Meta Soft Label Generation for Noisy Labels

Date

2021-01-01

Author

Algan, Gorkem
Ulusoy, İlkay

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

143
views

0
downloads

The existence of noisy labels in the dataset causes significant performance degradation for deep neural networks (DNNs). To address this problem, we propose a Meta Soft Label Generation algorithm called MSLG, which can jointly generate soft labels using meta-learning techniques and learn DNN parameters in an end-to-end fashion. Our approach adapts the meta-learning paradigm to estimate optimal label distribution by checking gradient directions on both noisy training data and noise-free meta-data. In order to iteratively update soft labels, meta-gradient descent step is performed on estimated labels, which would minimize the loss of noise-free meta samples. In each iteration, the base classifier is trained on estimated meta labels. MSLG is model-agnostic and can be added on top of any existing model at hand with ease. We performed extensive experiments on CIFAR10, Clothing1M and Food101N datasets. Results show that our approach outperforms other state-of-the-art methods by a large margin. Our code is available at https://github.com/gorkemalgan/MSLG_noisy_label.

URI

https://hdl.handle.net/11511/92017

DOI

https://doi.org/10.1109/icpr48806.2021.9412490

Conference Name

25th International Conference on Pattern Recognition (ICPR)

Collections

Department of Electrical and Electronics Engineering, Conference / Seminar

Suggestions

OpenMETU
Core

MetaLabelNet: Learning to Generate Soft-Labels From Noisy-Labels Algan, Gorkem; Ulusoy, İlkay (2022-01-01) Real-world datasets commonly have noisy labels, which negatively affects the performance of deep neural networks (DNNs). In order to address this problem, we propose a label noise robust learning algorithm, in which the base classifier is trained on soft-labels that are produced according to a meta-objective. In each iteration, before conventional training, the meta-training loop updates soft-labels so that resulting gradients updates on the base classifier would yield minimum loss on meta-data. Soft-labels...
Non-Linear Filtering based on Observations from Student's t Processes Saha, Saikat; Orguner, Umut; Gustafsson, Fredrik (2012-03-10) We consider measurements from possibly zero-mean stochastic processes in a nonlinear filtering framework. This is a challenging problem, since it is only the second order properties of the measurements that bear information about the unknown state vector. The covariance function of the measurements can have both spatial and temporal correlation that depend on the state. Recently, a solution to this problem was presented for the case of Gaussian processes. We here extend the theory to Student's t processes. ...
Online state estimation for discrete nonlinear dynamic systems with nonlinear noise and interference Demirbaş, Kerim (2015-01-01) This paper presents a real-time recursive state filtering and prediction scheme (PR) for discrete nonlinear dynamic systems with nonlinear noise and random interference, such as undesired random jamming or clutter. The PR is based upon discrete noise approximation, state quantization, and a suboptimal implementation of multiple composite hypothesis testing. The PR outperforms both the sampling importance resampling (SIR) particle filter and auxiliary sampling importance resampling (ASIR) particle filter; wh...
Hierarchical multitasking control of discrete event systems: Computation of projections and maximal permissiveness Schmidt, Klaus Verner; Cury, José E.r. (null; 2010-12-01) This paper extends previous results on the hierarchical and decentralized control of multitasking discrete event systems (MTDES). Colored observers, a generalization of the observer property, together with local control consistency, allow to derive sufficient conditions for synthesizing modular and hierarchical control that are both strongly nonblocking (SNB) and maximally permissive. A polynomial procedure to verify if a projection fulfills the above properties is proposed and in the case they fail for a g...
Out-of-Sample Generalizations for Supervised Manifold Learning for Classification Vural, Elif (2016-03-01) Supervised manifold learning methods for data classification map high-dimensional data samples to a lower dimensional domain in a structure-preserving way while increasing the separation between different classes. Most manifold learning methods compute the embedding only of the initially available data; however, the generalization of the embedding to novel points, i.e., the out-of-sample extension problem, becomes especially important in classification applications. In this paper, we propose a semi-supervis...

Citation Formats

G. Algan and İ. Ulusoy, “Meta Soft Label Generation for Noisy Labels,” presented at the 25th International Conference on Pattern Recognition (ICPR), ELECTR NETWORK, 2021, Accessed: 00, 2021. [Online]. Available: https://hdl.handle.net/11511/92017.