| Author |
Message |
| < Technology ~ FPGAs |
|
 Posted:
Sun Jul 10, 2005 5:37 am
|
|
|
Mentor
Joined: Jun 21, 2005
Posts: 94
|
|
Has anyone played with these things yet? I been thinking about getting a developer's kit that interfaces with a PC PCI slot, runs about $1000 for a good one. A company has actually built a super computer with them. There's link below describing their work:
http://www.newscientist.com/article.ns?id=dn7448 |
|
|
|
 |
|
| Posted:
Sun Jul 10, 2005 6:01 pm
|
|
|
Rising Star
Joined: Dec 01, 2004
Posts: 19
|
|
| But... what do you want to do with it? |
|
|
|
|
|
| Posted:
Sun Jul 10, 2005 8:09 pm
|
|
|
Mentor
Joined: Jun 21, 2005
Posts: 94
|
|
| joseaugusto wrote: |
| But... what do you want to do with it? |
I want to build a fast backprop neural net. |
|
|
|
|
|
| Posted:
Mon Jul 11, 2005 8:15 am
|
|
|
Expert
Joined: Nov 17, 2004
Posts: 158
Location: Egypt
|
|
For what end!
_________________
Hemetis |
|
|
|
|
|
| Posted:
Mon Jul 11, 2005 8:39 am
|
|
|
Mentor
Joined: Jun 21, 2005
Posts: 94
|
|
| HEMETIS wrote: |
| For what end! |
Hemetis is that a cry of futility or are you really curious? |
|
|
|
|
|
| Posted:
Mon Jul 11, 2005 9:59 am
|
|
|
Expert
Joined: Nov 17, 2004
Posts: 158
Location: Egypt
|
|
I am more interested than you can begin to believe. 
_________________
Hemetis |
|
|
|
|
|
| Posted:
Mon Jul 11, 2005 10:36 pm
|
|
|
Mentor
Joined: Jun 21, 2005
Posts: 94
|
|
| HEMETIS wrote: |
| I am more interested than you can begin to believe. |
Something smarter than the Sony robotic dog!  |
|
|
|
|
|
| Posted:
Sat Jul 16, 2005 9:24 am
|
|
|
Expert
Joined: Nov 17, 2004
Posts: 158
Location: Egypt
|
|
The propaganda is very tricky, and FPGA would require building circuits that can utilise them and learning VHDL to program them to behave like neural networks. I have experience with those things and they are good for building custom logical circuits under Boolean Algebra rules to minimise the number of gates for effecting a specific set of outputs from a specific set of inputs.
Go ahead and enjoy experimenting but do not expect the FPGA to come to life.
_________________
Hemetis |
|
|
|
|
|
| Posted:
Sat Jul 16, 2005 4:41 pm
|
|
|
Mentor
Joined: Jun 21, 2005
Posts: 94
|
|
| Quote: |
| This is a three-dimensional problem with infinite solutions! How could the simple brain of roach figure this problem out so quickly? |
Actually I have excluded the possibility of the roach’s memory abilities. Considering a simple olfactory location mechanism combine with a random walk if a direct approach to the food cache is not possible, then the roach need have walk the path to the top of the table before. Then an associative memory, which ANNs do, could provide very quick indexing to running up the table to get to my dinner. Of course a roach's memory capacity need only store information for several days before it is reused requiring minimal neurons to store episodic memories. Why a roach's brain remains small.
After doing more research and discovering that tonic firing of neurons is really a simple means of converting chemical signaling to electrical signals. However the possibility of chemical computation is very real especially for pyramid type neurons that are bunched together. Such close approximation of cells would allow computation to operate using chemical reactions as opposed to sinusoidal and sigmodial electrical signaling that ANNs emulate currently. It may be that electrical signaling is the means to transmit information long distances but chemical computation is done between bunches of neurons.
Last edited by Frankinstien on Sun Jul 17, 2005 8:32 pm; edited 1 time in total |
|
|
|
|
|
| Posted:
Sun Jul 17, 2005 8:28 pm
|
|
|
Expert
Joined: Nov 17, 2004
Posts: 158
Location: Egypt
|
|
I am totally fascinated by your wonderful short story.
I will edit my post after dinner.
****
I am back. You have edited your post and deleted your story. 
However, you are quite correct. 
The memory of the roach is definitely demonstrated by its obvious knowledge of the path to your food scent that is always on the table. You were also correct on the fact that darkness triggers the roach to move unseen by you.
It depends on its antenna to seek food scent increasing concentration, which happens to be in agreement with its previously learned path to your table top.
The scent was the motivator, the darkness was the trigger, the concentration of the scent chemicals was the leading factor, realising that the floor path coincided with its previous memory was a confirmation that the food was in the previously well known same place (the table top). The roach then speeds up to the top while it is receiving feedback that constantly confirms its reaction up to your dish.
Isn't nature fabulous!
_________________
Hemetis |
|
|
|
|
|
| Posted:
Mon Jul 18, 2005 8:11 am
|
|
|
Mentor
Joined: Jun 21, 2005
Posts: 94
|
|
| HEMETIS wrote: |
| I am back. You have edited your post and deleted your story. |
It wasn't me who deleted the story!  |
|
|
|
|
|
| Posted:
Mon Jul 18, 2005 8:17 am
|
|
|
Expert
Joined: Nov 17, 2004
Posts: 158
Location: Egypt
|
|
| Frankinstien wrote: |
| HEMETIS wrote: |
| I am back. You have edited your post and deleted your story. |
It wasn't me who deleted the story! |
Most probably you clicked on edit rather than quote, the result of which is overwriting your post rather than posting a new one.
_________________
Hemetis |
|
|
|
|
|
| Posted:
Mon Jul 18, 2005 8:38 am
|
|
|
Mentor
Joined: Jun 21, 2005
Posts: 94
|
|
| HEMETIS wrote: |
However, you are quite correct.
The memory of the roach is definitely demonstrated by its obvious knowledge of the path to your food scent that is always on the table. You were also correct on the fact that darkness triggers the roach to move unseen by you.
It depends on its antenna to seek food scent increasing concentration, which happens to be in agreement with its previously learned path to your table top.
The scent was the motivator, the darkness was the trigger, the concentration of the scent chemicals was the leading factor, realising that the floor path coincided with its previous memory was a confirmation that the food was in the previously well known same place (the table top). The roach then speeds up to the top while it is receiving feedback that constantly confirms its reaction up to your dish. |
I don’t agree that the path to the coffee table was what the roach remembered. If the roach stored every path it had taken during a day as an episodic event it would quickly overload its tiny brain! The roach is better off remembering objects and episodes with respect to some egocentric reference. This allows information compression since only location of an object and the objects qualities are stored. With this approach the roach need only remember the legs of the coffee table as a means to get to the top. It then simply relies on using its olfactory and visual senses to get to the food cache. This implies that the roach must explore its environment inorder to become a proficient survivor.
What can operate in a similar fashion is an Adaptive Resonance Theory model. Below is a link to a pdf file explaining how these models work:
http://cns.bu.edu/Profiles/Grossberg/CarGro2003HBTNN2.pdf |
|
|
|
|
|
| Posted:
Mon Jul 18, 2005 9:22 am
|
|
|
Expert
Joined: Nov 17, 2004
Posts: 158
Location: Egypt
|
|
Let me expand a bit on what I was saying.
The roach's antenna is perfectly adapted to pick up parts per million of food odours.
If there was no food scent, then it is irrelevant if light was on or off, but if there was a food scent, then it stimulates a conditional algorithm in the roaches brain to evaluate the risks and survival emergency degree.
A dim lit room with certainly trigger a go-for-food action as a reaction to the picked up scent.
Now you were discussing the "How" was that food-seeking was carried out so efficiently.
The roach's memory must contain correlations between the "usual" location of food from past experiences while seeking the higher concentration of food odours in the air. It does not have detailed maps of the paths it took during exploration, no. What the roach memorised was significant patterns of order sensible by its vision and scent senses. They are typical to landmarks we use for navigating the streets to our work or other places we know and used to go to.
You should know that the roach must have a kind of consciousness, even if it was inferior to that of humans.
It is aware of sensory patterns being reduced into totality and while it seeks food it "zooms in on" the details of that pattern. This means that the roach is aware of an abstract table (including the one leg it prefers), so it seeks the direction that causes the sensed pattern to be zoomed in on as it is matching the one in the memory.
reaching the leg in itself triggers a command to change direction from horizontal to vertical and using the claws to perform what insects do best. On the table leg, it is a completely different "street" and now the roach is seeking an "edge". Once the roach reaches that edge, it changes direction once more to the horizontal state and recalls the table top plane from that particular perspective to detect new changes from an empty top. Recognising the presence of your dish triggers seeking scent and only scent concentrations right to the dish because that white dish is visually in its reach, hence it does not need to remember anything (for the moment). The roach now is compelled not to change direction, but to keep the escape root exactly behind it or else it must remember every move it makes with the six limbs. After it feeds or if you scared it by loud sound or by switching the lights on, the roach knows its retreat path in the same way it came to be firstly by running back to the edge to seek the leg, then finishing the leg to the floor and back to the place under the sofa where it has experience of safety.
What we need to learn from this scenario for neural networks mode of action, is that the roach needs only to recognise key patterns by remembering the details of those key patterns, and then comparing the extremely compressed data to the memorised ones and seeking a zoom-in-on that data to increase the matching of the details on key patterns only. That key pattern is the signal and all other background data is considered as noise. So what the consciousness of the roach is achieving here is increasing the signal to noise ratio, which is identical to zooming-in-on the physical location of the key pattern along the path.
The reflex equation is much more complex than that of course, but we are focused on a specific scenario for now.
_________________
Hemetis
Last edited by HEMETIS on Fri Jul 22, 2005 3:01 pm; edited 1 time in total |
|
|
|
|
|
| Posted:
Fri Jul 22, 2005 8:11 am
|
|
|
Mentor
Joined: Jun 21, 2005
Posts: 94
|
|
| HEMETIS wrote: |
| If there was no food scent, then it is irrelevant if light was on or off, but if there was a food scent, then it stimulates a conditional algorithm in the roaches brain to evaluate the risks and survival emergency degree. |
Here I disagree with any risk evaluation algorithm since it would be unnecessary, a simple light sensing trigger would due. Obviously dim light conditions are safer than brightly light conditions for the little roach. With that simple reactive circuit the roach has been a very good survivor for tens if not hundreds of millions of years!
| Quote: |
The roach's memory must contain correlations between the "usual" location of food from past experiences while seeking the higher concentration of food odours in the air. It does not have detailed maps of the paths it took during exploration, no. What the roach memorised was significant patterns of order sensible by its vision and scent senses. They are typical to landmarks we use for navigating the streets to our work or other places we know and used to go to. You should know that the roach must have a kind of consciousness, even if it was inferior to that of humans. |
I agree here to the extent of the landmarks as cues but there is still a need to perceive from different 3d perspectives. I don't believe this can be done with out some kind of reference, whether it is an egocentric map or the earth's magnetic pole.
| Quote: |
It is aware of sensory patterns being reduced into totality and while it seeks food it "zooms in on" the details of that pattern. This means that the roach is aware of an abstract table (including the one leg it prefers), so it seeks the direction that causes the sensed pattern to be zoomed in on as it is matching the one in the memory.
reaching the leg in itself triggers a command to change direction from horizontal to vertical and using the claws to perform what insects do best. On the table leg, it is a completely different "street" and now the roach is seeking an "edge". Once the roach reaches that edge, it changes direction once more to the horizontal state and recalls the table top plane from that particular perspective to detect new changes from an empty top. |
I'm not so sure that the roach has to change direction from horizontal to vertical. After all to a roach the whole world is really just flat, as long as there are no obstacles in its way crawling up the leg is no different than walking across the floor.
| Quote: |
| Recognising the presence of your dish triggers seeking scent and only scent concentrations right to the dish because that white dish is visually in its reach, hence it does not need to remember anything (for the moment). The roach now is compelled not to change direction, but to keep the escape root exactly behind it or else it must remember every move it makes with the six limbs. After it feeds or if you scared it by loud sound or by switching the lights on, the roach knows its retreat path in the same way it came to be firstly by running back to the edge to seek the leg, then finishing the leg to the floor and back to the place under the sofa where it has experience of safety. |
Here I disagree that the dish is triggered as a target the roach is relying on olfactory senses. Visual senses are merely to move around obstacles and look out for predators.
The roach's sense of fear is more reflexive, usually when roaches are afraid their run is chaotic or random. This makes sense since a chaotic run makes it difficult to determine the roach's trajectory. Again simpler solutions provide very powerful survivor traits.
BTW roaches, as most insects, are deaf.
| Quote: |
What we need to learn from this scenario for neural networks mode of action, is that the roach needs only to recognise key patterns by remembering the details of those key patterns, and then comparing the extremely compressed data to the memorised ones and seeking a zoom-in-on that data to increase the matching of the details on key patterns only. That key pattern is the signal and all other background data is considered as noise. So what the consciousness of the roach is achieving here is increasing the signal to noise ratio, which is identical to zooming-in-on the physical location of the key pattern along the path.
The reflex equation is much more complex than that of course, but we are focused on a specific scenario for now. |
Well the problem with your approach is figuring out which key patterns are important and roaches' have fixed focused lenses! Here I propose that the ability to resolve object patterns is purely a visual process of recognizing edges but what gives the object importance is its smell, thus it zooms in with smell not vision. |
|
|
|
|
|
News
Predictions
Community
Links and Downloads
![[Valid RSS]](http://www.zapfuture.com/images/valid-rss.GIF "Validate my RSS feed")
![[Valid Atom 0.3]](http://www.zapfuture.com/images/valid-atom.GIF "Validate my Atom 0.3 feed")