Comments on: The Talented Dr. Hebb, Part 1, Novelty Filtering http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/ The Peltarion Blog Wed, 07 Jan 2009 03:21:05 +0000 http://wordpress.org/?v=2.7 hourly 1 By: Luka (Peltarion) http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/comment-page-1/#comment-25172 Luka (Peltarion) Wed, 26 Nov 2008 23:05:27 +0000 http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/#comment-25172 Yes, Synapse: http://www.peltarion.com/products/synapse Yes, Synapse: http://www.peltarion.com/products/synapse

]]> By: rey hass http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/comment-page-1/#comment-25034 rey hass Sat, 22 Nov 2008 23:37:10 +0000 http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/#comment-25034 is there a stimulator that can facilate this? Thanks, Rey is there a stimulator that can facilate this? Thanks, Rey

]]> By: Synaptic » Blog Archive » The talented Dr.Hebb, Part 2, PCA http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/comment-page-1/#comment-19 Synaptic » Blog Archive » The talented Dr.Hebb, Part 2, PCA Mon, 19 Jun 2006 23:20:57 +0000 http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/#comment-19 [...] Besides novelty filtering that was covered in Part 1, there is another interesting function that a Hebbian Layer can perform and it is called Principal Component Analysis (PCA). This time we are going to take a closer look at PCA and see how it can be used in Synapse in combination with regular neural networks.PCA is a linear transformation that can be used to reduce, compress or simplify a data set. It does this by transforming the data to a coordinate system so that the greatest variance of the data by a projection of the data ends up on the first component (coordinate), the next one in line on the magnitude of variance ends up on the second component and so on. This way one can choose not to use all the components and still capture the most important part of the data. [...] [...] Besides novelty filtering that was covered in Part 1, there is another interesting function that a Hebbian Layer can perform and it is called Principal Component Analysis (PCA). This time we are going to take a closer look at PCA and see how it can be used in Synapse in combination with regular neural networks.PCA is a linear transformation that can be used to reduce, compress or simplify a data set. It does this by transforming the data to a coordinate system so that the greatest variance of the data by a projection of the data ends up on the first component (coordinate), the next one in line on the magnitude of variance ends up on the second component and so on. This way one can choose not to use all the components and still capture the most important part of the data. [...]

]]> By: Luka (Peltarion) http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/comment-page-1/#comment-11 Luka (Peltarion) Mon, 15 May 2006 16:29:39 +0000 http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/#comment-11 Anna, it is difficult to answer as the two most important things are the selection of input data and the interpretation of what constitutes "change" detected by the novelty filter. Both these things require domain knowledge (in this case in oncology) which I'm afraid I don't have. In theory your idea makes perfect sense, but in practice you will know better if the results are useful. A more conventional approach would be to use a regular static supervised neural network to perform a classification task with patient data as input and health as desired output. - JoLex, Giuseppe et al, sure we can include a stock market data example in part 2 of the Hebbian tutorial. Anna, it is difficult to answer as the two most important things are the selection of input data and the interpretation of what constitutes “change” detected by the novelty filter. Both these things require domain knowledge (in this case in oncology) which I’m afraid I don’t have. In theory your idea makes perfect sense, but in practice you will know better if the results are useful.

A more conventional approach would be to use a regular static supervised neural network to perform a classification task with patient data as input and health as desired output.

-

JoLex, Giuseppe et al, sure we can include a stock market data example in part 2 of the Hebbian tutorial.

]]> By: Anna M http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/comment-page-1/#comment-10 Anna M Mon, 15 May 2006 15:16:58 +0000 http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/#comment-10 Luka, I was wondering if this could be applied as a method of detecting anomalies in brain tumor data? We have large data sets describing healthy as well as damaged brain tissue and it is of interest to map it to patient data. What I am thinking of is that we could adapt an anti-Hebbian to data from healthy tissue and patients and then see if there is a match when it is applied to damaged brain tissue and affected patients. Does that make sense? Anna Luka, I was wondering if this could be applied as a method of detecting anomalies in brain tumor data? We have large data sets describing healthy as well as damaged brain tissue and it is of interest to map it to patient data.

What I am thinking of is that we could adapt an anti-Hebbian to data from healthy tissue and patients and then see if there is a match when it is applied to damaged brain tissue and affected patients. Does that make sense?

Anna

]]> By: Miguel Perez http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/comment-page-1/#comment-9 Miguel Perez Mon, 15 May 2006 14:31:19 +0000 http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/#comment-9 I would be interested in a practical example too. Thanks. Miguel I would be interested in a practical example too. Thanks.

Miguel

]]> By: Giuseppe http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/comment-page-1/#comment-8 Giuseppe Mon, 15 May 2006 13:26:09 +0000 http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/#comment-8 Trading not different to any other data. Hebbian is static, so don't worry about time series. Example of hebbian for trading is easy to make. Perhaps Peltarion can include example in the next part? Hebbain PCA is also relevant to trading. --G.C Trading not different to any other data. Hebbian is static, so don’t worry about time series. Example of hebbian for trading is easy to make. Perhaps Peltarion can include example in the next part? Hebbain PCA is also relevant to trading.

–G.C

]]> By: JoLex http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/comment-page-1/#comment-7 JoLex Mon, 15 May 2006 09:49:57 +0000 http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/#comment-7 Could you give us an easy-to-understand example of the hebbian applied to trading data? I understand in theory, but a practical example would be great. Could you give us an easy-to-understand example of the hebbian applied to trading data? I understand in theory, but a practical example would be great.

]]> By: Luka (Peltarion) http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/comment-page-1/#comment-6 Luka (Peltarion) Mon, 15 May 2006 08:37:54 +0000 http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/#comment-6 Richard, yes, Hebbian novelty filters could without any problems be applied to stock market data. The trick is as always choosing the right input features. Also, one should be aware that a novelty filter reacts to unseen patterns in input data, which doesn't say how the system will react to it. David, glad to hear that you found it useful. As long there is interest in the blog, we'll try to keep it updated with news and tutorials on a regular basis. Richard, yes, Hebbian novelty filters could without any problems be applied to stock market data. The trick is as always choosing the right input features. Also, one should be aware that a novelty filter reacts to unseen patterns in input data, which doesn’t say how the system will react to it.

David, glad to hear that you found it useful. As long there is interest in the blog, we’ll try to keep it updated with news and tutorials on a regular basis.

]]> By: David Loraén http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/comment-page-1/#comment-5 David Loraén Mon, 15 May 2006 06:24:00 +0000 http://blog.peltarion.com/2006/05/11/the-talented-dr-hebb-part-1-novelty-filtering/#comment-5 Thanks! Didn't know about the Hebbian before. It took us five minutes to put it to great use in finding changes in the deforestation data we've been modeling. I hope you'll find time to keep this blog up, because this kind of information is enormously useful for us who do not have PhD's in adaptive systems. Thanks! Thanks! Didn’t know about the Hebbian before. It took us five minutes to put it to great use in finding changes in the deforestation data we’ve been modeling.

I hope you’ll find time to keep this blog up, because this kind of information is enormously useful for us who do not have PhD’s in adaptive systems. Thanks!

]]>