Technical Name |
Low-power Deep Learning Accelerator / Image Semantic Segmentation Technology for Autonomous Driving |
Project Operator |
NATIONAL TSING HUA UNIVERSITY |
Summary |
1.Low-power Deep Learning Accelerator uses a small amount of computation and memory footprint to perform high-accuracy speech command recognition on the edge device. 2.The image Semantic Segmentation technology for autonomous driving can significantly reduce the memory data traffic, the required computing time and computing power. |
Scientific Breakthrough |
1.低功耗深度學習硬體加速器,整合神經網路設計與訓練、音訊處理、模型壓縮、神經網路硬體加速器等多項技術,使用少量的運算量及記憶體,在有限的硬體環境下,於終端裝置實現低功耗、高精準度的語音指令識別。 晶片架構的設計可支援多種神經網路架構,並具有可調整性和模組化設計。此技術大幅降低參數儲存所需之空間,並且可維持神經網路運算之精準度。 2.應用於自動駕駛的圖像語意分割技術,可大幅減少記憶體資料流量與所需的運算時間與運算功耗。 |
Industrial Applicability |
1.The speech command recognition can be widely used in communications, vehicles, home appliances, etc. 2.The Image Semantic Segmentation Technology can be used in autonomous driving. |