Causal Covariate Shift Correction using Fisher information penalty

Evolving feature densities across batches of training data bias cross-validation, making model selection and assessment unreliable (\cite{sugiyama2012machine}). This work takes a distributed density estimation angle to the training setting where data are temporally distributed. $\mathit{\text{Causal}}\mathit{\text{Covariate}}\mathit{\text{Shift}}\mathit{\text{Correction}}\mathit{\text{(}}{C}^3\mathit{\text{)}}$, accumulates knowledge about the data density of a training batch using Fisher Information, and using it to penalize the loss in all subsequent batches. The penalty improves accuracy by $12.9\mathrm{\%}$ over the full-dataset baseline, by $20.3\mathrm{\%}$ accuracy at maximum in batchwise and $5.9\mathrm{\%}$ at minimum in foldwise benchmarks.