Hi, 1/ out of curiosity, what's your Matlab implementation of HTM?
On Thu, Feb 25, 2016 at 1:37 AM, Yuwei Cui <[email protected]> wrote: > Hi Jianghao, > > You may find this video from Subutai useful. He talked about how to map > back to real world data given SDRs using the CLA classifier > https://www.youtube.com/watch?v=QZBtaP_gcn0 > > Also, in this recent paper on spatial pooler, the authors developed an > interesting method to reconstruct the input given the SP output (Fig. 8). > > http://arxiv.org/abs/1601.06116 > > -- > Yuwei Cui, PhD > > Research Engineer, Numenta Inc. > > On Wed, Feb 24, 2016 at 4:29 PM, Jianghao Shen <[email protected]> > wrote: > >> Hello everyone, I'm currently running HTM on Matlab, my question is after >> I transform the data in to SDR form and do the training and predicting , >> how to transform the predicted SDR back into the real world data structure? >> thanks >> > for this process you could use a so called Classifier that learns mapping of SDR to the raw input value. the implementation in Nupic is CLAClassifier. To do a real inverse you'd need a process "topDownCompute"(?) you can find an old thread and (unused) methods in encoders to do (parts of) that. Cheers, -- Marek Otahal :o)
