Shane, To me, our difference is more on the "target" of the discussion.
I have no interest in defining neural network. What I want to do is to formalize a concrete notion of "neural network", which is general enough to cover many models, and also specific enough so its strength and weakness can be analyzed in some detail. Eventually I want to compare it with my own work, so I don't have to do the comparison with each individual models. I understand that the model I gave in the memo looks too narrow to you. Can you, or anyone else, give me an alternative? What Ben proposed is still too general for my purpose. Of course, no such model can be perfect, but we still need them to make our discussions meaningful. What I don't like in the previous discussion about neural network is: it is hard to pin down the theoretical assumptions and commitments in this research paradigm, except the most obvious ones. If everything is allowed in neural networks, then the concept means nothing. Pei On 12/18/05, Shane Legg <[EMAIL PROTECTED]> wrote: > Hi Pei, > > Most of our disagreement seems to be about definitions and choices > of words, rather than facts. > > > > > (1) My memo is not intend to cover every system labeled as "neural > network" > > > > --- that is why I use a whole section to define what I mean by "NN > > model" discussed in the paper. I'm fully aware of the fact that given > ... > > My strategy is to first discuss the most typical models of the "neural > > network" family (or the "standard NN architectures", as Ben put it), > > My problem is this: At my research institute a large portion of the people > work > on neural networks. Things like, recurrent LSTM networks for continuous > speech recognition and evolved echo state networks for real time adaptive > control problems. I also follow research on computational liquids, > biologically > plausible neural networks, neural microcircuit research, and the ideas of > people like Jeff Hawkins. In my mind, this is all work on "neural > networks", and > the researchers themselves call it that, and publish in big NN conferences > like > ICANN, IJCNN and journals like "Neural Networks". However none of this work > is like the "NN model" you have defined. Thus to my mind, your "NN model" > does not represent modern neural network research. > > > (3) Neuroscience results cannot be directly used to support > > "artificial neural networks" > > I disagree as a number of the trends and new ideas in artificial neural > networks > that I follow are coming from neuroscience research. > > If I had to sum up our differences: I'd say that what you call "standard > neural > networks" and your "NN model", and most of the problems you describe, would > have been reasonable in 2000... but not now, 5 to 6 years later. > > Shane > > > > ________________________________ > To unsubscribe, change your address, or temporarily deactivate your > subscription, please go to > http://v2.listbox.com/member/[EMAIL PROTECTED] > > ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
