Skip to search formSkip to main contentSkip to account menu

Progressive Identification of True Labels for Partial-Label Learning

A novel estimator of the classification risk, theoretically analyze the classifier-consistency, and establish an estimation error bound are proposed, and a progressive identification algorithm for approximately minimizing the proposed risk estimator is proposed.
View PDF on arXiv (opens in a new tab)

Geometry-aware Instance-reweighted Adversarial Training

This paper finds even over-parameterized deep networks may still have insufficient model capacity, because adversarial training has an overwhelming smoothing effect, and argues adversarial data should have unequal importance: geometrically speaking, a natural data point closer to/farther from the class boundary is less/more robust, and the corresponding adversary data point should be assigned with larger/smaller weight.
View PDF on arXiv (opens in a new tab)

A Diffusion Theory For Deep Learning Dynamics: Stochastic Gradient Descent Exponentially Favors Flat Minima

This work develops a density diffusion theory (DDT) to reveal how minima selection quantitatively depends on the minima sharpness and the hyperparameters, and is the first to theoretically and empirically prove that, benefited from the Hessian-dependent covariance of stochastic gradient noise, SGD favors flat minima exponentially more than sharp minima.
View Paper (opens in a new tab)

Do We Need Zero Training Loss After Achieving Zero Training Error?

Flooding is proposed that intentionally prevents further reduction of the training loss when it reaches a reasonably small value, which is called the flooding level and is compatible with any stochastic optimizer and other regularizers.
View PDF on arXiv (opens in a new tab)

Provably Consistent Partial-Label Learning

This paper proposes the first generation model of candidate label sets, and develops two novel PLL methods that are guaranteed to be provably consistent, i.e., one is risk-consistent and the other is classifier- Consistent.
View PDF on arXiv (opens in a new tab)

Parts-dependent Label Noise: Towards Instance-dependent Label Noise

Empirical evaluations demonstrate this method is superior to the state-of-the-art approaches for learning from the instance-dependent label noise and the transition matrices for parts of the instance are learned by exploiting anchor points.
View PDF on arXiv (opens in a new tab)

Coupling-based Invertible Neural Networks Are Universal Diffeomorphism Approximators

A general theorem is proved to show the equivalence of the universality for certain diffeomorphism classes, a theoretical insight that is of interest by itself, and affirmatively resolve a previously unsolved problem: whether normalizing flow models based on affine coupling can be universal distributional approximators.
View PDF on arXiv (opens in a new tab)

Dual T: Reducing Estimation Error for Transition Matrix in Label-noise Learning

This paper introduces an intermediate class to avoid directly estimating the noisy class posterior of the transition matrix, and introduces the dual $T-estimator for estimating transition matrices, leading to better classification performances.
View PDF on arXiv (opens in a new tab)

Rethinking Importance Weighting for Deep Learning under Distribution Shift

This paper rethink IW and theoretically show it suffers from a circular dependency, and proposes an end-to-end solution dynamic IW that iterates between THE AUTHORS and WC and combines them in a seamless manner, and hence their THEY can also enjoy deep networks and stochastic optimizers indirectly.
View PDF on arXiv (opens in a new tab)

A Survey of Label-noise Representation Learning: Past, Present and Future

A formal definition of Label-Noise Representation Learning is clarified from the perspective of machine learning and the reason why noisy labels affect deep models' performance is figured out.
View PDF on arXiv (opens in a new tab)
...
By clicking accept or continuing to use the site, you agree to the terms outlined in our Privacy Policy (opens in a new tab), Terms of Service (opens in a new tab), and Dataset License (opens in a new tab)