Negative Control
A control method originating in experimental science, in which a control group known not to produce the expected effect is set up to verify whether the detection system itself has failed.
Timeline (last 90 days)
In offline reinforcement learning, unmeasured confounding makes the behavior policy correlated with outcomes, so directly using logged data for policy evaluation or optimization produces bias, requiring auxiliary structures such as instrumental variables, negative controls or proxy variables to achieve causal identification.
The paper combines transferability theory with instrumental variables, negative controls and proxy variables to build a unified framework for identifying causal quantities in the target domain.
Negative controls originate from experimental science and are used to confirm that the detection system itself has not failed.
The positive control is borrowed from experimental science and means testing a known necessarily positive control sample before testing the sample to prove the detection process works; the negative control confirms that records that should not be returned do not appear; neither carries information about the time dimension.
Negative controls require deliberately including inputs on which the system should fail or score low; if the metric passes them, the metric is broken rather than the system being good.
All Facts (5)
In offline reinforcement learning, unmeasured confounding makes the behavior policy correlated with outcomes, so directly using logged data for policy evaluation or optimization produces bias, requiring auxiliary structures such as instrumental variables, negative controls or proxy variables to achieve causal identification.
50%UnverifiedThe paper combines transferability theory with instrumental variables, negative controls and proxy variables to build a unified framework for identifying causal quantities in the target domain.
50%UnverifiedNegative controls originate from experimental science and are used to confirm that the detection system itself has not failed.
50%UnverifiedThe positive control is borrowed from experimental science and means testing a known necessarily positive control sample before testing the sample to prove the detection process works; the negative control confirms that records that should not be returned do not appear; neither carries information about the time dimension.
50%UnverifiedNegative controls require deliberately including inputs on which the system should fail or score low; if the metric passes them, the metric is broken rather than the system being good.
50%