|Name:||(08) An Efficient Sparse Matrix Multiplication for Deep Neural Network-Based Applications|
|Time:||Monday, June 23, 2014
05:11 pm - 05:18 pm
CCL - Congress Center Leipzig
|Presenter:||Renliang Zhao, University of Chinese Academy of Sciences|
|Abstract:||Deep Neural Network (DNN) is currently widely used in various applications, such as speech recognition, computer vision and natural language processing, etc. However, real-time recognition process has not been achieved yet as the performance of existing methods and software libraries is not as good as expected. Therefore, in this project, we propose a novel sparse matrix storage format, called BCSR & BCOO, and a thread-level scalable computing kernel for sparse matrix multiplication, called SpMM. We evaluate our proposed data structure and SpMM on a real application in DNN-based online speech recognition. The experimental results demonstrate up to 4x speedup over Intel MKL on a typical CPU-based multicore platform.
Yang Gao, Baidu Corp.; Ying Liu, Graduate University of Chinese Academy of Sciences; Renliang Zhao, University of Chinese Academy of Sciences; Steve Chiu; Idaho State University