Diapositive 20 sur 26
Now I would like submit to you some ideas about how such a method could be improved.
First, we need a criterion to evaluate an algorithm. This is the cost notion. As we will see, iteration number is usually not a good one.
Second, we have to wonder if an optimal initialization can be found and if more "freedom" for the particles would be fruitful.
At last, a few word about the sub-optimal use of the information we have.