1: \begin{abstract}

2: We present in this work a new family of kernels to compare positive

3: measures on arbitrary spaces $\Xcal$ endowed with a positive kernel

4: $\kappa$, which translates naturally into kernels between histograms

5: or clouds of points. We first cover the case where $\Xcal$ is

6: Euclidian, and focus on kernels which take into account the variance

7: matrix of the mixture of two measures to compute their similarity.

8: The kernels we define are semigroup kernels in the sense that they

9: only use the sum of two measures to compare them, and spectral in

10: the sense that they only use the eigenspectrum of the variance

11: matrix of this mixture. We show that such a family of kernels has

12: close bonds with the laplace transforms of nonnegative-valued

13: functions defined on the cone of positive semidefinite matrices, and

14: we present some closed formulas that can be derived as special cases

15: of such integral expressions. By focusing further on functions which

16: are invariant to the addition of a null eigenvalue to the spectrum

17: of the variance matrix, we can define kernels between atomic

18: measures on arbitrary spaces $\Xcal$ endowed with a kernel $\kappa$

19: by using directly the eigenvalues of the centered Gram matrix of the

20: joined support of the compared measures. We provide explicit

21: formulas suited for applications and present preliminary experiments

22: to illustrate the interest of the approach.

23: \end{abstract}

24: