Profile avatar
thomlake.bsky.social
Principal Scientist at Indeed. PhD Student at UT Austin. AI, Deep Learning, PGMs, and NLP.
6 posts 697 followers 409 following
comment in response to post
Due to the split between the inputs statements and query, the resulting model isn't a generic sequence processor like RNNs or transformers. However, if you were to process a sequence by treating each element as a new query, you'd get something that looks a lot like a transformer.
comment in response to post
MemNets first encode each input sentence/statement with a position embedding independently. These are the "memories". Finally, you encode the query and apply cross-attention between that and the memories. Rinse and repeat for some fixed depth. No for-loop over time here.
comment in response to post
The recurrence there is referencing depth-wise weight tying (see Section 2.2). > Layer-wise (RNN-like): the input and output embeddings are the same across different layers
comment in response to post
Memory networks were earlier, attention only, and had position embeddings, but were not word/token level: arxiv.org/abs/1503.08895 They were later elaborated with the key-value distinction which is, AFAIK, where this terminology arises: arxiv.org/abs/1606.03126
comment in response to post
👋