Marginalized Denoising Auto-encoders for Nonlinear Representations

By Minmin Chen, Kilian Q. Weinberger, Fei Sha, Yoshua Bengio. 2014

144 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.

Citations in the corpus

Towards Deep Neural Network Architectures Robust to Adversarial Examples (2014, methodology): “…ideas from denoising autoencoder (DAE), contractive autoencoder (CAE), and most recently marginalized denoising au-toencoder (mDAE) provide strong framework for training neural networks that are robust against adversarial noises Rifai et al. (2011b); Alain and Bengio (2012); Chen et al. (2014).”
Dropout Training of Matrix Factorization and Autoencoder for Link Prediction in Sparse Graphs (2015, methodology): “Previously, [21, 6] used the second order Taylor expansion to explain the eﬀect of feature noising in generalized linear models and AE, respectively.”
Towards Bayesian Deep Learning: A Survey (2016, methodology): “2, if we assume that both the clean input Xc and the corrupted input X0 are observed, similar to [2], [3], [9], [37], we can define the following generative process of generalized Bayesian SDAE:”
Collaborative Denoising Auto-Encoders for Top-N Recommender Systems (2016, background): “Also, the idea of marginalized DAE [3] might be able to speed up the training and improve the performance.”
Pattern Recognition (2015, unknown): “It should be noted that we freeze the parameters of optical-ﬂow sub-network at the beginning of training.”
Spectral Bisection Tree Guided Deep Adaptive Exemplar Autoencoder for Unsupervised Domain Adaptation (2016, methodology): “Second, although there is an explicit formulation for ∇2xinL , we further simplify it by dropping certain terms and removing off-diagonal elements, as suggested in (LeCun et al. 2012; Chen et al. 2014).”
Compressing Neural Networks with the Hashing Trick (2015, methodology): “We can encourage this effect in the hidden layers with sparsity inducing transition functions, e.g. rectiﬁed linear units (ReLU) (Glorot et al., 2011a) or through specialized regularization (Chen et al., 2014; Boureau et al., 2008).”
Deep Sequential Context Networks for Action Prediction (2017, background): “This scheme can be considered as a maskout/drop-out regularization [8, 3].”
A Survey on Bayesian Deep Learning (2016, methodology): “2, if we assume that both the clean input Xc and the corrupted input X0 are observed, similar to [4, 5, 13, 72], we can define the following generative process of the probabilistic SDAE: (1) For each layer l of the SDAE network, (a) For each column n of the weight matrixWl , drawWl,∗n ∼ N(0, λ−1 w I”