Analysis of Deep Neural Networks with Extended Data Jacobian Matrix

38 citations recorded by the source corpus.

How researchers see this paper

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

On the challenges of learning with inference networks on sparse, high-dimensional data (2017, methodology): “Our use of the spectra of the Jacobian matrix of log p ( x | z ) to inspect learned models is inspired by Wang et al. (2016).”
Detecting Memorization in ReLU Networks (2018, background): “The processing of a single input x by a ReLU network is equivalent to sampling a sub-network that is linear with respect to the sample (Wang et al., 2016).”
Orthogonal Deep Neural Networks (2019, methodology): “[55] propose Extended Data Jacobian Matrix (EDJM) as a network analyzing tool, and study how the 1354 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL.”
Bias Also Matters: Bias Attribution for Deep Neural Network Explanation (2019, background): “As long as the activation function is piecewise linear, the DNN is a piecewise linear function and is equivalent to a linear model at and near each input point x (Montufar et al., 2014; Wang et al., 2016).”
Tensor Switching Networks (2016, background): “Finally, conditional linearity of deep ReLU networks is also used by [14], mainly to analyze their performance.”
Analysis of Invariance and Robustness via Invertibility of ReLU-Networks (2018, background): “Additionally, each of these matrices A P can be written via a chain of weight matrix multiplications incorporating the effect of ReLU , see [Wang et al., 2016].”