Long Short-Term Memory Over Recursive Structures

Xiaodan Zhu, Parinaz Sobihani, Hongyu Guo
Proceedings of the 32nd International Conference on Machine Learning, PMLR 37:1604-1612, 2015.

Abstract

The chain-structured long short-term memory (LSTM) has showed to be effective in a wide range of problems such as speech recognition and machine translation. In this paper, we propose to extend it to tree structures, in which a memory cell can reflect the history memories of multiple child cells or multiple descendant cells in a recursive process. We call the model S-LSTM, which provides a principled way of considering long-distance interaction over hierarchies, e.g., language or image parse structures. We leverage the models for semantic composition to understand the meaning of text, a fundamental problem in natural language understanding, and show that it outperforms a state-of-the-art recursive model by replacing its composition layers with the S-LSTM memory blocks. We also show that utilizing the given structures is helpful in achieving a performance better than that without considering the structures.

Cite this Paper

BibTeX
@InProceedings{pmlr-v37-zhub15, title = {Long Short-Term Memory Over Recursive Structures}, author = {Zhu, Xiaodan and Sobihani, Parinaz and Guo, Hongyu}, booktitle = {Proceedings of the 32nd International Conference on Machine Learning}, pages = {1604--1612}, year = {2015}, editor = {Bach, Francis and Blei, David}, volume = {37}, series = {Proceedings of Machine Learning Research}, address = {Lille, France}, month = {07--09 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v37/zhub15.pdf}, url = {https://proceedings.mlr.press/v37/zhub15.html}, abstract = {The chain-structured long short-term memory (LSTM) has showed to be effective in a wide range of problems such as speech recognition and machine translation. In this paper, we propose to extend it to tree structures, in which a memory cell can reflect the history memories of multiple child cells or multiple descendant cells in a recursive process. We call the model S-LSTM, which provides a principled way of considering long-distance interaction over hierarchies, e.g., language or image parse structures. We leverage the models for semantic composition to understand the meaning of text, a fundamental problem in natural language understanding, and show that it outperforms a state-of-the-art recursive model by replacing its composition layers with the S-LSTM memory blocks. We also show that utilizing the given structures is helpful in achieving a performance better than that without considering the structures.} }
Endnote
%0 Conference Paper %T Long Short-Term Memory Over Recursive Structures %A Xiaodan Zhu %A Parinaz Sobihani %A Hongyu Guo %B Proceedings of the 32nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2015 %E Francis Bach %E David Blei %F pmlr-v37-zhub15 %I PMLR %P 1604--1612 %U https://proceedings.mlr.press/v37/zhub15.html %V 37 %X The chain-structured long short-term memory (LSTM) has showed to be effective in a wide range of problems such as speech recognition and machine translation. In this paper, we propose to extend it to tree structures, in which a memory cell can reflect the history memories of multiple child cells or multiple descendant cells in a recursive process. We call the model S-LSTM, which provides a principled way of considering long-distance interaction over hierarchies, e.g., language or image parse structures. We leverage the models for semantic composition to understand the meaning of text, a fundamental problem in natural language understanding, and show that it outperforms a state-of-the-art recursive model by replacing its composition layers with the S-LSTM memory blocks. We also show that utilizing the given structures is helpful in achieving a performance better than that without considering the structures.
RIS
TY - CPAPER TI - Long Short-Term Memory Over Recursive Structures AU - Xiaodan Zhu AU - Parinaz Sobihani AU - Hongyu Guo BT - Proceedings of the 32nd International Conference on Machine Learning DA - 2015/06/01 ED - Francis Bach ED - David Blei ID - pmlr-v37-zhub15 PB - PMLR DP - Proceedings of Machine Learning Research VL - 37 SP - 1604 EP - 1612 L1 - http://proceedings.mlr.press/v37/zhub15.pdf UR - https://proceedings.mlr.press/v37/zhub15.html AB - The chain-structured long short-term memory (LSTM) has showed to be effective in a wide range of problems such as speech recognition and machine translation. In this paper, we propose to extend it to tree structures, in which a memory cell can reflect the history memories of multiple child cells or multiple descendant cells in a recursive process. We call the model S-LSTM, which provides a principled way of considering long-distance interaction over hierarchies, e.g., language or image parse structures. We leverage the models for semantic composition to understand the meaning of text, a fundamental problem in natural language understanding, and show that it outperforms a state-of-the-art recursive model by replacing its composition layers with the S-LSTM memory blocks. We also show that utilizing the given structures is helpful in achieving a performance better than that without considering the structures. ER -
APA
Zhu, X., Sobihani, P. & Guo, H.. (2015). Long Short-Term Memory Over Recursive Structures. Proceedings of the 32nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 37:1604-1612 Available from https://proceedings.mlr.press/v37/zhub15.html.

Related Material