Notebook

Saliency Papers¶

Sam Greydanus | 2017 | MIT License

Summary¶

Year	Author	Name	Keywords
2014	Simonyan et al.	Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps	saliency (orig. paper)
2014	Zeiler, Fergus	Visualizing and Understanding Convolutional Networks	saliency, ablation study
2015	Springenberg et al.	Striving for Simplicity: the All Convolutional Net	guided backpropagation
2015	Bach et al.	On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation	layer-wise relevance propagation (LRP)
2016	Karpathy et al.	Visualizing and Understanding Recurrent Networks	visualization, LSTMs, RNNs, error analysis
2016	Ziyu et al.	Dueling Network Architectures for Deep Reinforcement Learning	saliency, atari, video
2016	Zahavy	Graying the black box: Understanding DQNs	atari, saliency, markov models
2016	Zhang et al.	Top-down Neural Attention by Excitation Backprop	excitation backprop
2017	Shrikumar et al.	Learning Important Features Through Propagating Activation Differences	DeepLIFT, decision boundaries
2017	Montavon et al.	Explaining NonLinear Classification Decisions with Deep Taylor Decomposition	deep taylor decomposition, LRP

Visualizing and Understaning Atari Agents ¶

Sam Greydanus

Curated abstract.

Math.

Pros.

Cons.

Visualizing and Understanding Recurrent Networks ¶

Karpathy, Johnson, and Fei-Fei

Curated abstract. Performance and limitations of RNNs are poorly understood. Using character-level language models as a testbed, authors analyze the internal representations, predictions and error types. Results include: 1) existence of interpretable cells that track long-range dependencies such as line lengths, quotes and brackets. 2) comparative analysis with finite horizon n-gram models shows LSTM improvements are due to longer time horizon. 3) analysis of the remaining errors with a pie chart error breakdown.

Math.

Pros.

Cons.

Explaining NonLinear Classification Decisions with Deep Taylor Decomposition ¶

Montavon, Bach, Binder, Samek, Müller

Math.

Words.

Pros.

Cons.

Striving for Simplicity: the All Convolutional Net ¶

Springenberg, Dosovitskiy, Brox, Riedmiller

Math.

Words.

Pros.

Cons.

Graying the black box: Understanding DQNs ¶

Zahavy, Baram, Mannor

Math.

Words.

Pros.

Cons.

In [ ]:

Saliency Papers¶

Summary¶

Visualizing and Understaning Atari Agents¶

Visualizing and Understanding Recurrent Networks¶

Explaining NonLinear Classification Decisions with Deep Taylor Decomposition¶

Striving for Simplicity: the All Convolutional Net¶

Graying the black box: Understanding DQNs¶

Visualizing and Understaning Atari Agents ¶

Visualizing and Understanding Recurrent Networks ¶

Explaining NonLinear Classification Decisions with Deep Taylor Decomposition ¶

Striving for Simplicity: the All Convolutional Net ¶

Graying the black box: Understanding DQNs ¶