Evidence Inference 2.0
A benchmark dataset constructed specifically for understanding RCT literature. It requires models to extract evidence from complete scientific papers and determine the direction of effect of an intervention on a given outcome. It is an important evaluation benchmark in the medical NLP domain.
Timeline (last 90 days)
Fiorillo v0.5 outputs probabilities for each candidate answer, addressing a three-way classification question on the Evidence Inference 2.0 benchmark: whether an intervention significantly increases, significantly decreases, or does not significantly change a given outcome measure relative to a control group.
The EI training set contains 2,657 articles, of which the researchers used only the 1,431 whose licenses explicitly permit reuse.
All Facts (2)
Fiorillo v0.5 outputs probabilities for each candidate answer, addressing a three-way classification question on the Evidence Inference 2.0 benchmark: whether an intervention significantly increases, significantly decreases, or does not significantly change a given outcome measure relative to a control group.
50%UnverifiedThe EI training set contains 2,657 articles, of which the researchers used only the 1,431 whose licenses explicitly permit reuse.
50%