Training Invariant Support Vector Machines using Selective Sampling

By Gaëlle Loosli, Stéphane Canu, L. Eon Bottou. 2005

How researchers see this paper

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

In Optimization & Training

Citations in the corpus

Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks (2015, methodology): “No data augmentation was applied to the images.”
Generalized Byzantine-tolerant SGD (2018, methodology): “Note that all the experiments of CNN on CIFAR10 show similar results with the experiments of MLP on MNIST.”
Deep Sets (2017, background): “Image MNIST8m dataset (Loosli et al., 2007) contains 8 million instances of 2828 grey-scale stamps of digits in { 0 , ..., 9 } .”
Scalable Kernel Methods via Doubly Stochastic Gradients (2014, neutral): “5M 54 {−1, 1} no (5) MNIST 8M [22] K-logistic 8M 1568 {0, .”
Adversarial Diversity and Hard Positive Generation (2016, methodology): “Research by Loosli et al. [14] took perturbation-enhanced training to a new level and developed an open source tool called InfiMNIST [14] that produces MNIST examples by applying small affine transformations and noise to the original images.”
Transformation Pursuit for Image Classification (2014, background): “An exception is the digit recognition task, where computing elastic deformation fields can be performed in an elegant and computationally efficient manner [16].”
Learning from Conditional Distributions via Dual Kernel Embeddings (2016, methodology): “The proposed algorithm bares some similarities to virtual sample techniques [24, 18] in the sense that they both create examples with prior knowledge to incorporate invariance.”
Visualizing Higher-Layer Features of a Deep Network (2009, methodology): “We used two datasets: the first is an extended version of the MNIST digit classification dataset, by Loosli et al. (2007), in which elastic deformations of digits are generated stochastically.”
A Kernel Perspective for Regularizing Deep Neural Networks (2018, unknown): “This section provides more details on the deformation stability penalties mentioned in Section 2.2, and the practical versions we use in our experiments on the Inﬁnite MNIST dataset (Loosli et al., 2007).”
One Network to Solve Them All — Solving Linear Inverse Problems Using Deep Projection Models (2017, methodology): “We evaluate the proposed framework on the MNIST dataset [30], MS-Celeb-1M dataset [23], ImageNet dataset [41], and LabelMe dataset [42], whose descriptions are listed in Table 1.”
Online Incremental Feature Learning with Denoising Autoencoders (2012, background): “…(SVMs) [5] have been one of the most popular discriminative methods for large-scale learning which includes selective sampling to reduce the number of support vectors [4, 24] and incremental support vectors to learn an SVM with a small number of examples in the early phase of training [7, 23].”
Regularization for Deep Learning: A Taxonomy (2017, background): “Error-maximizing rigid image transformations (Loosli et al., 2007; Fawzi et al., 2016) p(θ) Input Adaptive stochastic & constrained, respectively Training”
Learning from Conditional Distributions via Dual Embeddings (2016, methodology): “The proposed algorithm bears some similarities to virtual sample techniques (Niyogi et al., 1998; Loosli et al., 2007) in the sense that they both create examples with prior knowledge to incorporate invariance.”