1.2 Data Acquisition
Li Mu
Station B: https://space.bilibili.com/1567748478/channel/collectiondetail?sid=28144
Course homepage: https://c.d2l.ai/stanford-cs329p/
1. How to obtain the data
- Is there enough data?
- If applicable: data processing
- If not: Can you find the data, the source of the data?
- Yes: discover data, merge data
- No: Can the data be generated or the method of generating the data?
- Yes: generate data
When there is not enough data, we can try to find additional other data and integrate them together as a data set.
If there is no way to obtain other additional data, we can also use data generation methods to increase data, such as: data enhancement (rotation, stretching, etc.) and use GAN to generate similar data (this will consider the issue of cost).
- How to find available datasets
- Datasets do exist, academic papers, conferences, competitions
- Find a benchmark dataset to evaluate performance: tuning: dataset diversification; deep network structure: large-scale dataset