Jie Zhang’s IJCAI-ECAI 2018 Paper is Accepted (Oral), Congratulation!

Jie Zhang, Xiaolong Wang, Dawei Li, Yalin Wang

IJCAI-ECAI-2018 (Oral)

We introduce a novel Recurrent neural networks (RNNs) model compression approach DirNet based on an optimized fast dictionary learning algorithm, which 1) dynamically mines the dictionary atoms of the projection dictionary matrix within layer to adjust the compression rate 2) adaptively changes the sparsity of sparse codes cross the hierarchical layers. Experimental results on language model and an ASR model trained with a 1000h speech dataset demonstrate that our method reduces the size of original model by eight times with real-time model inference and negligible accuracy loss.