Like arrows on a target: clustered but off-centre, or centred on average but scattered.
Direct answer
Bias is error from a model being too simple, so it misses patterns (underfitting). Variance is error from a model being too sensitive to training data, so it does not generalise (overfitting). The bias-variance trade-off is finding a model complex enough to learn the pattern but not so complex that it memorises noise.
Comparison
| High bias | High variance | |
|---|---|---|
| Model | too simple | too complex |
| Training error | high | low |
| Test error | high | high |
| Fix | more complex model, better features | more data, simplify, regularise |
How to diagnose
Compare training and test errors. Both high suggests bias. Low training and high test suggests variance.
Where Teyro fits
Teyro's Machine Learning for Beginners course turns this into short lessons you can play in a few minutes a day. It has 200 lessons, and you can start free.
The bottom line
Aim for the sweet spot where test error is lowest.
