Just came across a the Nengo SNN library on YouTube. And it comes with a vision to neural spike encoder. Maybe we can use it on HTM too? Just a thought.
@marty1885: the sdr output of normally distributed sparse encoder looks like what Eric Laukien does with his ImageEncoder in OgmaNeo repositories. We can put this sdr directly into TM for classification or even image prediction
Sure, we can definitely do that. I just want to put the infomation out here so others intrested in the topic can see. As I’m working on implementing the Thousand Brains Theory and others making other improvments. There might be someone interested in vision and vision sequences.
I love your work.
Bump, but relevant.
Found something similar as well within Nengo, except where they’re trying to show off their LMU concept (a strong competitor to LSTM and GRU), they provide a method for flattening and encoding MNIST samples into SDRs.
Maybe we can use it on HTM too? Just a thought.