmup
• PRIMA: General and Precise Neural Network Certification via Scalable Convex Hull Approximations
• Make Sure You're Unsure: A Framework for Verifying Probabilistic Specifications
• Bayesian Inference with Certifiable Adversarial Robustness
• Fast and Complete: Enabling Complete Neural Network Verification with Rapid and Massively Parallel Incomplete Verifiers
• Adversarial Training and Provable Robustness: A Tale of Two Objectives
• Towards a Theoretical Understanding of the Robustness of Variational Autoencoders
• Probabilistic Safety for Bayesian Neural Networks
• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• Uncertainty Quantification with Statistical Guarantees in End-to-End Autonomous Driving Control
• Towards Stable and Efficient Training of Verifiably Robust Neural Networks
• An Introduction to Variational Autoencoders
• Statistical Guarantees for the Robustness of Bayesian Neural Networks
• On the Effectiveness of Interval Bound Propagation for Training Verifiably Robust Models
• Sorting out Lipschitz function approximation
• Efficient Formal Safety Analysis of Neural Networks
• Training for Faster Adversarial Robustness Verification via Inducing ReLU Stability
• Adversarial Attacks on Variational Autoencoders
• Provable defenses against adversarial examples via the convex outer adversarial polytope
• Neural Discrete Representation Learning
• Formal Verification of Piece-Wise Linear Feed-Forward Neural Networks
• Adversarial examples for generative models
• Reluplex: An Efficient SMT Solver for Verifying Deep Neural Networks
• Universal adversarial perturbations
• End to End Learning for Self-Driving Cars
• The MNIST Database of Handwritten Digit Images for Machine Learning Research [Best of the Web]
• Machine learning for medical diagnosis: history, state of the art and perspective

Related

• Re-parameterizing VAEs for stability
• Simple and Effective VAE Training with Calibrated Decoders
• Robust VAEs via Generating Process of Noise Augmented Data
• Diagnosing Vulnerability of Variational Auto-Encoders to Adversarial Attacks
• Diagnosing and Enhancing VAE Models