ISyNet: Convolutional Neural Networks design for AI accelerator

Letunovskiy, A.; Korviakov, V.; Polovnikov, V.; Kargapoltseva, A.; Mazurenko, I.; Xiong, Y.

Авторы: Letunovskiy Alexey, Korviakov Vladimir, Polovnikov Vladimir, Kargapoltseva Anastasiia, Mazurenko Ivan, Xiong Yepan
Журнал: arXiv.org [cs.CV]
Год издания: 2021
Номер статьи: arXiv:2109.01932 [cs.CV]
Аннотация: In recent years Deep Learning reached significant results in many practical problems, such as computer vision, natural language processing, speech recognition and many others. For many years the main goal of the research was to improve the quality of models, even if the complexity was impractically high. However, for the production solutions, which often require real-time work, the latency of the model plays a very important role. Current state-of-the-art architectures are found with neural architecture search (NAS) taking model complexity into account. However, designing of the search space suitable for specific hardware is still a challenging task. To address this problem we propose a measure of hardware efficiency of neural architecture search space - matrix efficiency measure (MEM); a search space comprising of hardware-efficient operations; a latency-aware scaling method; and ISyNet - a set of architectures designed to be fast on the specialized neural processing unit (NPU) hardware and accurate at the same time. We show the advantage of the designed architectures for the NPU devices on ImageNet and the generalization ability for the downstream classification and detection tasks.
Добавил в систему: Мазуренко Иван Леонидович

	ИСТИНА	Войти в систему Регистрация
	ИПМех РАН
	Главная Поиск Статистика О проекте Помощь

ИСТИНА

ИПМех РАН

ISyNet: Convolutional Neural Networks design for AI acceleratorстатья

Прикрепленные файлы