Pre-registration
A research methodology requiring researchers to publicly declare hypotheses, methods, and evaluation criteria before experiments begin, with results reported regardless of outcome. It aims to enhance research credibility and reduce publication bias.
Timeline (last 90 days)
Preregistration requires researchers to write their hypotheses, predictions, evaluation metrics and the thresholds for declaring a hypothesis dead into a file and commit it to disk before running the data, in order to sever post-hoc narratives.
The SHA pre-registration mechanism originates from pre-registration practices in clinical trials, requiring hypotheses, metrics, and statistical analysis plans to be publicly registered before data collection, after which they must not be changed at will.
Pre-registration originates from academic research practice and refers to publicly locking hypotheses, methods, and evaluation metrics before data collection to prevent p-hacking or HARKing.
Before the new data were generated, the author publicly published in advance the precise transit times expected to occur and the criteria for pass or fail judgment, achieving pre-registration.
Pre-registration originated from the reform movement following the replication crisis in psychology, in which researchers submit their hypotheses, sample sizes, experimental procedures, and analysis methods to a public platform before data collection, forming a timestamped record.
Pre-registration has not yet become standard practice in the machine learning field, partly because experimental debugging is itself highly exploratory, making it difficult to fully predetermine hyperparameter search and result interpretation in advance.
All Facts (6)
Preregistration requires researchers to write their hypotheses, predictions, evaluation metrics and the thresholds for declaring a hypothesis dead into a file and commit it to disk before running the data, in order to sever post-hoc narratives.
50%UnverifiedThe SHA pre-registration mechanism originates from pre-registration practices in clinical trials, requiring hypotheses, metrics, and statistical analysis plans to be publicly registered before data collection, after which they must not be changed at will.
50%UnverifiedPre-registration originates from academic research practice and refers to publicly locking hypotheses, methods, and evaluation metrics before data collection to prevent p-hacking or HARKing.
50%UnverifiedBefore the new data were generated, the author publicly published in advance the precise transit times expected to occur and the criteria for pass or fail judgment, achieving pre-registration.
50%UnverifiedPre-registration originated from the reform movement following the replication crisis in psychology, in which researchers submit their hypotheses, sample sizes, experimental procedures, and analysis methods to a public platform before data collection, forming a timestamped record.
50%UnverifiedPre-registration has not yet become standard practice in the machine learning field, partly because experimental debugging is itself highly exploratory, making it difficult to fully predetermine hyperparameter search and result interpretation in advance.
50%