PMDG Jetstream 41

I got myself a new toy... added a Jetstream 41 sim to FSX and it is just a plain fun aircraft to fly. This is also the first time I am using an all 3D cockpit simulator. It takes some getting used to, but really it seems like it is worth it.

This is one screen from the cockpit after having landed and parked at EFTU on a short trip from EFMA.

RIA Services

I must say that I have not catched up on the new technologies emerging now. RIA Services seems like a real effectivity booster when using Silverlight applications. It is not constrained to data centric applications since it can use POCO objects. Need to look into this more.

ReMIX 2009 Helsinki

Today I am attending the ReMIX 2009 seminar in Helsinki. There are two tracks (web developer or web designer) and I would rather have attended both but I had to choose and chose the dev track because of the MVC session in the afternoon.

ProLiant DL380

Just got myself a new toy...

Our cellar is now complete with a real server. This one is 19" rack mounted server with 6 raided hot-swap disks and two Xeon processors. I might admit that it is a bit too much for a home server but I got a very good deal.

F-Secure a performance thief

We have had F-Secure as our standard anti-virus solution for a number of years now. Almost all the users have complained about their computers being sluggish. Today I got an old XP based computer that I was going to fix. It was really painfully slow! Guess what! Uninstalling f-secure fixed the problem. So we will be looking for another anti-virus solution that does not kill our machines.

F-Secure will not enter these doors again, that is for sure. I just felt that I got a new computer today at the cost of a new anti-virus solution. :-)

Windows 7

Ahh... finally!

I have really been waiting for this moment. I know that many have reported that the RC of Windows 7 have been very stable but I was able to resist installing it until the RTM version was available. Now that is is here, I went about installing it. According to the instructions, an upgrade of my Vista installation was recommended and it was commenced. It took 6 hours! But it was worth the wait. A lot of improvments, good improvements as I have seen thus far.

This is really what Vista should have been from the beginning.

Artificial Neural Networks

At work we are using very extensive modelling techniques to try to model the world we are trying to measure. We have so far been very successful in achieving this. Lately, a new project will add a very large amout of parameters to the model and I have really started to question if this model will really be possible to maintain. The problem with models is that they are only valid for a number of set conditions. Once a condition is not set, we don't really know exactly how to use the model or if the values produced by the model can be used in this "out of condition" situation.

I have come to realize that ANN or Artificial Neural Networks might be another way of trying to avoid the complexity of modelling the very complex world and instead concentrate on trying to measure, for each input into the system, how good or bad the result was. If this is done continously, we will very soon have a good knowledge of what consititutes good or bad behaviour. The good thing with this is that the system might also be able to learn to achieve better in situations where the traditional "model" would be out of condition, i.e. certain conditions are outside the bounds specified by the model itself. This is usually refered to by the term Artificial Intelligence in where the system seemlessly seems to be able to adapt to changing conditions. It will not be able to deduce new information from old knowledge but the adaption part is still a very good step forward.

The tricky step in using ANNs is to code the reward function properly. This function needs to learn the ANN what is good and what is bad and usually at levels inbetween the good and bad. In our situation I belive that it is easier to determine the good and bad compared to trying to find a model properly models the world. However, I see one critical aspect of using ANNs. If the reward function does not work properly the system will learn the wrong things. This is of course a very bad thing since the effect of that will be long lasting. Much like if we humans learn to perform a thing the worng way it is very difficult to re-learn.

ANNs are a very interesting concept for replacing complex models with a completely different approach. Unfortualtelly it seams that this is not yet generally accepted by software engineers. ANNs appears yet mostly be University products rather than products delivered from component vendors and such.