Great to see your interest, Aseem! I watched the video and you propose two things. The first is predicting input bits by summing the connection weights of the predicted columns for each input bit and taking the top X bits as the predicted bits. This seems very similar to reconstruction, a prediction technique used prior to the CLA classifier. It is perhaps a bit more elegant but showed inferior accuracy from what I have heard. There is a very brief wiki page on it here: https://github.com/numenta/nupic/wiki/Reconstruction
The second thing you propose is extending TP connections to predict multiple steps into the future. This has definitely been discussed previously but I don't know what the current stance is. I will see if I can get a response from someone more familiar with it. On Thu, Aug 22, 2013 at 6:09 PM, Aseem Hegshetye <[email protected]>wrote: > Hi Jeff sir, > Its an honor for me to have a reply from you ! > You have been inspiring me for Last 4 years. I am in love with > neuroscience because of you. > Even I dont like traditional AI algorithms and mathematically designed > neural networks. > I am interested in discussing algorithms consistent in building > intelligent machines by modelling cortex. > CLA is awesome. I have also thought of some things for CLA. > > > http://www.youtube.com/watch?v=7PGFivWS3Kk&feature=c4-overview&list=UULvxEnIf98-hQuhG6oljjHg > > I dont know if its right, but i just thought it might work. > > I want to work on CLA and design algorithms that can make machines think > and predict better than humans. > No lab in my university (UTD) does this, so nupic is the only place for me > to learn. > > Regards > Aseem > > > _______________________________________________ > nupic mailing list > [email protected] > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org >
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
