Understanding Deep Architectures by Visual Summaries

Carletti, Marco; Godi, Marco; Aghaei, Maedeh; Giuliari, Francesco; Cristani, Marco

doi:10.48550/arxiv.1801.09103

Cited by 2 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Meaningful-Perturbation (MP) [22] Fong et al proposed to find a minimal Gaussian blur mask of radius b R such that when applied over the input image would produce a blurred version that has a near-zero classification score. MP is the basis for many extensions [56,40,15,57,54,9]. In this paper, we evaluate MP sensitivity to three common hyperparameters: the blur radius b R , the number of steps N iter , and the random seed, which determines the random initialization.…”

Section: Methods and Related Workmentioning

confidence: 99%

“…Meaningful-Perturbation is sensitive to the number of iterations, the Gaussian blur radius, and the random seed MP [22] is a representative of a family of methods that attempt to learn an explanation via iterative optimization [56,40,15,57,54,9]. However, in practice, optimization problems are often non-convex and thus the stopping criteria for iterative solvers are heuristically set.…”

Section: 32mentioning

confidence: 99%

See 1 more Smart Citation

SAM: The Sensitivity of Attribution Methods to Hyperparameters

Bansal

Agarwal

Nguyen

2020

2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

Attribution methods can provide powerful insights into the reasons for a classifier's decision. We argue that a key desideratum of an explanation method is its robustness to input hyperparameters which are often randomly set or empirically tuned. High sensitivity to arbitrary hyperparameter choices does not only impede reproducibility but also questions the correctness of an explanation and impairs the trust of end-users. In this paper, we provide a thorough empirical study on the sensitivity of existing attribution methods. We found an alarming trend that many methods are highly sensitive to changes in their common hyperparameters e.g. even changing a random seed can yield a different explanation! Interestingly, such sensitivity is not reflected in the average explanation accuracy scores over the dataset as commonly reported in the literature. In addition, explanations generated for robust classifiers (i.e. which are trained to be invariant to pixel-wise perturbations) are surprisingly more robust than those generated for regular classifiers.

show abstract

Section: Methods and Related Workmentioning

confidence: 99%

Section: 32mentioning

confidence: 99%

SAM: The Sensitivity of Attribution Methods to Hyperparameters

Bansal

Agarwal

Nguyen

2020

2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

show abstract

“…Extensive work has been done to explain the decisions of image classifiers and segmentation [12,35,40,27,3,11,15,12,35,40,27,3,11,15,37]. Other research directions on image classification try to find answers inside a network architecture.…”

Section: Related Workmentioning

confidence: 99%

A Tour of Convolutional Networks Guided by Linear Interpreters

Michelini

Jiang

2019

2019 IEEE/CVF International Conference on Computer Vision (ICCV)

View full text Add to dashboard Cite

Convolutional networks are large linear systems divided into layers and connected by non-linear units. These units are the "articulations" that allow the network to adapt to the input. To understand how a network manages to solve a problem we must look at the articulated decisions in entirety. If we could capture the actions of non-linear units for a particular input, we would be able to replay the whole system back and forth as if it was always linear. It would also reveal the actions of non-linearities because the resulting linear system, a Linear Interpreter, depends on the input image. We introduce a hooking layer, called a LinearScope, which allows us to run the network and the linear interpreter in parallel. Its implementation is simple, flexible and efficient. From here we can make many curious inquiries: how do these linear systems look like? When the rows and columns of the transformation matrix are images, how do they look like? What type of basis do these linear transformations rely on? The answers depend on the problems presented, through which we take a tour to some popular architectures used for classification, super-resolution (SR) and image-to-image translation (I2I). For classification we observe that popular networks use a pixel-wise vote per class strategy and heavily rely on bias parameters. For SR and I2I we find that CNNs use wavelet-type basis similar to the human visual system. For I2I we reveal copy-move and template-creation strategies to generate outputs.

show abstract

Understanding Deep Architectures by Visual Summaries

Cited by 2 publications

References 0 publications

SAM: The Sensitivity of Attribution Methods to Hyperparameters

SAM: The Sensitivity of Attribution Methods to Hyperparameters

A Tour of Convolutional Networks Guided by Linear Interpreters

Contact Info

Product

Resources

About