HTM or SparseCNNs for video footage classification problem

Hello, this is my first post, and I hope not the last one. I’m here to ask for advice as I am starting my dissertation for a masters degree and I want to implement the HTM concept in it.

I am working with videos from surveillance cameras to implement the concept of Continual Semi-Supervised Learning for a classification problem, where we have 8 categories (person enters in a building, car arrives to the par, etc) that are annotated in a json file to be used as a ground truth.

The problem is to deal with unsupervised time series of data points that come for the model to be incrementally updated. My supervisor uses Hidden Markov models to deal with the unlaballed data but in my case, once I discovered the concept of HTM I felt that I needed to give it a try to improve the results.

My key point is: regarding that my level of programming is not very high, I am trying to see the differences between using HTM full implementation or using CNNs that contain some HTM concepts such as sparsity, modularity or hierarchy, which seem less hard to implement. I guess that I need to do more research but some recommendations would be highly appreciated.

Thank you very much for your attention.


Please see if this helps you,

Please share your results if you make any progress.

1 Like