Timezone: »
Back-propagation (BP) is the foundation for successfully training deep neural networks. However, BP sometimes has difficulties in propagating a learning signal deep enough effectively, e.g., the vanishing gradient phenomenon. Meanwhile, BP often works well when combining with ``designing tricks'' like orthogonal initialization, batch normalization and skip connection. There is no clear understanding on what is essential to the efficiency of BP. In this paper, we take one step towards clarifying this problem. We view BP as a solution of back-matching propagation which minimizes a sequence of back-matching losses each corresponding to one block of the network. We study the Hessian of the local back-matching loss (local Hessian) and connect it to the efficiency of BP. It turns out that those designing tricks facilitate BP by improving the spectrum of local Hessian. In addition, we can utilize the local Hessian to balance the training pace of each block and design new training algorithms. Based on a scalar approximation of local Hessian, we propose a scale-amended SGD algorithm. We apply it to train neural networks with batch normalization, and achieve favorable results over vanilla SGD. This corroborates the importance of local Hessian from another side.
Author Information
Huishuai Zhang (Microsoft Research Asia)
Wei Chen (Microsoft Research)
Tie-Yan Liu (Microsoft Research Asia)
Tie-Yan Liu is an assistant managing director of Microsoft Research Asia, leading the machine learning research area. He is very well known for his pioneer work on learning to rank and computational advertising, and his recent research interests include deep learning, reinforcement learning, and distributed machine learning. Many of his technologies have been transferred to Microsoft’s products and online services (such as Bing, Microsoft Advertising, Windows, Xbox, and Azure), and open-sourced through Microsoft Cognitive Toolkit (CNTK), Microsoft Distributed Machine Learning Toolkit (DMTK), and Microsoft Graph Engine. He has also been actively contributing to academic communities. He is an adjunct/honorary professor at Carnegie Mellon University (CMU), University of Nottingham, and several other universities in China. He has published 200+ papers in refereed conferences and journals, with over 17000 citations. He has won quite a few awards, including the best student paper award at SIGIR (2008), the most cited paper award at Journal of Visual Communications and Image Representation (2004-2006), the research break-through award (2012) and research-team-of-the-year award (2017) at Microsoft Research, and Top-10 Springer Computer Science books by Chinese authors (2015), and the most cited Chinese researcher by Elsevier (2017). He has been invited to serve as general chair, program committee chair, local chair, or area chair for a dozen of top conferences including SIGIR, WWW, KDD, ICML, NIPS, IJCAI, AAAI, ACL, ICTIR, as well as associate editor of ACM Transactions on Information Systems, ACM Transactions on the Web, and Neurocomputing. Tie-Yan Liu is a fellow of the IEEE, and a distinguished member of the ACM.
More from the Same Authors
-
2021 : AI X Science »
Tie-Yan Liu -
2021 Poster: On the Generative Utility of Cyclic Conditionals »
Chang Liu · Haoyue Tang · Tao Qin · Jintao Wang · Tie-Yan Liu -
2021 Poster: Curriculum Offline Imitating Learning »
Minghuan Liu · Hanye Zhao · Zhengyu Yang · Jian Shen · Weinan Zhang · Li Zhao · Tie-Yan Liu -
2021 Poster: Speech-T: Transducer for Text to Speech and Beyond »
Jiawei Chen · Xu Tan · Yichong Leng · Jin Xu · Guihua Wen · Tao Qin · Tie-Yan Liu -
2021 Poster: Stylized Dialogue Generation with Multi-Pass Dual Learning »
Jinpeng Li · Yingce Xia · Rui Yan · Hongda Sun · Dongyan Zhao · Tie-Yan Liu -
2021 Poster: Distributional Reinforcement Learning for Multi-Dimensional Reward Functions »
Pushi Zhang · Xiaoyu Chen · Li Zhao · Wei Xiong · Tao Qin · Tie-Yan Liu -
2021 Poster: Optimizing Information-theoretical Generalization Bound via Anisotropic Noise of SGLD »
Bohan Wang · Huishuai Zhang · Jieyu Zhang · Qi Meng · Wei Chen · Tie-Yan Liu -
2021 Poster: Co-evolution Transformer for Protein Contact Prediction »
He Zhang · Fusong Ju · Jianwei Zhu · Liang He · Bin Shao · Nanning Zheng · Tie-Yan Liu -
2021 Poster: Stable, Fast and Accurate: Kernelized Attention with Relative Positional Encoding »
Shengjie Luo · Shanda Li · Tianle Cai · Di He · Dinglan Peng · Shuxin Zheng · Guolin Ke · Liwei Wang · Tie-Yan Liu -
2021 Poster: Learning Causal Semantic Representation for Out-of-Distribution Prediction »
Chang Liu · Xinwei Sun · Jindong Wang · Haoyue Tang · Tao Li · Tao Qin · Wei Chen · Tie-Yan Liu -
2021 Poster: Object-Aware Regularization for Addressing Causal Confusion in Imitation Learning »
Jongjin Park · Younggyo Seo · Chang Liu · Li Zhao · Tao Qin · Jinwoo Shin · Tie-Yan Liu -
2021 Poster: FastCorrect: Fast Error Correction with Edit Alignment for Automatic Speech Recognition »
Yichong Leng · Xu Tan · Linchen Zhu · Jin Xu · Renqian Luo · Linquan Liu · Tao Qin · Xiangyang Li · Edward Lin · Tie-Yan Liu -
2021 Poster: Do Transformers Really Perform Badly for Graph Representation? »
Chengxuan Ying · Tianle Cai · Shengjie Luo · Shuxin Zheng · Guolin Ke · Di He · Yanming Shen · Tie-Yan Liu -
2021 Poster: R-Drop: Regularized Dropout for Neural Networks »
xiaobo liang · Lijun Wu · Juntao Li · Yue Wang · Qi Meng · Tao Qin · Wei Chen · Min Zhang · Tie-Yan Liu -
2021 Poster: Recovering Latent Causal Factor for Generalization to Distributional Shifts »
Xinwei Sun · Botong Wu · Xiangyu Zheng · Chang Liu · Wei Chen · Tao Qin · Tie-Yan Liu -
2020 Poster: Semi-Supervised Neural Architecture Search »
Renqian Luo · Xu Tan · Rui Wang · Tao Qin · Enhong Chen · Tie-Yan Liu -
2020 Poster: RD$^2$: Reward Decomposition with Representation Decomposition »
Zichuan Lin · Derek Yang · Li Zhao · Tao Qin · Guangwen Yang · Tie-Yan Liu -
2020 Poster: MPNet: Masked and Permuted Pre-training for Language Understanding »
Kaitao Song · Xu Tan · Tao Qin · Jianfeng Lu · Tie-Yan Liu -
2019 : Break / Poster Session 1 »
Antonia Marcu · Yao-Yuan Yang · Pascale Gourdeau · Chen Zhu · Thodoris Lykouris · Jianfeng Chi · Mark Kozdoba · Arjun Nitin Bhagoji · Xiaoxia Wu · Jay Nandy · Michael T Smith · Bingyang Wen · Yuege Xie · Konstantinos Pitas · Suprosanna Shit · Maksym Andriushchenko · Dingli Yu · Gaël Letarte · Misha Khodak · Hussein Mozannar · Chara Podimata · James Foulds · Yizhen Wang · Huishuai Zhang · Ondrej Kuzelka · Alexander Levine · Nan Lu · Zakaria Mhammedi · Paul Viallard · Diana Cai · Lovedeep Gondara · James Lucas · Yasaman Mahdaviyeh · Aristide Baratin · Rishi Bommasani · Alessandro Barp · Andrew Ilyas · Kaiwen Wu · Jens Behrmann · Omar Rivasplata · Amir Nazemi · Aditi Raghunathan · Will Stephenson · Sahil Singla · Akhil Gupta · YooJung Choi · Yannic Kilcher · Clare Lyle · Edoardo Manino · Andrew Bennett · Zhi Xu · Niladri Chatterji · Emre Barut · Flavien Prost · Rodrigo Toro Icarte · Arno Blaas · Chulhee Yun · Sahin Lale · YiDing Jiang · Tharun Kumar Reddy Medini · Ashkan Rezaei · Alexander Meinke · Stephen Mell · Gary Kazantsev · Shivam Garg · Aradhana Sinha · Vishnu Lokhande · Geovani Rizk · Han Zhao · Aditya Kumar Akash · Jikai Hou · Ali Ghodsi · Matthias Hein · Tyler Sypherd · Yichen Yang · Anastasia Pentina · Pierre Gillot · Antoine Ledent · Guy Gur-Ari · Noah MacAulay · Tianzong Zhang -
2019 Poster: Neural Machine Translation with Soft Prototype »
Yiren Wang · Yingce Xia · Fei Tian · Fei Gao · Tao Qin · Cheng Xiang Zhai · Tie-Yan Liu -
2019 Poster: FastSpeech: Fast, Robust and Controllable Text to Speech »
Yi Ren · Yangjun Ruan · Xu Tan · Tao Qin · Sheng Zhao · Zhou Zhao · Tie-Yan Liu -
2019 Poster: Fully Parameterized Quantile Function for Distributional Reinforcement Learning »
Derek Yang · Li Zhao · Zichuan Lin · Tao Qin · Jiang Bian · Tie-Yan Liu -
2019 Poster: Distributional Reward Decomposition for Reinforcement Learning »
Zichuan Lin · Li Zhao · Derek Yang · Tao Qin · Tie-Yan Liu · Guangwen Yang -
2019 Poster: Normalization Helps Training of Quantized LSTM »
Lu Hou · Jinhua Zhu · James Kwok · Fei Gao · Tao Qin · Tie-Yan Liu -
2018 Poster: Neural Architecture Optimization »
Renqian Luo · Fei Tian · Tao Qin · Enhong Chen · Tie-Yan Liu -
2018 Poster: Learning to Teach with Dynamic Loss Functions »
Lijun Wu · Fei Tian · Yingce Xia · Yang Fan · Tao Qin · Lai Jian-Huang · Tie-Yan Liu -
2018 Poster: Layer-Wise Coordination between Encoder and Decoder for Neural Machine Translation »
Tianyu He · Xu Tan · Yingce Xia · Di He · Tao Qin · Zhibo Chen · Tie-Yan Liu -
2018 Poster: FRAGE: Frequency-Agnostic Word Representation »
Chengyue Gong · Di He · Xu Tan · Tao Qin · Liwei Wang · Tie-Yan Liu -
2017 Poster: Decoding with Value Networks for Neural Machine Translation »
Di He · Hanqing Lu · Yingce Xia · Tao Qin · Liwei Wang · Tie-Yan Liu -
2017 Poster: Finite sample analysis of the GTD Policy Evaluation Algorithms in Markov Setting »
Yue Wang · Wei Chen · Yuting Liu · Zhi-Ming Ma · Tie-Yan Liu -
2017 Poster: Deliberation Networks: Sequence Generation Beyond One-Pass Decoding »
Yingce Xia · Fei Tian · Lijun Wu · Jianxin Lin · Tao Qin · Nenghai Yu · Tie-Yan Liu -
2017 Poster: LightGBM: A Highly Efficient Gradient Boosting Decision Tree »
Guolin Ke · Qi Meng · Thomas Finley · Taifeng Wang · Wei Chen · Weidong Ma · Qiwei Ye · Tie-Yan Liu -
2016 Poster: A Communication-Efficient Parallel Algorithm for Decision Tree »
Qi Meng · Guolin Ke · Taifeng Wang · Wei Chen · Qiwei Ye · Zhi-Ming Ma · Tie-Yan Liu -
2016 Poster: Dual Learning for Machine Translation »
Di He · Yingce Xia · Tao Qin · Liwei Wang · Nenghai Yu · Tie-Yan Liu · Wei-Ying Ma -
2016 Poster: LightRNN: Memory and Computation-Efficient Recurrent Neural Networks »
Xiang Li · Tao Qin · Jian Yang · Xiaolin Hu · Tie-Yan Liu -
2013 Poster: Estimation Bias in Multi-Armed Bandit Algorithms for Search Advertising »
Min Xu · Tao Qin · Tie-Yan Liu -
2012 Poster: Statistical Consistency of Ranking Methods in A Rank-Differentiable Probability Space »
Yanyan Lan · Jiafeng Guo · Xueqi Cheng · Tie-Yan Liu -
2012 Spotlight: Statistical Consistency of Ranking Methods in A Rank-Differentiable Probability Space »
Yanyan Lan · Jiafeng Guo · Xueqi Cheng · Tie-Yan Liu -
2010 Workshop: Machine Learning in Online Advertising »
James G Shanahan · Deepak Agarwal · Tao Qin · Tie-Yan Liu -
2010 Poster: Two-Layer Generalization Analysis for Ranking Using Rademacher Average »
Wei Chen · Tie-Yan Liu · Zhi-Ming Ma -
2010 Poster: A New Probabilistic Model for Rank Aggregation »
Tao Qin · Xiubo Geng · Tie-Yan Liu -
2009 Poster: Statistical Consistency of Top-k Ranking »
fen xia · Tie-Yan Liu · Hang Li -
2009 Poster: Ranking Measures and Loss Functions in Learning to Rank »
Wei Chen · Tie-Yan Liu · Yanyan Lan · Zhi-Ming Ma · Hang Li -
2008 Poster: Global Ranking Using Continuous Conditional Random Fields »
Tao Qin · Tie-Yan Liu · Xu-Dong Zhang · De-Sheng Wang · Hang Li -
2008 Oral: Global Ranking Using Continuous Conditional Random Fields »
Tao Qin · Tie-Yan Liu · Xu-Dong Zhang · De-Sheng Wang · Hang Li