Tuesday: advantages and disadvantages of SDS …

Today yet more PSO write up — will it never end?  — where today I have been writing up the advantages and disadvantages of stochastic diffusion search compared with particle swarm optimisation.  Hasn’t gone too badly in all honesty: I found some great links through the following Google Scholar search — link — in particularly this PhD thesis (http://www.cirg.reading.ac.uk/SDP/Download/DeMeyer-thesis.pdf) and that this 2006 swarm intelligence paper (http://www.cirg.reading.ac.uk/SDP/Download/DeMeyer-thesis.pdf).

I must admit I’m struggling at this moment: I’ve always been a dreadful that finally finishing things off, putting the final nails in the coffin so to speak, so all these final tweaks are very painful.  More annoyingly I keep coming across really interesting looking papers and keep getting distracted — rubbish!  So although it feels like crawling over broken glass at the moment, I am still (slowly) trudging forwards, or so I hope… However,  currently working in 2hr blocks, which seems to be quite good: read for 1hr, think during a break, and then come back and type up thoughts — nice!

Sunday: tracking people’s biographies down …

Found the following:

• Xiaohui Hu
• engineer
• research interests in: particle swarm optimisation; complex and chaos systems; artificial life; nonlinear multiobjective optimisation; artifical nerual networks, fuzzy logic and evoutionary computation
• links: see here –> http://www.swarmintelligence.org/xhu.php
• achievements: application of PSO to human tremor analysis; applicaiton of PSO to multiobjective optimisation
• Yuhui Shi
• engineer
• research interests: particle swarm optimisation; Electrical Vehicles; Cursive Script Word Recognition; Artificial Neural Networks; Evolutionary Computation; Digital Signal Processing; Biomedical Data Acquisition and Analysis; Fuzzy Logic http://www.engr.iupui.edu/~shi/
• ahcievements: similar to Hu — collaborated on many pieces together
• James Kennedy
• social psychologist
• link: http://www.particleswarm.info/jk/vita.html
• achievements: developed PSO method with Eberhart; ongoing developments of PSO
• Rusell Eberhart
• engineer
• research interests
• link
• achievements: developed PSO method with Kennedy;
• Rui Mendes
• Mendes has published papers involving particle swarms since 2002, including several regarding the effect of topologies on particle swarms. Study findings have included the relationship between the “choice of the social topology and the robestness of the algorithm to premature convergence” (Population Topologies and Their Influence in Particle Swarm Performance, 2004, see http://www.di.uminho.pt/~rcm/pubs.html)

Other stuff:

• link to using Kalman filters with PSO: link
• saw this in Hu X H, Eberhart R C, Shi Y H. Engineering optimization with particle swarm. IEEE Swarm Intelligence Symposium, 2003: 53-57 (http://citeseer.ist.psu.edu/hu03engineering.html): “PSO limitations: due to random origin orf evolutionary algorithms, difficult to deal with equity constrats. Almost impossible to find a group of initial solutions in the feasible space. Also applieds to problems with extremely small feasible space. 2nd for some engineering optimisation problems invlving interger variables, the algorithm is not stable occasionaly and can be trapped in local minima”

Still wondering what you need to do in a biography though …

Saturday: a quick one …

NB Website to accompany Kennedy and Eberhart’s book: link.

Thursday: finished the graphs!! …

I think I’ve finished the graphs — woohoo! Got them all done in SVG: now the challenge is going to be getting SVG images into a format I can whack into LaTeX ….

Tuesday: worked out how to do the re-doing …

Tonight been hitting Inkscape redrawing figures. Took me a while to work out what I needed to do, but have at last sussed it — I hope! Now just got to put the hours in to do the redrawing … sigh ;-)

Monday: re-doing …

Spent some time looking into how to go about redrawing images, including some of graphs. Experiemented with Powerpoint and Inkscape, but am starting to think the Excel could be the way forwards … Didn’t get too far, so still need to make some headway on this — like many other things!

Sunday: should I recode the simulation into SAS? ….

I’m been to-ing and fro-ing about this for a while now, and I’m wondering whether it would be worthwhile rewriting the Java simulation I have in SAS, to firstly remove the visual component of the simulation (hence speeding up the simulation), and secondly to ensure that the outputted data is ready for analysis in SAS, and does not need any additonal steps to analyse it.

The other idea is to take the Java simulation I have at the moment, and hack out all of the visual componentry. It would be useful to be able to “switch off” the visual bits as needed, to speed up the simulation, but be able to turn them back on when wanting to look at what happens.

My initial thought is to tranpose everything into SAS …. but the issue is that my SAS experience is not great, plus I have more experience with Java at this point in time. Hence, I will attempt to “un-visualise” the Java simulation for now.

From looking a the UML diagram for the Flocking3D simulation, the following classes are graphical and hence should be cut-able or need modifiable:

• ColouredTiles
• CheckeredFloor
• WrapFlocking3D
• BoidShapre
• Obstacles (?)

Been through the code and it looks like WrapFlocking3D does most of the calling of the visual parts … Also attempted to hack bits out of this as well, but needless to say the simulation has now stopped working!!

Wednesday: more rewriting …

Still more re-writing: today worked on some of the applications of PSO, to extend them somewhat.

Monday: more re-writing …

Getting there slowly, but still adding in more information around PSO.

NB saw this which could be useful later on: Critique of genetics algorithms:

Saturday: very quick points ….

Lyx stuff

• Insert > Math > AMS Falign to do equations that are aligned
• In an math insert, press Ctrl-M to get into text mode to add in “If, Then, Else, etc” — see http://www.stat.rice.edu/~helpdesk/howto/lyxguide.html
• To number an algorithm, move into the equationand type Alt-m n — see http://www.ee.iitm.ac.in/~dj/ec205/jan8.pdf for details

Toner and Tu flocking model  … ?