1: \begin{abstract}

2: The extraction of a common signal across many recordings is

3: difficult when each recording -- in addition to the signal -- contains 

4: large, unique variation components. This is observed for voltage

5: sensitive dye imaging (VDSI), an imaging technique used to measure neuronal

6: activity, for which the resulting 3D array data have a highly

7: heterogeneous noise structure. Maximin aggregation (magging) has

8: previously been proposed as a robust estimation method in the presence

9: of heterogeneous noise. We propose soft maximin aggregation as

10: a general methodology for estimating a common signal from heterogeneous

11: data. The soft maximin loss is introduced as an aggregation of

12: explained variances, and the estimator is obtained by minimizing the

13: penalized soft maximin loss. By establishing  

14:  properties of  the general soft maximin aggregation loss we are able to show

15:   convergence  of a  proximal

16: gradient based algorithm when applied to the estimation problem.  Using this we  obtain a time and memory

17: efficient algorithm for data with  tensor-array structure. The implementation is 

18: provided in the R package

19: \verb+SMMA+ available from CRAN. We demonstrate that soft

20: maximin aggregation performs well on a VSDI data set with 275 recordings,

21: for which magging does not work. 

22: \end{abstract}

23: