... Hi, Although I'm seriously considering Hayek as "source material" for techniques I will experiment with, I only thought about this so far, I didn't...
... From the FAQ: "Each node adds up the activation from all incoming nodes from the previous timestep." How do you add the activation from the previous time...
... Philip ... in ... I can see I totally mispoke. There *is* a memory at every node. What I meant to convey is that there is not a *special* memory just for...
I'm bringing this up in response to recent talk about how to graph networks and know where to put nodes and tell what links are recurrent. In Mat Buckland's...
There is another way (very similar to described one): Each neuron will have a layer identificator and input neurons will always be a 0 level. Then for each...
Recently I tried to evolve network that could implement XOR operation and added one extra input. Algorithm failed in approximately 10 runs. It's really weird....
Hi, I read this paper with some interest. As you may have gathered from my other posts, I'm interesting in anything that combines evolving components into...
This is mildly off-topic, and yet relevant... I have some data I want to visualize. The nature of the data is a network or mesh of associations. It is far...
Ian, ... Yes the network is essentially reset for each move. I did not allow recurrent connections in order to simplify the process of propagating activation,...
We did an implementation similar to this. We've also doen it with shortest distance to output layer. For the purpose of determining a best guess at which...
... operation ... runs. ... Yuri, Did you try the same thing with x3 being all 1's? It may just be that the network needs a bias (i.e. an input with a...
John, I think this is a very interesting idea, and could be a good way to graph a neural network, or any large network or topology for that matter. The one...
Hi, I have use this approach (briefly) for an unrelated type of graph. There's a few problems: 1) there's no guarantee that there _is_ an internally consistent...
Hi, Just worth mentioning that the one real-life example of something like this which uses big data is circuit-board/chip layout, and that's done by simulated...
Well, I've got it running. It takes a LONG time. Computationally intensive is putting it mildly. And that's after trimming it down to 146 nodes and ...
SharpNEAT, a C# .Net implementation of NEAT now has a rudamentary web site where you can download the current source code and .Net binaries. There is also a...
Colin, Very nice work, I'm looking forward to pretty pictures :) Would you like me to link to your software page from the various NEAT-related or NN-related...
Yes, but visualization can be made through ordering of neurons by their layers. Firstly we draw neurons with "pure" 1-st level (all connections to the input...
Excellent, thank you! A question...do I need to normalize my data? Do I have to specify integer coordinates that correspond to pixels? Or can I just put real...
Hi Joseph, ... Really? I would have thought it easier, just because of the "which stones are mine" angle. ... I don't see how it can distinguish its own...
... that ... No I haven't but it's a good idea, thank you. As for biases thay are defined in run-time. "Hang" neurons (that don't have inputs) can appear but...
I've been tinkering around with the graphical layout for networks. Ken has made a few suggestions along these lines. Here's the most recent version: ...
... No problem. :) ... There are no pixels in SVG...its a vector graphic format. Otherwise, I'm not sure I understand your questions. Basically, though, if...
... Hi Ken, Yes, if you could add a link to the SharpNEAT page that would be great, thanks! Although it's not exactly feature rich right now it still would be...
Colin, one suggestion I have is to make it possible to view champion networks playing the predator-prey domain, in other words to be able to see the predator...
... The same, except we start at the output nodes and move backward. The issue is that nodes are going to be placed in the layer corresponding to their...
Nevermind, answered my own question. This is pretty cool. Unfortunately, the size of the data is prohibitive. I used CSS to reduce the redundant information,...
... their layers. Firstly we draw ... Then (a bit further from inputs) only ... so on. ... in ... Have you tried placing them in in the layer of their average...
Having problems with message search?
Fill out this form to ensure your group is one of the first to be migrated to the new message search system.
