Robustness in Multilayer Perceptrons

By Philippe Kerlirzin, F. Vallet. 1993

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

The Hidden Vulnerability of Distributed Learning in Byzantium (2018, methodology): “…with to-days’ popular tools such as the dropout algorithm, which was initially derived with a (distributed computing) fault-tolerance purpose (Kerlirzin & Vallet, 1993), then independently re-discovered as a robust regularization scheme to reduce over-ﬁtting 20 years latter (Srivastava et…”
Deep learning in neural networks: An overview (2014, background): “Compare perturbation tolerance conditions (Minai and Williams, 1994; Murray and Edwards, 1993; Neti et al., 1992; Matsuoka, 1992; Bishop, 1993; Kerlirzin and Vallet, 1993; Carter et al., 1990).”
When Neurons Fail (2017, background): “In fact, it has been experimentally observed that over-provisioning leads to robustness [24], [25], [26], [22].”
Ares: A framework for quantifying the resilience of deep neural networks (2018, background): “It has been shown that eliminating individual nodes or parameters leads to a graceful degradation in model accuracy [6].”
Perturbation response in feedforward networks (1994, background): “Minai, University of Cincinnati, Electrical and Computer Engineering Dept., Cincinnati, OH 45221; E-mail: ali.minai@uc.edu. One advantage often claimed for these networks is their robustness, that is, their ability to withstand faults, perturbations, lesions, etc. A broad, systematic study of these ”
On the Robustness of a Neural Network (2017, methodology): “…networks of 8 neurons each, using the ReLu activation function and the other category is constituted of single-layered networks of 48 neurons using the ReLu activation as well, and trained with the dropout procedure [27] that consists of randomly switching off neurons during the training phase.”
Flat Minima (1997, unknown): “…8 ij : v ij = w ij g. Flatness condition 1 is a \robustness condition" (or \fault tolerance condition", or \perturbation tolerance condition" { see, e.g., Minai & Williams, 1994; Murray & Edwards, 1993; Neti et al., 1992; Matsuoka, 1992; Bishop, 1993; Kerlirzin & Vallet, 1993; Carter et al., 1990).”
When Neurons Fail - Technical Report (2016, background): “In particular, in the case of multilayer (often called deep) neural networks, our results provide a guidance on how to distribute synaptic weights over layers, and how to tune the neuron activation function to increase robustness.”