Training deep neural-networks using a noise adaptation layer

By Jacob Goldberger, Ehud Ben-Reuven. 2016

831 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

Using Pre-Training Can Improve Model Robustness and Uncertainty (2019, background): “Also note that most prior works (Goldberger & Ben-Reuven, 2017; Ma et al., 2018; Han et al., 2018) only experiment with small-scale images since label corruption demonstrations can require training hundreds of models (Hendrycks et al., 2018).”
Co-teaching: Robust training of deep neural networks with extremely noisy labels (2018, methodology): “1 Introduction Learning from noisy labels can date back to three decades ago [1], and still keeps vibrant in recent years [13, 31].”
Generalized Cross Entropy Loss for Training Deep Neural Networks with Noisy Labels (2018, methodology): “Several methods have been proposed to estimate it [11, 13, 31, 16].”
MentorNet: Learning Data-Driven Curriculum for Very Deep Neural Networks on Corrupted Labels (2017, background): “To address the issue, Goldberger et al. (2017) proposed to add an additional softmax layer end-to-end with the base model.”
Model-Agnostic Approaches To Handling Noisy Labels When Training Sound Event Classifiers (2019, background): “, the probability of each true label being flipped into another [12, 13].”
Probabilistic harmonization and annotation of single‐cell transcriptomics data with deep generative models (2019, unknown): “Our approach is similar to previous work in robustness to noisy labels (Goldberger & Ben-Reuven, 2017) and hierarchical multi-labels flavors of classification problems (Wehrmann et al , 2018).”
Enabling a Single Deep Learning Model for Accurate Gland Instance Segmentation: A Shape-Aware Adversarial Learning Framework (2020, methodology): “Goldberger and Ben-Reuven [25] proposed to estimate the correct labels by using the EM algorithm.”
NLNL: Negative Learning for Noisy Labels (2019, background): “[28, 14, 5] modeled the noise transition matrix by adding an additional layer.”
Identifying Mislabeled Data using the Area Under the Margin Ranking (2020, background): “ied [e.g. 16, 17, 42, 67]. Here we note several prior works, but please refer to [1] for a complete review. Within the context of deep learning, researchers have proposed novel training architectures [18], model-based curriculum learning schemes [26, 50], label correction [61], and robust loss funct”
From Synthetic to Real: Unsupervised Domain Adaptation for Animal Pose Estimation (2021, background): “To reduce the negative effect of noisy labels, some approaches focus on training noise robust models by designing robust losses [12, 34, 44] or by correcting the loss with a transition matrix [27, 13, 37].”
SELFIE: Refurbishing Unclean Samples for Robust Deep Learning (2019, methodology): “A typical method is using “loss correction” that corrects the loss of training samples based on the estimated noise transition matrix (Zhang et al., 2017; Goldberger & Ben-Reuven, 2017; Patrini et al., 2017; Chang et al., 2017).”
Meta-Weight-Net: Learning an Explicit Mapping For Sample Weighting (2019, methodology): “For instance, practically collected training samples always contain corrupted labels [10, 11, 12, 13, 14, 15, 16, 17].”
Understanding and Utilizing Deep Neural Networks Trained with Noisy Labels (2019, methodology): “Several methods focus on estimating the noise transition matrix and correcting the objective function accordingly, e.g., forward or backward correction (Patrini et al., 2017), S-model (Goldberger & Ben-Reuven, 2017).”
Learning from Rules Generalizing Labeled Exemplars (2020, unknown): “Another way of relating clean and noisy labels is via an instance-independent confusion matrix learned jointly with the classiﬁer (Khetan et al., 2018; Goldberger & Ben-Reuven, 2016; Han et al., 2018b;a).”