Scalar encoder to SDRs


I am beginner in HTM, i have some question about scalar encoder. i watched video.

  1. I dont understand , for a binary image, how scalar encoder creat SDRs. how we know it is sematic similarity.
  2. i read some document, after overlapping, we creat active cell, and that is called SDRs.
    so what different bettween scalar encoder and overlapping.

Many thanks.


If two SDRs have semantic similarity, that means they have overlapping bits. The more overlapping bits, the more the SDRs look like each other, the more they approach the exact same representation. I’m not sure I can explain scalar encoding much better in words than I did in the HTM School video. I suggest you perhaps watch it again and read Encoding Data for HTM Systems if you haven’t yet.