Temporal Masking
Temporal masking is an evaluation technique that truncates time-series data to a specific point in time to prevent a model from accessing future information beyond the decision point. This simulates genuine prospective decision-making scenarios and reduces the influence of hindsight bias on model evaluation.
Timeline (last 90 days)
Temporal masking can reduce hindsight bias without compromising accuracy.
Temporal masking refers to physically removing all clinical events after the temporal cutoff from the sequence when inputting case information, ensuring the model can only see information at and before the decision point.
The study found that even when prompts explicitly instruct the model to answer from a prospective perspective, the model is still influenced by outcome information already present in the context, and physical truncation is the effective means of blocking hindsight bias.
All Facts (3)
Temporal masking can reduce hindsight bias without compromising accuracy.
50%UnverifiedTemporal masking refers to physically removing all clinical events after the temporal cutoff from the sequence when inputting case information, ensuring the model can only see information at and before the decision point.
50%UnverifiedThe study found that even when prompts explicitly instruct the model to answer from a prospective perspective, the model is still influenced by outcome information already present in the context, and physical truncation is the effective means of blocking hindsight bias.
50%