𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗶𝗻 𝗛𝗶𝗴𝗵 𝗗𝗶𝗺𝗲𝗻𝘀𝗶𝗼𝗻 𝗔𝗹𝘄𝗮𝘆𝘀 𝗔𝗺𝗼𝘂𝗻𝘁𝘀 𝘁𝗼 𝗘𝘅𝘁𝗿𝗮𝗽𝗼𝗹𝗮𝘁𝗶𝗼𝗻
High dimensional data changes how machines learn.
Most models work well on training data. They fail when they meet new patterns. This happens because of high dimensions.
In high dimensions, data points sit far apart. Empty spaces exist between your data points. The model has to guess what happens in these empty spaces.
This guessing is extrapolation.
When a model extrapolates, it makes assumptions. If those assumptions are wrong, the model fails.
Key points to remember:
- High dimensions create vast empty spaces.
- Models must fill these gaps to make predictions.
- Predictions in these gaps rely on extrapolation.
- Extrapolation increases the risk of errors.
Understanding this helps you build better models. You must know when your model is guessing.
Source: https://dev.to/paperium/learning-in-high-dimension-always-amounts-to-extrapolation-1nk
Optional learning community: https://t.me/GyaanSetuAi