[Adapt] A Tour Over Memory Networks
13701970825 at 126.com
Wed Oct 28 11:25:28 CST 2020
Traditional structures like encoder-decoder(or with attention) can not store too much information. Their memory (encoded by hidden states and weights) is typically too small. So we can’t accurately remember facts from the past. The problem is more severe for other tasks in vision or audio where a long-term memory is required to watch a movie and answer questions about it.
The central idea of memory networks is to combine the successful learning strategies developed in the machine learning/deeplearning literature for inference with a memory component that can be read and written to.
I will give a tour over classic memory networks, mainly focus on their ideas. I'll also make some comparations.
Time: Wed 4:00pm
Venue: SEIEE 3-414
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Adapt