View unanswered posts | View active topics It is currently Thu Apr 26, 2018 3:36 am

 Page 3 of 3 [ 39 posts ] Go to page Previous  1, 2, 3
 Print view Previous topic | Next topic
Author Message
Guru

Joined: Sat Mar 01, 2008 12:52 pm
Posts: 1030
Re: File read and write functions
hi,
thx, so far I've got it.

But what actually are the neurons in a CANN (continuous attractor neural network)?
How does a neuron in the CANN look like,
where do it's dendrits come from and the axon go to,
what is it's calculating algorithm,
and how many neurons do you need depending on the number of sensors and the number of squares in your map?

To show why this point is still obscure to me, let me try to explain this by three examples:
the a-star
the BUG/BUG2
an artificial neural net like a backpropagation net.

An approximating system also is used by the a-star and the BUG(BUG2) algorithm.

a-star is useful if the map is nearly complete (most of the locations of obstacles are known),
BUG/-2 simply drives to the estimated direction and rounds obstacles just if any of them appear.
In case of the a-star you need a detailed map,
in case of the BUG/-2 you don't.

And in both cases you don't need any a neuron.

If you have a small neural net, you may train behavior rules depending on sensor values; e.g.:
if distance sensors show "no obstacles ahead" try output "two motors forward",
if distance sensors show "obstacles right" try output "motor right forward, motor left float",
if distance sensors show "obstacles right and front" try output "motor right forward, motor left reverse".
Notify, that you don't program these behaviors in advance, but train it in case a event happens, and a shown spontaneous behavior was wrong (or maybe even conditionalize it if it was right). Now: what happens inside the neurons?
Step by step the neuron inputs (sensor values) are differently weighted by each neuron cell and lead to a vector room (R^[inputs] x R^[outputs] ) of input-output-patterns with local error maximums and minimums. If you reach a global minimum of errors, the net has learnt it's lesson.
The result of this training may not be very different from a bug algorithm if you just train him to round an obstacle. You also don't need to know anything about your environment, you don't need to have a map, you just let the robot react spontaneously to singular events (an obstacle appears wherever). The only difference is that at a BUG you have to program (via if..then..else) all events in advance, while at a NN you let all things happen and just educate the robot if anything was incidentally wrong (or right).

Like other nets like "Elman nets", "Sigma-Pi-neurons", "self organizing maps (SOM, Kohone Feature Maps)", the "Hopfield nets", the "Boltzmann machine", or the "Adaptive Resonance Theory (ART)" , surely even the CANN, every net has it's very typical network structure. But although all nets are constructed in a very different way, and although they use different functions to lead to the error minima: the construction of each neuron and the weighting and activation algorithms (net and act) of each neuron all are still quite similar.

So back to my starting question:
what actually are the neurons in a CANN?
How does a neuron in the CANN look like,
where do it's dendrits come from and the axon go to,
what is the calculating algorithm,
and how many neurons do you need depending on the number of sensors and the number of squares in your map?

_________________
regards,
HaWe aka Ford
#define S sqrt(t+2*i*i)<2
#define F(a,b) for(a=0;a<b;++a)

Sun Sep 21, 2008 5:57 am
Expert

Joined: Sat Sep 20, 2008 12:16 pm
Posts: 106
Location: Curently fighting aliens on P3X-828
Re: File read and write functions
I guess that it would take a HUGE amount of data to program something like that. btw, programing something like that would take a lot of time...

_________________

 Quote:What the world needs is more geniuses with humility, there are so few of us left.

Agreed

Sun Sep 21, 2008 6:42 pm
Guru

Joined: Sat Mar 01, 2008 12:52 pm
Posts: 1030
Re: File read and write functions
no, surprisingly not!
e.g., I've built a backpropagation and an Elman net - it's the same issue if I configure it
for 5 neurons with 3 inputs and 2 outputs or
for 4000 neurons with 3000 inputs and 100 outputs *).
The net simply calculates 1 neuron after the other in a series of layer arrays, you just have to adjust the connections like
for (i=0; i< nLayer1; i++) do (NeuronLayer0[i].input=(NeuronLayer1[i].output); **)

The rest is trivial ***)

*) but due to the 15 k memory limit at RobotC the size is physically limited to ~40 neurons over all for nxt based neural nets
**) this surely works only as far as the compiler doesn't always produce nan errors
***) trivial = the most hated word among mathematicians

_________________
regards,
HaWe aka Ford
#define S sqrt(t+2*i*i)<2
#define F(a,b) for(a=0;a<b;++a)

Sun Sep 21, 2008 7:05 pm
Rookie

Joined: Wed Apr 30, 2008 7:05 am
Posts: 18
Location: Brisbane, QLD
Re: File read and write functions
success - i have ratSLAM working on the lego NXT (using the the entire available memory) with experience mapping and loop closure occuring. Thesis is done

_________________
experiencing slow death by thesis

Thu Oct 16, 2008 1:00 am
Guru

Joined: Sat Mar 01, 2008 12:52 pm
Posts: 1030
Re: File read and write functions
what does this mean?
what has this to do with file r/w functions?
what can be done with this?
what has been improved?

_________________
regards,
HaWe aka Ford
#define S sqrt(t+2*i*i)<2
#define F(a,b) for(a=0;a<b;++a)

Thu Oct 16, 2008 4:01 am
Rookie

Joined: Wed Apr 30, 2008 7:05 am
Posts: 18
Location: Brisbane, QLD
Re: File read and write functions
This means i have implemented ratSLAM on the lego nxt.

This is just following on from the previous topics in the read/write sections.

It is able to generate a real time/online map of an area using local view, pose cells and experience to do this. It demonstrates loop closure in the map and demonstrates some ability to handle being 'kidnapped'.

Improvements...well think about it. I have taken ratSLAM which currently running on a pioneer DXE wirelessly connected to a PC to share the computation to running it on it onboard a lego NXT.

_________________
experiencing slow death by thesis

Thu Oct 16, 2008 4:06 am
Guru

Joined: Sat Mar 01, 2008 12:52 pm
Posts: 1030
Re: File read and write functions
what has this to do with file r/w functions?
what can be done with this?

what has it do do with this and what are the neurons like using ratslam?
I don't see the point.

_________________
regards,
HaWe aka Ford
#define S sqrt(t+2*i*i)<2
#define F(a,b) for(a=0;a<b;++a)

Thu Oct 16, 2008 4:11 am
Rookie

Joined: Wed Apr 30, 2008 7:05 am
Posts: 18
Location: Brisbane, QLD
Re: File read and write functions
it was in regard to the CANNs and the doubters who believed it couldn't be done. A full version of the code and results can be made available on the proviso that my thesis supervisor agrees. Besides i started this thread i write wat i want

_________________
experiencing slow death by thesis

Thu Oct 16, 2008 6:32 am
Guru

Joined: Sat Mar 01, 2008 12:52 pm
Posts: 1030
Re: File read and write functions
surely you can write what you want
I just didn't see the relation to the starting question.

_________________
regards,
HaWe aka Ford
#define S sqrt(t+2*i*i)<2
#define F(a,b) for(a=0;a<b;++a)

Thu Oct 16, 2008 6:40 am
Display posts from previous:  Sort by
 Page 3 of 3 [ 39 posts ] Go to page Previous  1, 2, 3

#### Who is online

Users browsing this forum: No registered users and 2 guests

 You cannot post new topics in this forumYou cannot reply to topics in this forumYou cannot edit your posts in this forumYou cannot delete your posts in this forumYou cannot post attachments in this forum

Search for:
 Jump to:  Select a forum ------------------ ROBOTC Applications    ROBOTC for LEGO MINDSTORMS       Third-party sensors    ROBOTC for CORTEX & PIC    ROBOTC for VEX IQ    ROBOTC for Arduino    Robot Virtual Worlds    Multi-Robot Communications    Issues and Bugs Competitions & Partners    Mini Urban Challenge    CS2N Robot Virtual Worlds Competitions       VEX Skyrise Competition 2014-2015       VEX Toss Up 2013-2014       FTC Block Party! 2013-2014    Competitions using VEX - BEST, TSA, VEX, and RoboFest!    FTC Programming    RoboCup Junior and Other ROBOT Competitions Virtual Brick Robotics Discussions    General Discussions    Project Discussions Off-Topic ROBOTC Forum & ROBOTC.net Suggestions/Feedback    ROBOTC Forums Suggestions/Comments    ROBOTC.net Suggestions/Comments       NXT Programming: Tips for Beginning with ROBOTC       VEX Programming: Tips for Beginning with ROBOTC    2013 Robotics Summer Of Learning       VEX Toss Up Programming Challenge       FTC Ring It Up! Programming Challenge    International Forums       Spanish Forums          ROBOTC for MINDSTORMS          ROBOTC for VEX       French Forums          ROBOTC pour Mindstorms          ROBOTC pour IFI VEX       Japanese Forums （日本語のフォーラム）       German Forums    2015 Spring Carnival Event    PLTW (Project Lead The Way)    Robotics Merit Badge    2014 Robotics Academy Summer of Learning