更多数据提升模型性能
本文探讨了为何在低偏差学习模型中需要更多数据以改善模型表现，并介绍了如何判断何时需要增加数据量的方法。此外，还提供了多种获取额外数据的途径，如人工合成数据、手动收集标注等。
Get More Data
1. Why We Need More Data?
In many situations (low bias learning model), more data usually means better performance of the model.
2. When We Need More Data?
Usually, we should plot the learning curve by using part of the training data (1/10). If we have low bias curve, then we are safely increase the training data to get better machine learning model.
3. How to Get More Data?
- Artificial data synthesis (e.g., rotation, crop, change background, etc)
- Collect and label the data manually
- Hire other company to label (e.g., Amazon Mechanical Turk)
Usually to make the original data 10 times larger won’t take so much effort, but it will make the performance of the model much better.
扫一扫
1万+
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
