Low-rank bases for factorized hidden layer adaptation of DNN acoustic models

Author(s):  
Lahiru Samarakoon ◽  
Khe Chai Sim
2020 ◽  
Vol 34 (04) ◽  
pp. 6145-6152
Author(s):  
Di Wang ◽  
Feiqing Huang ◽  
Jingyu Zhao ◽  
Guodong Li ◽  
Guangjian Tian

Autoregressive networks can achieve promising performance in many sequence modeling tasks with short-range dependence. However, when handling high-dimensional inputs and outputs, the massive amount of parameters in the network leads to expensive computational cost and low learning efficiency. The problem can be alleviated slightly by introducing one more narrow hidden layer to the network, but the sample size required to achieve a certain training error is still substantial. To address this challenge, we rearrange the weight matrices of a linear autoregressive network into a tensor form, and then make use of Tucker decomposition to represent low-rank structures. This leads to a novel compact autoregressive network, called Tucker AutoRegressive (TAR) net. Interestingly, the TAR net can be applied to sequences with long-range dependence since the dimension along the sequential order is reduced. Theoretical studies show that the TAR net improves the learning efficiency, and requires much fewer samples for model training. Experiments on synthetic and real-world datasets demonstrate the promising performance of the proposed compact network.


2014 ◽  
Vol 59 (2) ◽  
pp. 509-516
Author(s):  
Andrzej Olajossy

Abstract Methane sorption capacity is of significance in the issues of coalbed methane (CBM) and depends on various parameters, including mainly, on rank of coal and the maceral content in coals. However, in some of the World coals basins the influences of those parameters on methane sorption capacity is various and sometimes complicated. Usually the rank of coal is expressed by its vitrinite reflectance Ro. Moreover, in coals for which there is a high correlation between vitrinite reflectance and volatile matter Vdaf the rank of coal may also be represented by Vdaf. The influence of the rank of coal on methane sorption capacity for Polish coals is not well understood, hence the examination in the presented paper was undertaken. For the purpose of analysis there were chosen fourteen samples of hard coal originating from the Upper Silesian Basin and Lower Silesian Basin. The scope of the sorption capacity is: 15-42 cm3/g and the scope of vitrinite reflectance: 0,6-2,2%. Majority of those coals were of low rank, high volatile matter (HV), some were of middle rank, middle volatile matter (MV) and among them there was a small number of high rank, low volatile matter (LV) coals. The analysis was conducted on the basis of available from the literature results of research of petrographic composition and methane sorption isotherms. Some of those samples were in the form (shape) of grains and others - as cut out plates of coal. The high pressure isotherms previously obtained in the cited studies were analyzed here for the purpose of establishing their sorption capacity on the basis of Langmuire equation. As a result of this paper, it turned out that for low rank, HV coals the Langmuire volume VL slightly decreases with the increase of rank, reaching its minimum for the middle rank (MV) coal and then increases with the rise of the rank (LV). From the graphic illustrations presented with respect to this relation follows the similarity to the Indian coals and partially to the Australian coals.


Author(s):  
An Wang ◽  
Donglin Chen ◽  
Shan Cheng ◽  
Xuepeng Jiao ◽  
Wenwei Chen
Keyword(s):  
Flue Gas ◽  

2021 ◽  
Author(s):  
Mathieu Le Provost ◽  
Ricardo Baptista ◽  
Youssef Marzouk ◽  
Jeff Eldredge
Keyword(s):  
Low Rank ◽  

Author(s):  
Daniel Povey ◽  
Gaofeng Cheng ◽  
Yiming Wang ◽  
Ke Li ◽  
Hainan Xu ◽  
...  

Sign in / Sign up

Export Citation Format

Share Document