Safer Classification by Synthesis

By William Wang, Angelina Wang, Aviv Tamar, Xi Chen, Pieter Abbeel. 2017

41 citations recorded by the source corpus.

The discriminative approach to classification using deep neural networks has become the de-facto standard in various fields. Complementing recent reservations about safety against adversarial examples, we show that conventional discriminative methods can easily be fooled to provide incorrect labels with very high confidence to out of distribution examples. We posit that a generative approach is the natural remedy for this problem, and propose a method for classification using generative models. At training time, we learn a generative model for each class, while at test time, given an example to classify, we query each generator for its most similar generation, and select the class corresponding to the most similar one. Our approach is general and can be used with expressive models such as GANs and VAEs. At test time, our method accurately "knows when it does not know," and provides resilience to out of distribution examples while maintaining competitive performance for standard examples.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Generative Models, 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 Generative Models

In Safety, Robustness & Interpretability

Citations in the corpus

Generalization in anti-causal learning (2018, background): “Experiments: Wang et al. (2017); Samangouei et al. (2018); Song et al. (2017); Ilyas et al. (2017) increase robustness against adversarial examples, by learning a generative (causal) model and synthesizing images until a match for the input is found.”
Towards neural networks that provably know when they don't know (2019, methodology): “Second, modification of the classifier training by integrating generative models like a VAE or GAN in order to discriminate out-distribution from in-distribution data (Lee et al., 2018a; Wang et al., 2018; Lee et al., 2018b) or approaches Code at https://github.”
Why ReLU Networks Yield High-Confidence Predictions Far Away From the Training Data and How to Mitigate the Problem (2018, methodology): “The existing approaches basically either use adjustment techniques of the softmax outputs [9, 22] by temperature rescaling [12] or they use a generative model like a VAE or GAN to model boundary inputs of the true distribution [20, 33] in order to discriminate in-distribution from out-of-distributio”
PAQ: 65 Million Probably-Asked Questions and What You Can Do With Them (2021, methodology): “Figure 3 shows a risk-coverage plot (Wang et al., 2018) for RePAQ and FiD, where we use FiD’s answer log probability for its answer confidence.”
Learning Plannable Representations with Causal InfoGAN (2018, methodology): “Following [45], we performed a search over the latent space to find the best latent state mapping s∗(o): s∗(o) = arg mins mins′,z ‖o−G(s, s′, z)‖(2).”
Does Your Model Know the Digit 6 Is Not a Cat? A Less Biased Evaluation of "Outlier" Detectors (2018, background): “[30] present an alternative “safe” classification paradigm based on generative adversarial networks (GANs) [31].”
Training Discriminative Models to Evaluate Generative Ones (2018, background): “They can also be used for data augmentation [20], for safety against adversarial example [27], or to produce labeled data [24] in order to improve the training of discriminative models.”
Exemplar-Supported Generative Reproduction for Class Incremental Learning (2018, methodology): “Besides generating high-quality images [8, 10, 22, 24], GANs have been used in image classification [34], zeroshot learning [3, 35] etc.”