My Neural Nets Phase

Every tech-curious person should have one of these.  Mine benefits from the framework of SpringThing, which is a network generator I started writing in 2006 instead of doing systems characterisation homework.  It is a great platform for visualising the interaction of connections and nodes.

Here are my introductory thoughts on neural networks:

• Training nets with data which has a known confidence interval or statistical basis -- the known accuracy of the inputs could be used to determine a convergence criteria (i.e. no need to be an expert on Mickey Mouse).
• Generally, thinking about the concept that the higher the input/output ratio, the stronger or more robust the output.
• Explicit, parallel co-update of connections and nodes.
• Connection weights could be visualised by adjusting the connection length, giving a visual clustering, and animating the learning process.
• Adaptive networking which adds connection "short-cuts" along heavily trafficked routes from input to output;  and generally adaptive networking.
• The irrelevance of directionality in an explicitly (in the temporal sense) integrated network, which can be thought of more as a "sponge" than a serial machine.
• Because my framework does away with directionality, there are not clear input/output groups for each node.  Thus I'll try having two variables in each node - an "internal" value which is the sum of the connected nodes times their respective weights, and an "external" value which is presented to the neighbouring nodes.  The internal value is processed using a basis function to determine the external value, which is then used by adjacent nodes to make decisions.
• Input and output nodes need not be ordered, or even adjacent.
• Simultaneous training and guessing... no need to separate these phenomena apart from the fact that when it is possible to estimate the error, corrections are attempted.
• The use of an array of basis functions as a meta-degree of freedom is something I'm willing to try.  At least globally at first, then maybe locally.  Would likely disturb any gradient-based optimisation too much.
• Lastly, after talking to Karin about my maturity level, I realised that human pride is similar to a "learning speed" parameter I've read about being used to stabilise the learning process.  How, um, useful...

One of my first efforts will be to have the mouse produce a yellow dot.  The image will be fed to the net, which will estimate the mouse's [X, Y] position.  The actual position will be used to train the network.  This will occur within the animation loop of SpringThing, and thus the (hopefully) studious reaction of the network to the moving mouse will be animated in realtime, either by colouration of the connections by weight, or by adjustment of connection length.

Check out SpringThing (sample shown at right) to see what I'll be starting with.

Here's a great recent interview with a guy doing hardware studies:
http://www.nanotech-now.com/news.cgi?story_id=37079

There are a lot of really great looking packages that I'm avoiding for now.  I plan to get some training and optimisation ideas from them though, as that seems to be the most difficult aspect of this project.
http://www.ire.pw.edu.pl/~rsulej/NetMaker/

Magic Point Rock Slink

Botany Bay Storm

I planned a dive off Cape Solander (aka "The Steps") but it was milky, so I read my kindle on a cliff instead.  It was so sunny.  I finished a nice chicken pot pie I got at the malabar bakery, and headed down to Oak Park.  There, I could see in the distance a huge storm was forming.  It came on fast an hard.

 I ran for a cliff overhang instead of back to the car, for fun.  The gulls didn't want to come in under the cliff with me.  The just pointed their beaks into the hail.

 Suddenly I could only see about 30'.  Sailboats disappeared.

 This is the cliff I hung under.

 Right after the front passed, an inflatable boat with two spearfishermen hooting and hollering came zipping by.

 And then it was gone.  The sun came out and I headed back to the other side of the bay (a 40 minute drive) to check out Bare Island.  I think I dove then.  Haven't been diving too much.

Tasmania - Big Quiet Dunes

This post is in reverse-chronological order.  So we "start" in Queenstown.  Queenstown, which I think should perhaps be named "Queen'stown" to be grammatically correct, was destroyed by unfettered backwater mining operations over the course of the last 100-odd years.  It is a treeless landscape, because the trees all died due to effluent and emissions from the town's ginormous smelting operations.  Here's a pit mine that had a mine-sponsored placard with information concerning historical events, such as how some things might not be as bad as they seem.  Something about how some miners died in the 60's and how it wasn't really the company's fault. 

This is a stream rolling through a town where the children can't play in their local stream.  Because it is toxic now, and forever more because of a huge tailings pile that wasn't properly managed.  There is a wonderful juxtaposition of orange flowers...

This is a "headframe" which is a mining elevator works.  One of these failed at Olympic Dam.  The shaft there is about 1km deep!

An overlook.  Just great.  Note: reverse chronological order means that we are now outside of Queen'stown, approaching it.

... and the dunes.  The namesake of this post.  So great.  What can I say?  A lot!

TASMANIAN DEVILS EVERYWHERE!  With cute paw-prints.  And tail-swagger prints!

Karin in our tent.

... our tent alone at sunset.

We camped on top of a large pointy dune that was unlikely to be driven up by bogan hoonage.  They tended to camp down closer to the water, because they didn't have to hike in the gear over loose sand dunes.  Oh man it was so worth it.  Let me just list a few things that made this place an excellent night, our best in Tasmania in fact:

1. Stunning visual beauty - vast expanses of huge white dunes.
2. You could hear (and see) the ocean in the distance.  
3. A prevailing breeze meant no mozzies.
4. No people.
5. Free!

We woke up in the morning to find tassie devil tracks going past our tent.  I tracked them deep into the dune complex but found no actual devils or den holes.  There is some thick-as scrub in the lowlands where moisture collects.

There's Karin and the ocean.  Karin is brushing her tuffy toofers and the ocean is doing it's thing.  Authentic gorilla chest.

Aforementioned tail-swagger tracks from a fat devil:

Us, before bedding down for the night in a fantastic dune.

Here they are in case you want to enjoy this place too:

View Larger Map