ConvNets and ImageNet Beyond Accuracy: Understanding Mistakes and Uncovering Biases
arXiv:1711.11443
Abstract
ConvNets and Imagenet have driven the recent success of deep learning for image classification. However, the marked slowdown in performance improvement combined with the lack of robustness of neural networks to adversarial examples and their tendency to exhibit undesirable biases question the reliability of these methods. This work investigates these questions from the perspective of the end-user by using human subject studies and explanations. The contribution of this study is threefold. We first experimentally demonstrate that the accuracy and robustness of ConvNets measured on Imagenet are vastly underestimated. Next, we show that explanations can mitigate the impact of misclassified adversarial examples from the perspective of the end-user. We finally introduce a novel tool for uncovering the undesirable biases learned by a model. These contributions also show that explanations are a valuable tool both for improving our understanding of ConvNets' predictions and for designing more reliable models.
arXiv commentECCV 2018 camera-ready
References in corpus (14)
- Intriguing properties of neural networks
- mixup: Beyond Empirical Risk Minimization
- Methods for Interpreting and Understanding Deep Neural Networks
- Convolutional Sequence to Sequence Learning
- Ensemble Adversarial Training: Attacks and Defenses
- Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
- Human pose estimation via Convolutional Part Heatmap Regression
- Countering Adversarial Images using Input Transformations
- Understanding Adversarial Training: Increasing Local Stability of Neural Nets through Robust Optimization
- Parseval Networks: Improving Robustness to Adversarial Examples
- DeeperCut: A Deeper, Stronger, and Faster Multi-Person Pose Estimation Model
- DeepFool: a simple and accurate method to fool deep neural networks
- Towards Interpretable Deep Neural Networks by Leveraging Adversarial Examples
- Residual Convolutional CTC Networks for Automatic Speech Recognition
Cited by in corpus (12)
- Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)
- Deep Clustering for Unsupervised Learning of Visual Features
- Deep k-Nearest Neighbors: Towards Confident, Interpretable and Robust Deep Learning
- Women also Snowboard: Overcoming Bias in Captioning Models
- Deceiving End-to-End Deep Learning Malware Detectors using Adversarial Examples
- Does Object Recognition Work for Everyone?
- What Do Compressed Deep Neural Networks Forget?
- Characterising Bias in Compressed Models
- Balanced Datasets Are Not Enough: Estimating and Mitigating Gender Bias in Deep Image Representations
- Distribution Density, Tails, and Outliers in Machine Learning: Metrics and Applications
- xGEMs: Generating Examplars to Explain Black-Box Models
- Breaking Batch Normalization for better explainability of Deep Neural Networks through Layer-wise Relevance Propagation