Jump to content

User:Diffeomorphicvoodoo/sandbox/Multiplicative weights update method

From Wikipedia, the free encyclopedia
The printable version is no longer supported and may have rendering errors. Please update your browser bookmarks and please use the default browser print function instead.

Motivating example

Multiplicative weights update algorithm

We have n experts who give advice about some decision each day. Each decision incurs a certain loss which is revealed only after we make the decision. The loss incurred at the tth day if we follow expert i's advice is given by . The following algorithm picks the experts in a way such that the total loss incurred over T days is approximately the loss incurred by the best expert.

Algorithm

  1. for i = 1 to n
  2. .
  3. for t = 1 to T
    Sample an i from the probability distribution and follow expert i's advice.
    Observe the loss obtained by the experts -

Analysis

Theorem: For any , the expected loss incurred by the above algorithm is bound as follows:

(The inequality holds for all i = 1 to n, and, in particular, for the expert that minimizes the losses).

Proof:

.

Since , for ,

.

Since for all ,

.

Therefore, by induction,

Furthermore, .

The theorem follows immediately from the above two inequalities.

Matrix multiplicative weights update algorithm

This is a matrix generalization of the multiplicative weights update algorithm. We associate an expert to every unit vector . The loss incurred by an expert $v$ on day t is given by , where is the loss matrix for day t.

Algorithm

  1. for i = 1 to n
  2. .
  3. for t = 1 to T
    Follow expert's advice according to
    Observe the loss matrix for day t -

Analysis

Theorem: For any and for any , the expected loss incurred by the above algorithm after T rounds is bound as follows:

Category:Algorithms