Retraction Note: Simplified neural network for generalized least absolute deviation

Yawei Li, Xingbao GaoNeural 2024 · DOI · OpenAlex

Introduction

In 2018, the journal Neural Computing and Applications published a study titled "Simplified neural network for generalized least absolute deviation," which aimed to optimize neural networks for robust statistics. However, following a comprehensive investigation by the publisher, Springer Nature, the article has been officially retracted. This retraction note serves as a formal invalidation of the study's findings due to significant breaches of scientific integrity.

Grounds for Retraction

The decision to retract the paper was based on several critical irregularities identified during a post-publication review. The investigation revealed three primary areas of concern:

  • Tortured Phrases: The text contained numerous instances of "non-standard phrasing." In the context of research integrity, these are often referred to as "tortured phrases"—technical terms replaced with synonyms (e.g., using "profound learning" instead of "deep learning") to bypass plagiarism detection software.
  • Citation Manipulation: A citation analysis showed that the referenced literature did not support the claims made within the paper. This pattern often suggests the presence of citation rings or the inclusion of irrelevant references to artificially inflate metric scores.
  • Image Irregularities: Forensic analysis of the figures and data visualizations provided in the original 2018 article uncovered unexplained irregularities, further undermining the reliability of the reported results.

The Investigation Process

Springer Nature’s internal investigation utilized a multi-faceted approach to evaluate the work. This included textual forensics to identify automated paraphrasing and a rigorous check of the bibliography against the core arguments of the paper. The Editor-in-Chief concluded that these cumulative issues resulted in a total loss of confidence in the study’s conclusions.

Despite the evidence presented by the publisher, the authors, Yawei Li and Xingbao Gao, formally disagreed with the retraction. This disagreement highlights the frequent tension between editorial oversight and authors when research integrity is questioned.

Broader Implications for the Field

This case serves as a cautionary tale for the machine learning community, particularly regarding the "paper mill" era of the late 2010s. The use of automated tools to generate or paraphrase technical content poses a significant threat to the reliability of the scientific record. For researchers working on robust regression and generalized least absolute deviation (GLAD), this retraction is a reminder to critically evaluate the provenance of older literature in the field.

Conclusion

The results and conclusions of the 2018 paper are no longer considered part of the reliable scientific record. While the publisher maintains a neutral stance regarding jurisdictional claims and institutional affiliations, the technical content of the paper is officially invalidated. Future research in this domain should look toward verified and peer-reviewed methodologies that adhere to standard academic integrity practices.

Summary updated

BibTeX
@inproceedings{li2024retraction,
  author = {Yawei Li and Xingbao Gao},
  title = {{Retraction Note: Simplified neural network for generalized least absolute deviation}},
  booktitle = {Neural},
  year = {2024},
  doi = {10.1007/s00521-024-10124-2},
}
Figure