Gregory's Server@yogthos there’s some of the same ideas from his GLOMM paper in here, especially the contrastive positive/negative training and analogy to sleep.
I need to read it again in more detail (and again, and again), but I’ve been really enjoying his recent thought experiments here.
|
Top-level
2 comments
@yogthos yeah, neat stuff! Between the other feed-forward ideas like the synthetic gradient paper, and the hierarchical GLOM proposal, there’s lots of neat potentials to explore for alternative model structures. |
@erikavaris yeah there's been a lot of exciting developments recently