An ablation study compares a computational system with variants in which a component, input feature or training choice is removed or replaced. In machine learning, the purpose is to investigate which parts contribute to the measured behavior under a specified training and evaluation procedure. A useful comparison names the change, the metric and the conditions held fixed.
A measured example appears in Attention Is All You Need by Ashish Vaswani and others. Its transformer model comparisons include removing label smoothing: the base system's development-set BLEU is , versus without label smoothing. Meanwhile, wordpiece perplexity changes from to , an improvement for that metric. The same removal can therefore improve one measurement and worsen another. These are English-to-German machine translation results on the named development set, not universal effects of label smoothing.[ref]
A synthetic example shows why effects need not add. Suppose accuracy is with neither component A nor B, with A alone, with B alone, and with both. Removing either from the full system changes accuracy by zero percentage points. Adding either to the system with neither improves accuracy by ten points. The components provide redundant benefits: a zero removal effect does not imply that a component could never be useful. The statistical interaction is visible in percentage points.
For an interpretable ablation study, record the data split, metric, training budget and any settings changed to keep the variant viable. Retraining a reduced artificial neural network and suppressing a component only during evaluation answer different questions. Repeated training runs help assess variability; one difference alone does not establish statistical reliability. Broader claims require further comparisons.
New to topics? Read the docs here!