Show HN: Word2Bits – Quantized Word Vectors • Hacker News Board

Re: Show HN: Word2Bits – Quantized Word Vectors

#81

Post by rspeer » Tue, Mar 20, 2018, 3:31 PM UTC

BjoernKW wrote: post #56

rspeer wrote: post #17

It's interesting and slightly uncomfortable that the illustration for similar words uses the word "man" as an example, given the gender biases that result from learning word vectors solely from word distributions. To explain, although I’m sure the author himself is familiar with the issue: For any word that is disproportionately associated with one gender in the corpus, the model will learn that gender difference as…

In modern linguistics grammar is used descriptively rather than prescriptively, i.e. it's about describing how speakers actually use language rather than being about telling them how to use language 'properly'. If there's gender bias in certain words it might be interesting to point that out but it's not the linguists' job to 'de-bias' the grammar and the underlying model. There's an undeniable gender bias in certain…

The corpus alone can't be sufficient. The corpus displays associations that are disproportionate for how we want to use it, because corpora are from the past. But the point of ML is to build something that will be used in the future. [1]

There is no objective way to describe this difference between your training set and your test set, because you don't have data from the future. It requires a conscious decision. Enforcing that the future should be like the past is a conscious decision, and it's a lazy and unfortunate one.

This is just one instance of a problem that ML is already quite familiar with, and has been since "ML" was called "statistics": observed data produces a biased estimator of the actual distribution. You always have to correct your objectively-measured distributions for what you can't measure objectively. Here I mean "biased" in the mathematical sense; when it comes to human biases encoded in word embeddings, it also produces bias in the ethical sense. But there are many simpler instances of this.

For example: a maximum-likelihood language model assigns a probability of 0 to any word it has never seen. Maximum likelihood is, when measured, a better model of the input data than anything else. If you implement this completely objective language model, it would be useless when applied; it would output the impossible probability of 0 for most inputs. Instead you have to "smooth" and "correct" it for the fact that other words exist.

[1] https://twitter.com/random_walker/status/975700725807439879

Re: Show HN: Word2Bits – Quantized Word Vectors

#82

Post by loxias » Tue, Mar 20, 2018, 3:34 PM UTC

morelandjs wrote: post #76

tmalsburg2 wrote: post #63

Interesting to see that "science" and "fiction" are so similar according to this metric. Not surprising, though, given how often they co-occur in text, but this clearly shows the limitations of the method. The unit of interest is not really character strings but lexical entries and there can be multiple lexical entries associated with one character string. For instance, the word "bank" can mean "financial institution…

One method to partially rectify the problem you mention is to add additional preprocessing to the text to identify ngrams and add part of speech tags to words. For example [The, river, bank] might be restructed as [The, river bank (bigram)] where "river bank" is embedded as its own word vector. This can also be used to disambiguate words like "hit" which can be used as a verb and a noun. You just replace "hit" with "…

Yes! And, also, I swear I've seen this method done before. I can't remember the paper. Can anyone help?

Basically, instead of making neural word embeddings, much like you describe, the objects being embedded in a vector space were "hit|noun(1)" "hit|noun(2)" "hit|verb", and so on.

I believe they used WordNet or some ontology, or maybe a POS corpus like Penn Treebank....

Re: Show HN: Word2Bits – Quantized Word Vectors

#83

Post by patelajay285 » Tue, Mar 20, 2018, 3:35 PM UTC

creichenbach wrote: post #15

Only tangentially related, but we've recently tried to find an encoding of text that's "stable" with regards to it's characters (as opposed to stable wrt semantic meaning as here). That is, similar words (or fragments) such as "carg" and "cargo" should yield a simliar encoding. To our surprise, we couldn't find any example or description of someone doing this before. Is this such an uncommon problem or did we just no…

We recently open sourced a library that does exactly this:

https://github.com/plasticityai/magnitude

For word2vec, GloVE, and fastText. It is able to generate vectors for out-of-vocabulary words through "fragments" of words or subword character n-grams rather.

Re: Show HN: Word2Bits – Quantized Word Vectors

#84

Post by rspeer » Tue, Mar 20, 2018, 3:41 PM UTC

yorwba wrote: post #79

IAmEveryone wrote: post #70

Earlier quoted context omitted.

I’m a white, able-bodied, mostly heterosexual man, and I reject the theory that (non)discrimination is a zero-sum game.

Debiasing word vectors is not guaranteed to yield less discriminatory results in every application. In many languages, gender is expressed grammatically and there are different words for e.g. male and female scientists. If you debias the word vectors to remove associations with gender, it becomes impossible to distinguish those two. Now if you translate between two languages with that distinction, the model will have…

Ah yes. This approach to ML bias is popular in some large corporate research groups:

- Say a lot of nice things about fairness

- Talk about the importance of de-biasing and doing it as well as possible

- Show (legitimately) that the farther downstream in your ML process you apply de-biasing, the more sound the results are

- Assume that any successful ML model will be used downstream in something else

- Therefore, never de-bias anything, except the final output of something where your company benefits from showing fairness: that is, a demonstration at a conference talk about fairness

- The time to de-bias real applications is always "later" and everyone can say they are working on doing the right thing

People who release data artifacts have the ability to de-bias now instead of later. It's not perfect, but it is good. If you rely on the developer after you to do the de-biasing that you're not doing, you're ensuring that de-biasing won't happen, because they won't do it either.

Re: Show HN: Word2Bits – Quantized Word Vectors

#85

Post by yorwba » Tue, Mar 20, 2018, 4:05 PM UTC

rspeer wrote: post #84

yorwba wrote: post #79

Earlier quoted context omitted.

Debiasing word vectors is not guaranteed to yield less discriminatory results in every application. In many languages, gender is expressed grammatically and there are different words for e.g. male and female scientists. If you debias the word vectors to remove associations with gender, it becomes impossible to distinguish those two. Now if you translate between two languages with that distinction, the model will have…

Ah yes. This approach to ML bias is popular in some large corporate research groups: - Say a lot of nice things about fairness - Talk about the importance of de-biasing and doing it as well as possible - Show (legitimately) that the farther downstream in your ML process you apply de-biasing, the more sound the results are - Assume that any successful ML model will be used downstream in something else - Therefore, nev…

My point is that the appropriate debiasing is task-specific. You can completely eradicate bias one one task, but as soon as someone builds on top of your results, the bias is going to creep right back in, and likely to be worse due to the information loss in previous layers. If you just hand a bias-conscious developer a bunch of debiased word vectors with the implication that it will make their model less biased, I don't think that's really helpful. Like security, bias in machine learning needs to be addressed on the level of complete systems, it's not something you can leave to some simple input sanitizer.

Re: Show HN: Word2Bits – Quantized Word Vectors

#86

Post by loxias » Tue, Mar 20, 2018, 4:11 PM UTC

creichenbach wrote: post #66

opportune wrote: post #40

Earlier quoted context omitted.

You could modify word2vec to embed word fragments as part of the training process. You could also use stemming before training, or if you have a decent amount of computational resources you could embed trie entries with word2vec representations of word fragments and probabilistic models of the likely next character/syllable/word, which would allow you to use something like a markov process For word completion, I woul…

Our problem is not about auto-completion (we're not dealing with that much data to need sophisticated algorithms for that). What we're doing with our NN is ordering the set of results (matches) we already have. In other words, we're assigning a relevance number in [0, 1] to each result, based on the query string and training based on past user choices (clicking a result). In order to maintain some consistency and rob…

creichenbach: I believe I have an algorithm with working code, for you, but don't want to spam this thread. send me an email if you want to. (on my profile).

[Also. I hate having to spam discussion threads with personal user-to-user comments.. but there's no message user feature. This message will self destruct once it's goal has been achieved.]

Re: Show HN: Word2Bits – Quantized Word Vectors

#87

Post by jdonaldson » Tue, Mar 20, 2018, 4:12 PM UTC

loxias wrote: post #82

morelandjs wrote: post #76

Earlier quoted context omitted.

One method to partially rectify the problem you mention is to add additional preprocessing to the text to identify ngrams and add part of speech tags to words. For example [The, river, bank] might be restructed as [The, river bank (bigram)] where "river bank" is embedded as its own word vector. This can also be used to disambiguate words like "hit" which can be used as a verb and a noun. You just replace "hit" with "…

Yes! And, also, I swear I've seen this method done before. I can't remember the paper. Can anyone help? Basically, instead of making neural word embeddings, much like you describe, the objects being embedded in a vector space were "hit|noun(1)" "hit|noun(2)" "hit|verb", and so on. I believe they used WordNet or some ontology, or maybe a POS corpus like Penn Treebank....

I think you're talking about sense2vec

Re: Show HN: Word2Bits – Quantized Word Vectors

#88

Post by Maybestring » Tue, Mar 20, 2018, 4:18 PM UTC

maxlam wrote: post #25

andreyk wrote: post #20

Very cool! I like the visualizations a lot. Did you try to get an interpretation for what each quantized vector dimension means (have just skimmed, not read)? Also, I am curious why you chose to go straight to publishing on Arxiv? I am actually also in CS224N right now and have a project me and my collaborator feel is publication worthy, but our plan is to go the normal route of submitting to a conference and only pu…

Definitely tried to figure out if the dimensions mean anything -- as far as I can tell they don't really mean much :(

If you want them to be meaningful without changing the model...

Couldn't you rotate the basis to minimize the distance between each basis vector and it's nearest neighbor?

Re: Show HN: Word2Bits – Quantized Word Vectors

#89

Post by fvdessen » Tue, Mar 20, 2018, 4:18 PM UTC

rspeer wrote: post #77

fvdessen wrote: post #53

Earlier quoted context omitted.

Personally I am rather uncomfortable with the idea of 'correcting' the learning process to have the machine output what we think it should instead of what it objectively learns from the dataset. It is trading a bias against an ideal world with a bias against reality.

Where did you get the idea that data , of all things, is objective? Data is shit! Data is the noxious raw material that we have to process with great difficulty into something useful! If you are familiar at all with machine learning, you should recognize that human decisions affect every step of the process, especially the part where the data is produced and collected. It is not an oracle of objective truth. And let…

IIRC, if you remove the 'uh' and 'hmm' from audio samples to train for speech recognition, you get lesser performance than if you leave them in. So it is not that easy to differentiate a-priori between actual data and noise. If you look at the history of NLP we went from 'sentences are instances of a well defined grammar' to 'sentences are a bunch of statistically related tokens'. Embracing the mess is how we got better results !

And as for de-biasing gender terms, It makes the assumption that the gender differences in language only have nefarious purposes, but without an actual proof that this is the case, you may very well throw the baby with the bathwater.

A good example of this is when the french government mandated that Resumes must be anonymous and not mention sex, gender, etc. It actually resulted in worse outcome for people from lower socio-economic background. It turned out that on average people were more forgiving of bad Resumes if it came from people who could be expected to be disadvantaged.

Re: Show HN: Word2Bits – Quantized Word Vectors

#90

Post by fvdessen » Tue, Mar 20, 2018, 4:32 PM UTC

yorwba wrote: post #85

rspeer wrote: post #84

Earlier quoted context omitted.

Ah yes. This approach to ML bias is popular in some large corporate research groups: - Say a lot of nice things about fairness - Talk about the importance of de-biasing and doing it as well as possible - Show (legitimately) that the farther downstream in your ML process you apply de-biasing, the more sound the results are - Assume that any successful ML model will be used downstream in something else - Therefore, nev…

My point is that the appropriate debiasing is task-specific. You can completely eradicate bias one one task, but as soon as someone builds on top of your results, the bias is going to creep right back in, and likely to be worse due to the information loss in previous layers. If you just hand a bias-conscious developer a bunch of debiased word vectors with the implication that it will make their model less biased, I d…

That's because what's called 'debiasing' is not actually removing bias but adding another opposite one at the superficial layers we can actually notice.