Move from Perturbed scheme to exponential weighting average
This work provides incremental insights into algorithm equivalence and efficiency for online learning, benefiting researchers in optimization and decision theory.
The paper tackles the problem of online decision-making by comparing exponential weighting and follow-the-perturbed-leader algorithms, showing they share common properties and can be identical under specific perturbations, and demonstrates that follow-the-leader algorithms extend more efficiently to structured online problems.
In an online decision problem, one makes decisions often with a pool of decision sequence called experts but without knowledge of the future. After each step, one pays a cost based on the decision and observed rate. One reasonal goal would be to perform as well as the best expert in the pool. The modern and well-known way to attain this goal is the algorithm of exponential weighting. However, recently, another algorithm called follow the perturbed leader is developed and achieved about the same performance. In our work, we first show the properties shared in common by the two algorithms which explain the similarities on the performance. Next we will show that for a specific perturbation, the two algorithms are identical. Finally, we show with some examples that follow-the-leader style algorithms extend naturally to a large class of structured online problems for which the exponential algorithms are inefficient.