Salient Deconvolutional Networks

By Aravindh Mahendran, Andrea Vedaldi. 2016

132 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Generative Models

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Generative Models

Citations in the corpus

Sanity Checks for Saliency Maps (2018, background): “Along similar lines, Mahendran and Vedaldi [28] also showed that some backpropagation-based saliency methods lack neuron discriminativity.”
Interpretable Explanations of Black Boxes by Meaningful Perturbation (2017, methodology): “Consider the following variant of rule From left to right: original image with ground truth bounding box, learned mask sub-tracted from 1 (our method), gradient-based saliency [15], guided backprop [16, 8], contrastive excitation backprop [20], Grad-CAM [14], and occlusion [19].”
Toward Trustworthy AI Development: Mechanisms for Supporting Verifiable Claims (2020, background): “, the reliability of existing methods is an active area of research [85] [86] [87] [88] [89] [90] [91] [92] [93].”
Opportunities and obstacles for deep learning in biology and medicine (2017, methodology): “Researchers are still actively identifying weaknesses [339], and new methods are being developed to address them [169,340,341].”
Grad-CAM++: Generalized Gradient-Based Visual Explanations for Deep Convolutional Networks (2017, neutral): “These visualization techniques were compared in [9].”
Action Genome: Actions As Compositions of Spatio-Temporal Scene Graphs (2019, background): “Amongst numerous techniques, saliency prediction has emerged as a key mechanism to interpret machine learning models [54, 65, 67].”
A Theoretical Explanation for Perplexing Behaviors of Backpropagation-based Visualizations (2018, neutral): “Our experiments have conﬁrmed previous observations (Mahendran & Vedaldi, 2016; Selvaraju et al., 2016; Samek et al., 2017) that saliency map is indeed very sensitive to the change of class labels, while GBP and DeconvNet, though their visualization results are much cleaner than saliency map,…”
Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization (2016, unknown): “These approaches are compared in [34].”
Debugging Tests for Model Explanations (2020, result): “The ﬁnding that methods that modify backprop combined with positive aggregation are invariant to higher layer parameters corroborates the recent work of Sixt et al. [60] along with previous evidence by Nie et al. [44] and Mahendran and Vedaldi [38].”
An Introduction to Deep Visual Explanation (2017, methodology): “thod is proposed in [19], the authors assign a weight to each feature with respect to class y. The drawback of this approach is that, it is computationally expensive. Other algorithms are proposed in [7,14,17, 20]. 2 Another interesting type of explanation is based on network activation; two popular”
Explainable AI: Interpreting, Explaining and Visualizing Deep Learning (2019, methodology): “[16] suggest several methods, such as establishing a quantitative approximate measure (e.”
Using KL-divergence to focus Deep Visual Explanation (2017, background): “Recently, many methods have been proposed to address this trade-off of interpret-ability [8-10].”
Opportunities and Challenges in Explainable Artificial Intelligence (XAI): A Survey (2020, methodology): “Saliency Maps [67], [71], [73], [96] Visualizing gradients, neural activation of individual layers using DeConv nets, guided backpropagation, etc.”
Ablation-CAM: Visual Explanations for Deep Convolutional Network via Gradient-free Localization (2020, unknown): “These works are compared in [11].”
Understanding Deep Networks via Extremal Perturbations and Smooth Masks (2019, background): “Several papers have shown that some (but not all) backpropagation-based methods produce the same saliency map regardless of the output neuron being analysed [14], or even regardless of network parameters [2].”
Using Machine Learning Safely in Automotive Software: An Assessment and Adaption of Software Process Requirements in ISO 26262 (2018, background): “, [87]) and newer model agnostic approaches [85, 86].”