A Randomized Mirror Descent Algorithm for Large Scale Multiple Kernel Learning

We consider the problem of simultaneously learning to linearly combine a very large number of kernels and learn a good predictor based on the learnt kernel. When the number of kernels $d$ to be combined is very large, multiple kernel learning methods whose computational cost scales linearly in $d$ are intractable. We propose a randomized version of the mirror descent algorithm to overcome this issue, under the objective of minimizing the group $p$-norm penalized empirical risk. The key to achieve the required exponential speed-up is the computationally efficient construction of low-variance estimates of the gradient. We propose importance sampling based estimates, and find that the ideal distribution samples a coordinate with a probability proportional to the magnitude of the corresponding gradient. We show that in the case of learning the coefficients of a polynomial kernel, the combinatorial structure of the base kernels to be combined allows sampling from this distribution in $O(łog(d))$ time, making the total computational cost of the method to achieve an $\epsilon$-optimal solution to be $O(łog(d)/\epsilon^2)$, thereby allowing our method to operate for very large values of $d$. Experiments with simulated and real data confirm that the new algorithm is computationally more efficient than its state-of-the-art alternatives.