A Causal View on Robustness of Neural Networks
arXiv:2005.01095
Abstract
We present a causal view on the robustness of neural networks against input manipulations, which applies not only to traditional classification tasks but also to general measurement data. Based on this view, we design a deep causal manipulation augmented model (deep CAMA) which explicitly models possible manipulations on certain causes leading to changes in the observed effect. We further develop data augmentation and test-time fine-tuning methods to improve deep CAMA's robustness. When compared with discriminative deep neural networks, our proposed model shows superior robustness against unseen manipulations. As a by-product, our model achieves disentangled representation which separates the representation of manipulations from those of other latent causes.
arXiv commentNeurIPS 2020
References in corpus (19)
- The Effectiveness of Data Augmentation in Image Classification using Deep Learning
- Stochastic Backpropagation and Approximate Inference in Deep Generative Models
- Ensemble Adversarial Training: Attacks and Defenses
- A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks
- Disentangling by Factorising
- Technical Report on the CleverHans v2.1.0 Adversarial Examples Library
- Detecting Adversarial Samples from Artifacts
- Efficient Neural Network Robustness Certification with General Activation Functions
- Invariant Risk Minimization
- Towards the first adversarially robust neural network model on MNIST
- On Learning Invariant Representation for Domain Adaptation
- Adversarial vulnerability for any classifier
- On Causal and Anticausal Learning
- Disentangled Sequential Autoencoder
- DIVA: Domain Invariant Variational Autoencoders
- Anchor regression: heterogeneous data meets causality
- Are Generative Classifiers More Robust to Adversarial Attacks?
- Generalization in anti-causal learning
- Neuropathic Pain Diagnosis Simulator for Causal Discovery Algorithm Evaluation
Cited by in corpus (7)
- Adversarial Machine Learning: Bayesian Perspectives
- Out-of-distribution Prediction with Invariant Risk Minimization: The Limitation and An Effective Fix
- Proactive Pseudo-Intervention: Causally Informed Contrastive Learning For Interpretable Vision Models
- Embracing the Disharmony in Medical Imaging: A Simple and Effective Framework for Domain Adaptation
- Enhancing Model Robustness and Fairness with Causality: A Regularization Approach
- Causally Constrained Data Synthesis for Private Data Release
- Decoder-free Robustness Disentanglement without (Additional) Supervision