#sparsely-gated MOE, world models, & spatial MoE
8 messages · Page 1 of 1 (latest)
arxiv dot org / abs / 2211.13491v1
arxiv dot org / abs / 2301.04104v1
just dropping the dreamerv3 world model & spatial MoE papers
sparsely-gated MOE, world models, & spatial MoE
this is also a relevant discussion for the recent chatgpt + wolframalpha + langchain stack. seems like having a preexisting knowledge graph is significantly more dependable than what might crawled samples of authoritative opinion
there should probably be a pinned post for current SOTA advances in ML
just posted some thoughts here /r/MLQuestions/comments/10gair7/need_a_sanity_check_on_world_vs_spatial_moe_models/
wondering whether there's a hierarchical NN that's sparser and takes into account that some humans are better generalists/specialists and this should be weighted in the reward model