The world is being quietly rearranged by people who write very long documents.


The title they went with An LP-based Sampling Policy for Multi-Armed Bandits with Side-Observations and Stochastic Availability Noisy translates that to

Algorithm learns faster when some options disappear unpredictably


Researchers developed a better decision-making algorithm for situations where you can't always access all your options, and some choices give you information about others. In real systems like social networks where users come and go, this approach learns faster which options are actually best by smartly choosing which available choices to test.
Most algorithms that learn by trial-and-error assume all options stay available forever — but in the real world, options vanish and reappear randomly, forcing you to learn from incomplete information; this work shows you can still learn efficiently by mathematically planning which temporary options to sample.

If you insist
Read the original →