Some optimisation processes don’t start from any particular configuration, but from the configuration space in general, and iterate their way towards a better configuration
Deterministic
A particular kind of stochastic optimisation process
Destination macrostate should be a pareto improvement over the source macrostate according to some sensible measure
My current measure candidate is the worst case performance of the destination macrostate
This is because I define the macrostate of a given configuration as the set of all configurations that are at least as good according to the objective function(s), so the measure is simply the performance of the destination configuration
“Performance” here being something like the vector of all values a configuration obtains for the objective functions
Stochastic
Induces a different probability measure over configuration space?
Does the starting macrostate matter for the induced probability measure?
Desiderata/Thoughts on Optimisation Process:
Optimisation Process
Map on configuration space
Between distinct configurations?
Between macrostates?
Currently prefer this
Some optimisation processes don’t start from any particular configuration, but from the configuration space in general, and iterate their way towards a better configuration
Deterministic
A particular kind of stochastic optimisation process
Destination macrostate should be a pareto improvement over the source macrostate according to some sensible measure
My current measure candidate is the worst case performance of the destination macrostate
This is because I define the macrostate of a given configuration as the set of all configurations that are at least as good according to the objective function(s), so the measure is simply the performance of the destination configuration
“Performance” here being something like the vector of all values a configuration obtains for the objective functions
Stochastic
Induces a different probability measure over configuration space?
Does the starting macrostate matter for the induced probability measure?
Separate into its own post?
Thoughts on stochastic optimisation processes:
Definition: ???
Quantification
Compare the probability measure induced by the optimisation process to the baseline probability measure over the configuration space
Doesn’t give a static definition of work done by a particular process?
Maybe something more like “force” of optimisation?
Do I need to rethink the “work” done by optimisation?
Need to think on this more