Главная
» Электронная библиотека » Enhancing Neural Network Performance with a Pre-Trained Low-Level Feature Detection Unit
Enhancing Neural Network Performance with a Pre-Trained Low-Level Feature Detection Unit
| Категория: | техника | |
| Авторы (ПГУ): | Гудков Павел Анатольевич [50]  , Подмарькова Екатерина Михайловна [50] | |
| Аннотация: | The paper proposes a new approach to designing neural network architectures, inspired by neurophysiological findings on the visual cortex, to enhance both training efficiency and pattern recognition accuracy. The foundation of this method lies in emulating the function of detector neurons in the brain, which are specialized for recognizing elementary visual features such as simple lines and shapes. To integrate this biological principle into artificial systems, the study introduces an additional block of pre-trained neurons into a conventional neural network model. This block is specifically dedicated to the low-level processing of fundamental image components. The results demonstrate that the incorporation of this specialized block yields the benefit: it improves the final quality of image recognition by the network. A key advantage of this architecture is that the pre-trained neurons do not consume computational resources during the main network's training phase, as their parameters are already optimized. The research investigates the dependency of training speed and recognition quality on the presence and specific composition of this lines-recognizing block. In conclusion it is indicates that approach to optimizing the architecture of neural networks has been developed and experimentally validated: applying of additional block of pretrained neurons that simulate the operation of detector neurons in the primary visual cortex leads to increased efficiency of existing neural networks. | |
| Тип: | Статья | |
| Вид: | электронная копия бумажного издания | |
| Количество частей: | 1 | |
| Год издания: | 2025 | |
| Издательство: | IEEE Publisher | |
| Где опубликовано: | Журнал | |
| Место опубликования: | APEIE 2025 | |
| Индексация: | Scopus без квартиля (тип публикации Proceeding Paper) | |
| Целевая аудитория: | Исследователь | |
| Целевое назначение: | Научное | |
| Лицензия: | Нет ограничений | |
| Ограничения: | Нет ограничений | |
| Правообладатель: | IEEE | |
| DOI: | 10.1109/APEIE66761.2025.11289270 | |
| Библиографическая ссылка: | P. Gudkov and E. Podmarkova, "Enhancing Neural Network Performance with a Pre-Trained Low-Level Feature Detection Unit," 2025 IEEE XVII International Scientific and Technical Conference on Actual Problems of Electronic Instrument Engineering (APEIE), Novosibirsk, Russian Federation, 2025. | |
| Внешняя веб-ссылка: | https://ieeexplore.ieee.org/document/11289270 | |
| Язык: | Английский | |
| Место хранения: | Вне ПГУ | |
| Дата размещения: | 22.06.2026 | |
| Владелец: | Гудков Павел Анатольевич | |
| Ссылка на ресурс: | http://elib.pnzgu.ru/library/1782100375  | |
| Формат: | Не указан |