This Springer chapter cited once contains several tortured phrases that make some passages hard to parse. These typically result from an attempt to avoid plagiarism detection using a paraphrasing software. So far, the following have been spotted:
- convolutional neural system instead of the established ‘convolutional neural network (CNN)’
- profound neural system instead of the established ‘deep neural network’
- yield layer AND neural instead of the established ‘output layer’
Can the authors explain why they departed from the established phrases?
How come these incorrect wordings survived proofreading by the coauthors, editors, referees, copy editors, and typesetters?
Found by the Problematic Paper Screener.
The source text appears to be Malov D., Letenkov M. (2020) Synthetic Data Generation Approach for Face Recognition System. In: Ronzhin A., Shishlakov V. (eds) Proceedings of 14th International Conference on Electromechanics and Robotics “Zavalishin's Readings”. Smart Innovation, Systems and Technologies, vol 154. Springer, Singapore. DOI link. A few fragments for comparison are below.
This chapter:
- It’s important to have as huge dataset as it’s conceivable to prepare an exact classifier, particularly if profound neural systems are being used. It’s expensive to compose delegate dataset physically – to snap a picture of each individual from each conceivable edge with each conceivable light condition. This is the reason why generation of a synthetic data for training a classifier, utilizing least genuine information is so urgent.
Likely source:
- It is necessary to have a large dataset as it is possible to train an accurate classifier, especially if deep neural networks are being used. It is expensive to organize representative dataset manually to take a photograph of every person from every possible angle with every possible light condition. This is why the generation of synthetic data for training a classifier, using minimum real data is so urgent.
This chapter:
- Considering the class of assignments related with computer vision, and, specifically, with pattern recognition, when working with classes of objects not beforehand represented in wide access, there emerges the issue of deciding the right for building a calculation for creating datasets for preparing a model.
Likely source:
- Considering the class of tasks associated with computer vision, and, in particular, with pattern recognition, when working with classes of objects not previously represented in wide access, there arises the problem of determining the most correct strategy for constructing an algorithm for generating datasets for training a model.
This chapter:
- Tragically, this methodology has various disadvantages, which comprise in an elevated level of equipment necessities for the machine which register the training generations. Utilizing such free programming packages as Blender (the free-permit appropriated three-dimensional editorial manager) and AliceVision, which actualizes the execution of the specialized piece of photogrammetry forms, it gets conceivable to accomplish the accompanying results of time and equipment subordinate qualities.
Likely source:
- Unfortunately, this approach has a number of drawbacks, which consist in a high level of hardware requirements for the machine which compute the generation process. Using such free software packages as Blender (the free-license distributed three-dimensional editor) and AliceVision, which implements the execution of the technical part of photogrammetry processes, it becomes possible to achieve the following results of time and hardware-dependent characteristics.