Limited stochastic meta-descent for kernel-based online learning

To improve the single-run performance of online learning and reinforce its stability, we consider online learning with limited adaptive learning rate in this letter. The letter extends convergence proofs for NORMA to a range of step sizes, then employs support vector learning with stochastic meta-descent (SVMD) limited to that range for step size adaptation, so as to obtain an online kernel algorithm that combines theoretical convergence guarantees with good practical performance. Experiments on different data sets corroborate theoretical results well and show that our method is another promising way for online learning.

Accelerating IMRT optimization by voxel sampling

This paper presents a new method for accelerating intensity-modulated radiation therapy (IMRT) optimization using voxel sampling. Rather than calculating the dose to the entire patient at each step in the optimization, the dose is only calculated for some randomly selected voxels. Those voxels are then used to calculate estimates of the objective and gradient which are used in a randomized version of a steepest descent algorithm. By selecting different voxels on each step, we are able to find an optimal solution to the full problem. We also present an algorithm to automatically choose the best sampling rate for each structure within the patient during the optimization. Seeking further improvements, we experimented with several other gradient-based optimization algorithms and found that the delta-bar-delta algorithm performs well despite the randomness. Overall, we were able to achieve approximately an order of magnitude speedup on our test case as compared to steepest descent.