English
Overview
This paper introduced the sequence-to-sequence (Seq2Seq) framework using LSTM networks. The encoder-decoder architecture maps variable-length input sequences to variable-length output sequences, establishing the foundation for neural machine translation.
Key Contributions
- Encoder-Decoder Architecture: Two LSTMs for encoding and decoding
- Variable Length: Handles sequences of different lengths
- End-to-End Training: Single model trained on input-output pairs
- Reversed Input: Reversing source sentences improved performance
Architecture
- Encoder LSTM: Reads input sequence, produces fixed-size context vector
- Decoder LSTM: Generates output sequence from context vector
- Teacher Forcing: Uses ground truth during training
Impact
Seq2Seq became the standard for：
- Machine translation
- Text summarization
- Dialogue systems
- Any sequence transformation task
Later enhanced with attention mechanisms.
Paper Link
.3215中文
概述
本文介绍了使用LSTM网络的序列到序列（Seq2Seq）框架。编码器-解码器架构将可变长度的输入序列映射到可变长度的输出序列，为神经机器翻译奠定了基础。
主要贡献
- 编码器-解码器架构：两个LSTM分别用于编码和解码
- 可变长度：处理不同长度的序列
- 端到端训练：单一模型在输入-输出对上训练
- 反转输入：反转源句子提高了性能
架构
- 编码器LSTM：读取输入序列，产生固定大小的上下文向量
- 解码器LSTM：从上下文向量生成输出序列
- 教师强制：训练时使用真实标签
影响
Seq2Seq成为以下任务的标准：
- 机器翻译
- 文本摘要
- 对话系统
- 任何序列转换任务
后来通过注意力机制得到增强。