Filters








742,208 Hits in 3.8 sec

Compression of End-to-End Models

Ruoming Pang, Tara Sainath, Rohit Prabhavalkar, Suyog Gupta, Yonghui Wu, Shuyuan Zhang, Chung-Cheng Chiu
2018 Interspeech 2018  
This work explores the problem of compressing end-to-end models with the goal of satisfying device constraints without sacrificing model accuracy.  ...  End-to-end models, which directly output text given speech using a single neural network, have been shown to be competitive with conventional speech recognition models containing separate acoustic, pronunciation  ...  We therefore examine the impact of these techniques on compressing a specific end-to-end model: the attention-based encoder-decoder architecture (LAS) [9] .  ... 
doi:10.21437/interspeech.2018-1025 dblp:conf/interspeech/PangSPGWZC18 fatcat:lv3la3qv45ab3mccbfkw27bfim

Iterative Compression of End-to-End ASR Model Using AutoML

Abhinav Mehrotra, Łukasz Dudziak, Jinsu Yeo, Young-yoon Lee, Ravichander Vipperla, Mohamed S. Abdelfattah, Sourav Bhattacharya, Samin Ishtiaq, Alberto Gil C.P. Ramos, SangJeong Lee, Daehyun Kim, Nicholas D. Lane
2020 Interspeech 2020  
Past research have shown that AutoML-based Low Rank Factorization (LRF) technique, when applied to an end-to-end Encoder-Attention-Decoder style ASR model, can achieve a speedup of up to 3.7×, outperforming  ...  However, we show that current AutoML-based search techniques only work up to a certain compression level, beyond which they fail to produce compressed models with acceptable word error rates (WER).  ...  T [s j :] 5 M j ←M j 6 end 7 if i == K then 8 s * ← s 9 end 10 Retrain the model using M for initialization 11 end 12 for j ← 1 to L do 16 Retrain the model using M for initialization 13 U, Σ, V T ←  ... 
doi:10.21437/interspeech.2020-1894 dblp:conf/interspeech/MehrotraDYLVABI20 fatcat:rn7mrtqncrhmbgq7snxhnouo2q

Iterative Compression of End-to-End ASR Model using AutoML [article]

Abhinav Mehrotra, Łukasz Dudziak, Jinsu Yeo, Young-yoon Lee, Ravichander Vipperla, Mohamed S. Abdelfattah, Sourav Bhattacharya, Samin Ishtiaq, Alberto Gil C. P. Ramos, SangJeong Lee, Daehyun Kim, Nicholas D. Lane
2020 arXiv   pre-print
Past research have shown that AutoML-based Low Rank Factorization (LRF) technique, when applied to an end-to-end Encoder-Attention-Decoder style ASR model, can achieve a speedup of up to 3.7x, outperforming  ...  However, we show that current AutoML-based search techniques only work up to a certain compression level, beyond which they fail to produce compressed models with acceptable word error rates (WER).  ...  T [s j :] 5 M j ←M j 6 end 7 if i == K then 8 s * ← s 9 end 10 Retrain the model using M for initialization 11 end 12 for j ← 1 to L do 16 Retrain the model using M for initialization 13 U, Σ, V T ←  ... 
arXiv:2008.02897v1 fatcat:xlhigg6l6jgtjca72rthlrq2uy

ShrinkML: End-to-End ASR Model Compression Using Reinforcement Learning

Łukasz Dudziak, Mohamed S. Abdelfattah, Ravichander Vipperla, Stefanos Laskaridis, Nicholas D. Lane
2019 Interspeech 2019  
In this paper, we build an AutoML system that uses reinforcement learning (RL) to optimize the per-layer compression ratios when applied to a state-of-the-art attention based end-to-end ASR model composed  ...  Finally, we present accuracy results on LibriSpeech of the model compressed by our AutoML system, and we compare it to manually-compressed models.  ...  Introduction End-to-end automatic speech recognition (ASR) models have outperformed traditional ASR systems.  ... 
doi:10.21437/interspeech.2019-2811 dblp:conf/interspeech/DudziakAVLL19 fatcat:x5v6r2flhndwxmwbolvsymshai

Pilot Study to Compare the Use of End‐Tidal Carbon Dioxide–Guided and Diastolic Blood Pressure–Guided Chest Compression Delivery in a Swine Model of Neonatal Asphyxial Cardiac Arrest

Caitlin E. O'Brien, Michael Reyes, Polan T. Santos, Sophia E. Heitmiller, Ewa Kulikowicz, Sapna R. Kudchadkar, Jennifer K. Lee, Elizabeth A. Hunt, Raymond C. Koehler, Donald H. Shaffner
2018 Journal of the American Heart Association : Cardiovascular and Cerebrovascular Disease  
We compared hemodynamic parameters during cardiopulmonary resuscitation in which either end-tidal carbon dioxide ( ETCO 2) or diastolic blood pressure ( DBP ) levels were used to guide chest compression  ...  The American Heart Association recommends use of physiologic feedback when available to optimize chest compression delivery.  ...  The remaining authors have no disclosures to report.  ... 
doi:10.1161/jaha.118.009728 pmid:30371318 pmcid:PMC6404892 fatcat:4qurmun45jcczmhf6cexmol35q

End-to-End Rate-Distortion Optimization for Bi-Directional Learned Video Compression [article]

M. Akin Yilmaz, A. Murat Tekalp
2020 arXiv   pre-print
Learned video compression allows end-to-end rate-distortion optimized training of all nonlinear modules, quantization parameter and entropy model simultaneously.  ...  individually due to combinatorial nature of the end-to-end optimization problem.  ...  DVC [3] is the first end-to-end deep video compression model that jointly learns all components of the video compression framework.  ... 
arXiv:2008.05028v1 fatcat:tycf2rizbng37eqdzah73kryey

End-to-End Facial Deep Learning Feature Compression with Teacher-Student Enhancement [article]

Shurun Wang, Wenhan Yang, Shiqi Wang
2020 arXiv   pre-print
In this paper, we propose a novel end-to-end feature compression scheme by leveraging the representation and learning capability of deep neural networks, towards intelligent front-end equipped analysis  ...  We verify the effectiveness of the proposed model with the facial feature, and experimental results reveal better compression performance in terms of rate-accuracy compared with existing models.  ...  Motivated by the recent development of end-to-end image compression [19] , an end-to-end model by imposing l 1 norm as the sparsity constraint is trained for feature compression.  ... 
arXiv:2002.03627v1 fatcat:yb6vleg2ejebnhuy64ykagw32i

Learning End-to-End Lossy Image Compression: A Benchmark [article]

Yueyu Hu, Wenhan Yang, Zhan Ma, Jiaying Liu
2021 arXiv   pre-print
Despite great progress, a systematic benchmark and comprehensive analysis of end-to-end learned image compression methods are lacking.  ...  With this survey, the main challenges of image compression methods are revealed, along with opportunities to address the related issues with recent advanced learning methods.  ...  With inspiration from the technical merits, we propose a coarse-to-fine hyperprior framework for image compression, trying to address the issues of existing methods in multiresolution context modeling.  ... 
arXiv:2002.03711v4 fatcat:47d2ybnvmbhvtjp3lxqkkvxjq4

Constitutive model of scale effects in uniaxial compression for gas-saturated coal

Song Liang, Liu Weiqun, Jin Cuijun, Liang Haonan
2011 Procedia Engineering  
We discuss the effects of interfacial friction during testing and modify our model according to the testing results.  ...  Results show that, the ultimate compression strength of coal samples decreases as gas pressure and the ratio of height to diameter increase.  ...  Acknowledgements The research is supported by National Natural Science Foundation of China (no. 41074040) and "973 Program" of China (2009CB219605).  ... 
doi:10.1016/j.proeng.2011.11.2155 fatcat:ix3c33637jhhrjaooiawqbrsqu

Numerical and Experimental Study on End Effect of Waste-Soil Samples under Uniaxial Compression

Yukai Wang, Xiaoli Liu, Bo Pang, Yang Yu
2022 Geofluids  
In order to investigate the influence of end effect on the waste-soil sample strength under uniaxial compression, the influence law of end effect on the uniaxial compression strength is numerically simulated  ...  Based on the above model, the influence of end effect on the lateral displacement and stress state of waste-soil samples is simulated, and the formation mechanism of end effect under uniaxial compression  ...  To further study the formation mechanism of the end effect, a uniaxial compression model under different friction coeffi-cients is established firstly.  ... 
doi:10.1155/2022/3014164 fatcat:yarqlzejlfb6fbhbk6ydp72ql4

A COMPUTATIONAL APPROACH FOR EVALUATING HELICAL COMPRESSION SPRINGS

MacArthur L. Stewart .
2014 International Journal of Research in Engineering and Technology  
To this end, researchers have developed finite element analysis (FEA) modeling methods to simulate the design performance of helical compression springs.  ...  Specifically, commercially available FEA software was used to construct a structural model of a helical compression spring to simulate its full range of compression.  ...  Fig -5: Helical compression spring FEA model: hexahedral FEA mesh representation of the active coils, glue contact elements were used to attach the end coil surfaces to the FE mesh (left); integrated end  ... 
doi:10.15623/ijret.2014.0312029 fatcat:yanv7mdkejephha7b6fh4uh62q

BBNet: A Novel Convolutional Neural Network Structure in Edge-Cloud Collaborative Inference

Hongbo Zhou, Weiwei Zhang, Chengwei Wang, Xin Ma, Haoran Yu
2021 Sensors  
However, the in-layer data size of DNN is usually larger than the original data, so the communication time to send intermediate data to the cloud will also increase end-to-end latency.  ...  For example, when the upload bandwidth is only 20 kb/s, the end-to-end latency of BBNet is increased by 38.89× compared with the cloud-only approach.  ...  Conflicts of Interest: The authors declare no conflict of interest. Abbreviations The following abbreviations are used in this manuscript:  ... 
doi:10.3390/s21134494 pmid:34209400 fatcat:q7ieun26sjhujdei2auh73dira

End-to-end Optimized Video Compression with MV-Residual Prediction [article]

XiangJi Wu, Ziwen Zhang, Jie Feng, Lei Zhou, Junmin Wu
2020 arXiv   pre-print
We present an end-to-end trainable framework for P-frame compression in this paper.  ...  Finally, novel rate allocation and post-processing strategies are used to produce the final compressed bits, considering the bits constraint of the challenge.  ...  End-to-end Optimized Video Compression with MV-Residual Prediction Overview of the Proposed Model The proposed model is based on our CVPR 2019 CLIC framework in Low-rate compression [19] .  ... 
arXiv:2005.12945v1 fatcat:xcibnnubn5dqnm42j2juot7iri

Lightweight and Efficient End-to-End Speech Recognition Using Low-Rank Transformer [article]

Genta Indra Winata, Samuel Cahyawijaya, Zhaojiang Lin, Zihan Liu, Pascale Fung
2020 arXiv   pre-print
The LRT model outperforms those from existing works on several datasets in an end-to-end setting without using an external language model or acoustic data.  ...  Our approach reduces the number of parameters of the network by more than 50% and speeds up the inference time by around 1.35x compared to the baseline transformer model.  ...  End-to-end Speech Recognition Current end-to-end automatic speech recognition models are of two main types: (a) CTC-based models [10, 11] , and (b) Seq2Seq-based models, such as LAS [1] .  ... 
arXiv:1910.13923v3 fatcat:kj2ct2zlizfkxebhfmgc65btve

Design of Tapered Flatted End Tube for Tension and Compression Loading

2019 International Journal of Engineering and Advanced Technology  
Hollow circular tube joint either welding or bolting but welding aluminum structural alloys lead to unavoidable annealing which can reduce efficiency to the order of only 50%.  ...  The efficiency of riveted and bolted joints is usually assumed to be around 75% in both aluminum and steel construction.  ...  In the design modular of ansys work bench, Steel material of Fe250 is assigned to the exported tapered flatted end tubes.  ... 
doi:10.35940/ijeat.f8294.088619 fatcat:fpluk2g4hzd6fmnhi4ocph3j5i
« Previous Showing results 1 — 15 out of 742,208 results