[HN Gopher] Text Embeddings Reveal (Almost) as Much as Text
       ___________________________________________________________________
        
       Text Embeddings Reveal (Almost) as Much as Text
        
       Author : jxmorris12
       Score  : 48 points
       Date   : 2023-07-24 18:03 UTC (4 hours ago)
        
 (HTM) web link (openreview.net)
 (TXT) w3m dump (openreview.net)
        
       | Legend2440 wrote:
       | I think this is unsurprising, the point of embeddings is to
       | encode the information from the text. It's not encryption or
       | hashing; it's a compressed representation.
       | 
       | Even if you couldn't recover the original words I would expect to
       | be able to recover equivalent words with the same meaning.
        
       | Imnimo wrote:
       | For the experiment in section 5.3, where they try to recover
       | private information from embeddings of clinical notes, it's
       | interesting that the model has to also spend capacity trying to
       | reconstruct non-private information. I wonder if you could do
       | better at recovering names by first learning a custom distance
       | metric that tries to assign a low distance to embeddings of texts
       | that share a name, regardless of other content, and then using
       | this method to minimize that distance.
        
       | RC_ITR wrote:
       | This is interesting, but said differently
       | 
       | 'when we build models to do a really good job of representing 32
       | words/tokens as vectors, you can very easily backsolve for 28/32
       | words _just_ using the vectors. These results are not robust
       | above 32 words. '
        
       | fmeyer wrote:
       | Not a single reference to differential privacy?
        
       | autokad wrote:
       | is there any kind of embedding that does protect the privacy of
       | the initial words?
        
       | majormajor wrote:
       | If you recover 90% of text _exactly_ does that mean the
       | "semantic overlap" of the vectors isn't as good as it could be?
       | E.g. semantically identical but textually different words cause
       | more meaningful shifts than would be desired for certain use
       | cases?
        
       | sp332 wrote:
       | The security aspect is important to keep in mind, but a more
       | interesting use case for me is tweaking the embedding values and
       | finding token inputs that correspond. That will let me explore
       | the latent space.
        
       ___________________________________________________________________
       (page generated 2023-07-24 23:01 UTC)