Language: 漢語
08-01, 10:40–11:10 (Asia/Taipei), TR409-1
30
您是否知悉並同意如採遠端形式分享,需提供預錄影片(您需同意大會才能接受您的稿件) – yes Target audience –對於加速運算或深度學習以及開源框架有興趣的人
Difficulty –入門
講者所屬的公司或組織名稱 –none
講者所屬社群 –none
Translate Title –Performance optimaztion for inference of deep learning
slido url – hackmd url –https://hackmd.io/@coscup/ryH3GaPCu/%2F%40coscup%2Fry0qGTvCO
other info –none
Abstract –深度學習的研究在現今已非常火熱且應用領域十分廣泛。其中,如何將深度學習網路部屬在邊緣端進行運算 (edge computing),如手機,是深度學習應用於產品的重要關鍵。
在這段分享將會介紹如何在手機平台上加速深度學習網路模型的推理運算。
- 介紹適用於手機上的開源推理框架,如 TensorFlow Lite、NCNN、TNN 與 MACE 。
- 介紹硬體 (如GPU 與 DSP) 加速運算。
- 介紹如何使用 TensorFlow Lite 在手機上部屬深度網路並使用硬體進行加速。
The research of deep learning is very popular today and has a wide range of applications. Among them, how to apply deep neural networks on edge computing devices, such as mobile phones, is an important key to applying deep neural networks to products.
This sharing will introduce how to accelerate the inference of deep learning models on the mobile platform.
- Introduce open source inference frameworks suitable for mobile phones, such as TensorFlow Lite, NCNN, TNN and MACE.
- Introduce accelerated computing with hardware devices, such as GPU and DSP.
- Introduce how to use TensorFlow Lite to deploy the deep neural network on mobile phone and accelerate with hardware devices.