Identifying and Handling Mislabelled Instances

By Fabrice Muhlenbach, Stéphane Lallich, Djamel Abdelkader Zighed. 2004

159 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Information Systems

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Information Systems

Citations in the corpus

Learning kernel logistic regression in the presence of class label noise (2014, background): “The seemingly straightforward approach is by means of data preprocessing where any suspect samples are removed or relabelled [7,1,29,37,31,18].”
Classification in the Presence of Label Noise: A Survey (2014, methodology): “In [209] and [214], the i th instance is characterized by its local cut edge weight statistic J i , which is the sum of the weights of edges linking the instance to its neighbors with a different label.”
Automatic Adaptation of Object Detectors to New Domains Using Self-Training (2019, background): “A subset of these estimated labels that are most likely to be correct are selected and used to re-train the baseline model, and the process continues in an incremental fashion [39, 33, 37, 24, 63].”
On the characterization of noise filters for self-training semi-supervised in nearest neighbor classification (2014, methodology): “In [27], the authors propose the addition of a statistical filter [28] to the self-training process, naming this algorithm SETRED.”
Co-training Improves Prompt-based Learning for Large Language Models (2022, methodology): “The cut statistic is a ranking heuristic that uses the view geometry more than the model conﬁdence approach (Muhlenbach et al., 2004; Zhang & Zhou, 2011).”
Generalized Out-of-Distribution Detection: A Survey (2021, unknown): “Another set of methods uses the relationship among data points and constructs a neighborhood graph [342], [343] or its variants [344].”
Auror: defending against poisoning attacks in collaborative deep learning systems (2016, methodology): “Muhlenbach et al. ﬁlter samples that do not have same class as their neighbors, which needs access to raw training dataset [30].”
A brief introduction to weakly supervised learning (2018, background): “For example, a data editing approach [56] constructs a relative neighborhood graph where each node corresponds to a training example, and an edge connecting two nodes with different labels is called a cut edge.”