|
Title:
|
SUPPORT FOR COMMUNICATION WITH DEAF
AND DUMB PATIENTS VIA FEW-SHOT MACHINE
LEARNING |
|
Author(s):
|
Grigorii Shovkoplias, Mark Tkachenko, Arip Asadulaev, Olga Alekseeva, Natalia Dobrenko,
Daniil Kazantsev, Alexandra Vatian, Anatoly Shalyto and Natalia Gusarova |
|
ISBN:
|
978-989-8704-30-6 |
|
Editors:
|
Piet Kommers and Mário Macedo |
|
Year:
|
2021 |
|
Edition:
|
Single |
|
Keywords:
|
Sign Language, Machine Learning, Marker Method, Neural Networks |
|
Type:
|
Short |
|
First Page:
|
216 |
|
Last Page:
|
220 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
Improving healthcare quality and patient safety for patients with disabilities is one of the most important goals of e-Health.
A large percentage of these patients are persons with disabling hearing loss, i.e. partially or completely deaf and dumb. In
our work, we discuss the opportunity of fast Sign Language processing having only a small number of examples. We
investigated the possibility of classifying datasets with an extremely small number of samples of electromyograms of deaf
and dumb gestures using few-shots learning methods. Several such methods have been considered - Matching Networks,
Model-Agnostic Meta-Learning, and Prototypical Networks. The developed methodology makes it possible to train
electromyogram classifiers using an extremely small amount of data for other deaf and dumb sign languages. To do this, it
is enough to collect a small dataset for re- or additional training of these models for the classification of another language,
which is easy to accomplish in practice by means of a small session of recording the gestures of several
deaf-mute people - speakers of a particular sign language. In our study, we selected three main few-shot learning approaches
and compared them on small sign language dataset. Based on our results, one can choose the best models and their
modifications to adapt to the practice task. |
|
|
|
|
|
|
