阿里few shot learning文章的个人理解
先贴结构图： 1、每次C类，每类K样本，剩下的是测试集。 2、encoder层是BiLSTM+attention，得到编码向量，假设隐含层个数是96个，句子长度是30，暂不考虑batch_size，那么每个字的隐含层就是1*192向量，整个H就是30*192，那么attention的最后输出就是一个1*192的向量（计算公式如下，权重矩阵W是30*192的矩阵）。
相关文章
- 1. Zero-shot Learning / One-shot Learning / Few-shot Learning
- 2. Zero-shot / One-shot / Few-shot Learning
- 3. Few-shot-learning
- 4. few-shot learning几篇论文
- 5. 【论文理解】Shot in The Dark: Few-Shot Learning with No Base-Class Labels
- 6. few-shot learning总结
- 7. Zero-shot Learning / One-shot Learning / Few-shot Learning/传统 Learning
- 8. few-shot learning, zero-shot learning, one-shot learning,any-shot learning, C-way K-shot，Meta-learn
- 9. 深度学习: Zero-shot Learning / One-shot Learning / Few-shot Learning
- 10. Few-shot Learning with Graph Neural Networks
- 更多相关文章...
- • PHP 文件处理 - PHP教程
- • UDP报文格式详解 - TCP/IP教程
- • 使用阿里云OSS+CDN部署前端页面与加速静态资源
- • 三篇文章了解 TiDB 技术内幕——说存储
相关标签/搜索
每日一句
-
每一个你不满意的现在，都有一个你没有努力的曾经。
欢迎关注本站公众号,获取更多信息
相关文章
- 1. Zero-shot Learning / One-shot Learning / Few-shot Learning
- 2. Zero-shot / One-shot / Few-shot Learning
- 3. Few-shot-learning
- 4. few-shot learning几篇论文
- 5. 【论文理解】Shot in The Dark: Few-Shot Learning with No Base-Class Labels
- 6. few-shot learning总结
- 7. Zero-shot Learning / One-shot Learning / Few-shot Learning/传统 Learning
- 8. few-shot learning, zero-shot learning, one-shot learning,any-shot learning, C-way K-shot，Meta-learn
- 9. 深度学习: Zero-shot Learning / One-shot Learning / Few-shot Learning
- 10. Few-shot Learning with Graph Neural Networks