Stochastic Online Correlated Selection

We initiate the study of Stochastic Online Correlated Selection (SOCS), a family of online rounding algorithms for the general Non-IID model of Stochastic Online Submodular Welfare Maximization and its special cases such as Online Stochastic Matching, Stochastic Ad-Words, and Stochastic Display Ads. At each time step, the algorithm sees the type of an online item and a fractional allocation of the item, then immediately allocates the item to an agent. We propose a metric called the convergence rate that measures the quality of SOCS algorithms in the above special cases. This is cleaner than most metrics in the related Online Correlated Selection (OCS) literature and may be of independent interest. We propose a Type Decomposition framework that reduces the design of SOCS algorithms to the easier special case of two-way SOCS. First, we sample a surrogate type whose fractional allocation is half-integer. The rounding is trivial for a one-way surrogate type fully allocated to one agent. For a two-way surrogate type split equally between two agents, we round it using a two-way SOCS. We design the distribution of surrogate types to get two-way types as often as possible, while respecting the original fractional allocation in expectation. Following this framework, we make progress on nu-merous problems including two open questions related to AdWords.