The Mean-Thinnning and Relative-Threshold Processes

The \(\MeanThinning \) process is the process which samples a bin \(i_1\) and allocates to it iff \(x_{i_1}^t \leq \frac {t}{n}\); otherwise it allocates to another bin sampled uniformly at random.

The \(\MeanThinning \) process:
Iteration: At each step \(t = 1, 2, \ldots \),

  • • Sample two bins \(i_1 = i_1^t\) and \(i_2 = i_2^t\) uniformly at random.

  • • If \(x_{i_1}^t \leq \frac {t}{n}\), then allocate to \(i_1\).

  • • Otherwise, allocate to \(i_2\).

It was shown that this process achieves w.h.p. an \(\Oh (\log n)\) gap, which is tight for any \(m = \Omega (n \log n)\). This process makes in expectation \(2 - \epsilon \) samples per allocation.

Your browser does not support the HTML5 canvas tag.

.

Relative-Threshold processes

A generalisation of this process is the \(\RelativeThreshold (f)\) processes where \(f = f(n)\) is the fixed offset, so \(\MeanThinning \) has \(f(n) = 0\).

The \(\RelativeThreshold (f)\) process:
Iteration: At each step \(t = 1, 2, \ldots \),

  • • Sample two bins \(i_1 = i_1^t\) and \(i_2 = i_2^t\) uniformly at random.

  • • If \(x_{i_1}^t \leq \frac {t}{n} + f(n)\), then allocate to \(i_1\).

  • • Otherwise, allocate to \(i_2\).

Your browser does not support the HTML5 canvas tag.

.