Apprentissage sans labels

Notebooks

(à venir)

Lectures

Autoencoders - réduction de dimensionnalité

• Why Does Unsupervised Pre-training Help Deep Learning?

• Autoencoders

• Autoencoders, Unsupervised Learning, and Deep Architectures

• Generative Models, Adversarial Autoencoders

• Tutorial on Variational Autoencoders, Denoising Autoencoders (dA)

• Generative Adversarial Networks, NIPS 2016 Tutorial: Generative Adversarial Networks

• Adversarial Autoencoders

• Adversarial Autoencoders (with Pytorch)

• Marginalizing Stacked Linear Denoising Autoencoders

• What Regularized Auto-Encoders Learn from the Data-Generating Distribution

• Compressed sensing and single-pixel cameras

• Multi-Label Prediction via Compressed Sensing

• Inference in generative models using the Wasserstein distance, Coupling of Particle Filters

• Auto-Encoding Variational Bayes

No label, weak labels

• Unsupervised Supervised Learning I: Estimating Classification and Regression Errors without Labels

• Unsupervised Supervised Learning II: Margin-Based Classification without Labels, Unsupervised Supervised Learning II: Margin-Based Classification Without Labels (longer version)

• Large-scale Multi-label Learning with Missing Labels

• Reducing Label Complexity by Learning From Bags

• Learning from Corrupted Binary Labels via Class-Probability Estimation

• Generalized Expectation Criteria for Semi-Supervised Learning with Weakly Labeled Data

• Multitask Learning without Label Correspondences

• Training Highly Multiclass Classifiers

Online training

• Online Incremental Feature Learning with Denoising Autoencoders

• Fast Kernel Classifiers with Online and Active Learning, A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data

• Multi Kernel Learning with Online-Batch Optimization

Improving training set

• Data Programming: Creating Large Training Sets, Quickly

• Foolbox is a Python toolbox to create adversarial examples that fool neural networks.

Adversarial Examples

• The Limitations of Deep Learning in Adversarial Settings : l’article montre des limites de l’approche deep learning en construisant des exemples proches des exemples initiaux mais qui font dérailler le modèle.