semi-supervised noisy label augmentation

[dmarx]

• http://ruder.io/semi-supervised/
• Training Deep Neural Networks on Noisy Labels with Bootstrapping
• Training Convolutional Networks with Noisy Labels
• Learning from noisy labels with dropout regularization
• Learning from noisy labels with deep neural networks
• Dimensionality driven learning with noisy labels
• Toward Robustness against Label Noise in Training Deep Discriminative Neural Networks
• Learning with Noisy Labels
• Iterative learning with open-set noisy labels - siamese network

PU Learning: positive + unlabeled data

• https://www.cs.uic.edu/~liub/NSF/PSC-IIS-0307239.html

PU LEarning: cosine-rocchio PU algorithm, Li et al 2010 EMNLP

• Use cosine similarity and KNN (rochio classification) to filter likely false negatives from negative class to construct a "reliable negatives" dataset. Train SVM to further filter data. rinse and repeat

Perplexity based PU learning (stephen wan)

• use perplexity to find |pos_class| most dissimilar unlabeled instances to positive class
• train classifier to filter

[dmarx]

relevance models - ~2001/2002

LM ranking, top N -> another model to extract salient terms -> add to query, repeat
use KL divergence instead of perplexity