Interesting-DL-Papers

List of DL papers on Interpretability, Learning from Limited Data and Information Bottleneck

Visual Interpretation & Adversarial Attacks

[1] Paying more attention to attention: Improving the performance of convolutional neural networks via attention transfer

[2] Interpretable and Pedagogical Examples

[3] Interpretable Explanations of Black Boxes by Meaningful Perturbation

[4] Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization

[5] Grad-CAM++: Generalized Gradient-based Visual Explanations for Deep Convolutional Networks

[6] Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examples

Information Bottleneck

[1] Compressing Neural Networks using the Variational Information Bottleneck

[2] One Big Net For Everything

Model Level Changes

[1] Minimal gated unit for recurrent neural networks

[2] Gated Feedback Recurrent Neural Networks

[3] MinimalRNN: Toward More Interpretable and Trainable Recurrent Neural Networks

[4] A recurrent neural network without chaos

Limited Data