• Publications
  • Influence
Semi-Orthogonal Low-Rank Matrix Factorization for Deep Neural Networks
We introduce a factored form of TDNNs (TDNN-F) which is structurally the same as a TDNN whose layers have been compressed via SVD, but is trained from a random start with one of the two factors of each matrix constrained to be semi-orthogonal. Expand
  • 131
  • 17
  • PDF
Impacts of urbanization and industrialization on energy consumption/CO2 emissions: Does the level of development matter?
Urbanization and industrialization have significant impacts on energy consumption and CO2 emissions, but their relationship varies at different stages of economic development. Taking cognizance ofExpand
  • 238
  • 16
Quantifying atmospheric nitrogen deposition through a nationwide monitoring network across China
Abstract. A Nationwide Nitrogen Deposition Monitoring Network (NNDMN) containing 43 monitoring sites was established in China to measure gaseous NH3, NO2, and HNO3 and particulate NH4+ and NO3− inExpand
  • 192
  • 9
  • PDF
A Time-Restricted Self-Attention Layer for ASR
We use a time-restricted self-attention layer in TDNN and TDNN-LSTM structures and show that it can improve the performance in both setups as well as speed up decoding. Expand
  • 59
  • 7
  • PDF
Neural Network Language Modeling with Letter-Based Features and Importance Sampling
In this paper we describe an extension of the Kaldi software toolkit to support neural-based language modeling, intended for use in automatic speech recognition (ASR) and related tasks. Expand
  • 36
  • 5
  • PDF
Impact of energy conservation policies on the green productivity in China's manufacturing sector: Evidence from a three-stage DEA model
This study introduces an improved Malmquist–Luenberger productivity index to measure the green productivity growth of China’s manufacturing sector during the 11th Five-Year Period (2006–2010). AExpand
  • 172
  • 4
Recurrent Neural Network Language Model Adaptation for Conversational Speech Recognition
We propose two adaptation models for recurrent neural network language models (RNNLMs) to capture topic effects and longdistance triggers for conversational automatic speech recognition (ASR). Expand
  • 24
  • 3
  • PDF
Measuring green productivity growth of Chinese industrial sectors during 1998–2011
This study proposes an improved method for measuring green productivity growth in order to overcome the “discriminating power problem” and “technical regress” associated with the conventional dataExpand
  • 48
  • 3
A Pruned Rnnlm Lattice-Rescoring Algorithm for Automatic Speech Recognition
We propose a pruned lattice-rescoring algorithm for ASR, improving the n-gram approximation method. Expand
  • 50
  • 2
  • PDF
The nonlinear impacts of industrial structure on China's energy intensity
Adjusting industrial structural is crucial for Chinese government's effort to reduce energy intensity. This paper, based on the fact that the Chinese industrial structure's two V-pattern evolutionsExpand
  • 86
  • 2