Associative Memory
- We (at MDX) have been working on associative memory for a long time.
- However, we haven't put a lot of time into it, so our progress has
been quite slow.
- We did inheritence from overlapping CAs in 2004, and they
were learned from presentation of items.
- Last summer we presented a paper on learning to solve Quillian's
question answering task (a neuro cognitive model), and the Stroop
task (static synapses).
- I've just come up with a mechanism to encode (non-plastic) a
semantic net into neurons.
- This static stuff is based on binary CAs.
- I'm hoping to use a big version of this to encode word net, and
run it on the big SpiNNaker system with learning to make a system
that resolves prepositional phrase attachment ambiguity.
- Kailash (2012) did that on our CANT simulator, but overflowed the java
heap.
