Systematic Bias
A bias in data whose expected value is non-zero. It does not cancel out as sample size increases and requires targeted cleaning or correction. It is fundamentally different from random noise.
Timeline (last 90 days)
Random error can be averaged out by increasing sample size, but systematic bias is amplified with sample size, causing attack success rates to be structurally overestimated or underestimated.
The expectation of systematic bias is nonzero, and no amount of data can eliminate it through averaging; if the annotation team has a consistent misunderstanding of a certain type of sample, the network will learn that error as a true pattern.
In systematic bias, data leakage and duplication, extreme long-tail, and safety-sensitive scenarios, data cleaning remains irreplaceable, because the network cannot correct structural errors on its own.
All Facts (3)
Random error can be averaged out by increasing sample size, but systematic bias is amplified with sample size, causing attack success rates to be structurally overestimated or underestimated.
50%UnverifiedThe expectation of systematic bias is nonzero, and no amount of data can eliminate it through averaging; if the annotation team has a consistent misunderstanding of a certain type of sample, the network will learn that error as a true pattern.
50%UnverifiedIn systematic bias, data leakage and duplication, extreme long-tail, and safety-sensitive scenarios, data cleaning remains irreplaceable, because the network cannot correct structural errors on its own.
50%