HTM for streaming analysis; TM produces Memory Error

Engineering NuPIC Community Fork
1

Hi together,

I’m trying to set up a streaming analysis system using the htm.core community fork.

Like other members of the forum, I am facing the problem that the TM does not destroy synapses. At some point, the TM consumes too much memory resulting in a Memory Error.

Two years ago, Subutai proposed randomly destroying synapses in the TM: HTM anomaly detection model too large for streaming analysis.

However, it seems that this has not been tried yet (except in Grok?).

Has anyone here had experience with online analysis and a potentially infinite number of data samples?

Thanks in advance,
Qubitza

2 Likes