Semantic hashing by similarity is one way to create more sensible representations (Re: Proposition 2): http://www.cs.toronto.edu/~rsalakhu/papers/semantic_final

Semantic hashing by similarity is one way to create more sensible representations (Re: Proposition 2): http://www.cs.toronto.edu/~rsalakhu/papers/semantic_final.pdf http://www.cs.toronto.edu/~rsalakhu/papers/semantic_final.pd... This would also fit well into memory networks: http://arxiv.org/abs/1410.3916 http://arxiv.org/abs/1410.3916, SDM: http://en.wikipedia.org/wiki/Sparse_distributed_memory http://en.wikipedia.org/wiki/Sparse_distributed_memory or global workspace model: http://en.wikipedia.org/wiki/Global_Workspace_Theory http://en.wikipedia.org/wiki/Global_Workspace_Theory