BalayarCB 0 Report post Posted November 11, 2005 Most of computer lads have heard or learnt about NN. Can anyone elaborate how we can implement Neural Network in P2P environment? Feed forward Neural Network and BackPropagation? Quote Share this post Link to post Share on other sites
okinawa 0 Report post Posted November 19, 2005 Most of computer lads have heard or learnt about NN. Can anyone elaborate how we can implement Neural Network in P2P environment? Feed forward Neural Network and BackPropagation? may be this paper would be helpful to you: http://www.cs.mu.oz.au/~aharwood/online/Ha...ruong-2004a.pdf r u using market available ANN software or home-built? Quote Share this post Link to post Share on other sites
BS 0 Report post Posted November 19, 2005 My understanding of Neural Network is not related with this P2P stuff. I use NN for classifying satellite images and its interpretation. I will check the disucussion in this topic. Interesting. Quote Share this post Link to post Share on other sites
BalayarCB 0 Report post Posted November 19, 2005 Benktesh Ji, You're right. Neural Networks i.e. ANN(aritificial neural networks) are mostly used in pattern recognition and in cancer diagnosis too. But, what I am trying is to use ANN for discovering peers from pool of peers i.e. a new methodology for discovering devices in p2p environment. Can it be efficient we are working on it? Any discussion or ideas will be welcomed. and, any links ...too.. thanks. Quote Share this post Link to post Share on other sites
BS 0 Report post Posted November 19, 2005 Balayar jee, Thanks for the reply. Idea getting interesting. I used that about 3 years ago while I was doing my research. It took a week to set up the network environment with occasional network crash. But I did not that for image classification. Theoritically, you idea matches well. I cannot give any general idea but "encouragement" to DO IT. Good luck and let this forum know about the result. Thanks Quote Share this post Link to post Share on other sites
okinawa 0 Report post Posted November 21, 2005 (edited) I ve played with feed-forward, recurrent ANN as a forecasting tool. ANN is a good tool when the parameters are highly non-linear and complex. Donno much about P2P, may be interesting if ANN is applied. Give a try and see the result. Why not to search IEEE papers? go to sciencedirect.com or ISI web search. You might get few papers. Good luck. Edited November 21, 2005 by okinawa Quote Share this post Link to post Share on other sites
Limitation//Moon 0 Report post Posted November 21, 2005 Hi Balayar As far as I have invloved, P2P Networking ( if your are talking for core networking) has physical limitation. P2P has different implementation in office environment and internet environment. let me know if, you... Are you trying to research NN implementation on p2p as an hobbie at home ? Are you trying to research NN implementation on p2p as an academic/industrial research ? however in your own pleasure, if you want tools for NN , go http://www.mathworks.com/products/neuralnet/ ...ahead If you are just a starter (student) into the world of neuronal networks,I recommend also to study fluxes in a network in parallel. The later can be used in large scale "ethernet"/ (IEEE x.xx) networks and the correlations between the two are large. For a start, check Robert Sedgewick's Algorithms[1]. Well, in any case, Sedgewick's book is a must. have a TASTE at [1] http://annie.sourceforge.net/ [2] http://www.torch.ch/ [3] http://ieee.uow.edu.au/~daniel/software/li...sh/BPN_English/ Also, let me know, what programs/software/tools do you tend to use for algorithms ? MATLABS ( Goooooogle it) C++ ??? (hands up!!) .... C++ suck, not really I mean ...I suck Quote Share this post Link to post Share on other sites
Limitation//Moon 0 Report post Posted November 21, 2005 Furthermore, for BP, The common algorithmic technique is called back propagation, with its two variants: forward learning and backward re-weighting number [3] link of the above post. This remind me of cisco days duh!. However have forgotte them. Quote Share this post Link to post Share on other sites
BalayarCB 0 Report post Posted November 21, 2005 Guys, thxs for the links and suggestion, ideas. am too really a new bie in the field of Neural network, FUZZYzim. what we're trying in my lab is to find proper devices(peers/nodes) from 20-devices networked, subnetted in my lab using FFNN(feed forward neural network). As you all know, FFNN uses Backpropagation for training the network, we too uses backpropagation with gradient. Suppose, A peer(device) has 4-neighbors and those neighbors have their own neighbors. And, each device can have its own resources, hops(from the originator). Hence, we took neighbors, neighborsneighbors and hops etc.. as input to decide the OUTPUT 1 or 0(zero) which means forward or not forward. Suppose, a peer has a query and it needs help from others, it can execute FFNN for all the peers it knows(its neighbors) and with the RESULT 1 or 0, it decide from whom it can get help and DISTRIBUTE the task among those peers(It might be high computation or ...). THE PROBLEM is, to train the Network, training data(input parameters) and number of hidden neurons for getting most efficient result. Next, the problem is HOW it can be different from Current, BFS, Random Walks, Gnutella, Napster methods? The Comparision? Anyway, idea or suggestion will be highly appreciated. Quote Share this post Link to post Share on other sites
Limitation//Moon 0 Report post Posted November 22, 2005 You said you are trying to subnet 20 modes ??? Does 20 nodes needs to be subnetted ? ( EXCEPT OTHERWISE, SECURITY RELATED ) THEN, ACTUALLY you are not networking on P2P model. There will be a crash if you use only one hop in 20 nodes...lol, however you have not mentioned how maany hops you intended to use ?? Next, the problem is HOW it can be different from Current, BFS, Random Walks, Gnutella, Napster methods? The Comparision? The difference is due to the emerging technology. BFS, Random Walks, Gnutella, Napster are just the modified version of perl, and open source developers that does not need huge scientif or theoritical knowledge. However, Neural network is a principled network in terms of deterministic algorith theory, driven by flow chart, formula. If you are already familiar with the backpropagation algorithms, you may have noticed it is very slow/inefficient in realistic scenarios. A good paper[ http://www.cs.bham.ac.uk/~pxt/NC/schiffmann.bp.pdf ] dealing with its optimization has been written by Schiffmann (a Gooooogle for "schiffmann neural network" also is interesting). A kind of review can be found at the datamininglab[ http://www.datamininglab.com/pubs/Elder92_PRaGNO.pdf ], Quote Share this post Link to post Share on other sites