ALBERT: A Lite BERT for Self-Supervised Learning of Language Representations
ai.googleblog.com
ai.googleblog.com
Really impressed at the speed with which Hugging Face ported this to their transformers library -- Google released the model and source code Oct 21 [1] and it was available in the library just 8 days later [2].
[1] https://github.com/google-research/google-research/commit/b5...
[2] https://github.com/huggingface/transformers/commit/c0c208833...
Another interesting finding is that this speeds up training due to the smaller number of parameters.
However, my worry is that those of use that do not readily have access to TPUs will get even slower models when using CPUs for prediction due to the additional and wider hidden layers. (Of course, one could use ALBERT base, which still have 12 layers and a hidden layer size of 768, at a small loss.) Did anyone measure the CPU prediction performance of ALBERT models compared to BERT?
Edit: I guess one solution would be to use a pretrained ALBERT and finetune to get the initial model and then use model distillation to get a smaller, faster model.
In section 4.8 they compare accuracy at the same amount of training time for the biggest of each model and show that ALBERT is substantially better.
IME this is the way to go; take an ensemble of big, accurate models, then distill them down to the smallest model you can get away with. There are really good tricks in this paper
It's common knowledge not to bother running a deep model like ALBERT, BERT, or XLNet, etc without at least a GPU. Training and inference with models of this size on CPU's is typically considered to be intractable.
Obviously, this paper is from Google, so they have free access to TPU's, which is arguably optimal for an arbitrary TensorFlow model in terms of performance, but if you don't have access to TPU's, the clear choice is to use GPU's instead. It won't be quite as fast, but it is sufficient and GPU's are relatively cheap and available.
It's pretty common to run inference on CPUs. There are lots of operational and cost reasons why this makes sense in at least some cases.
I am not sure if this is common knowledge, and if it is, it is wrong. With read-ahead and sorting batches by length, we can easily reach 100 sentences per second on modern CPUs (8 threads) with BERT base. We use BERT in a multi-task setup, typically annotating 5 layers at the same time. This is many times faster than old HPSG parsers (which typically had exponential time complexity) and as fast as other neural methods used in a pipelined setup.
It's worth noting that TPUs are available for free on Google Colab.
Yes, and you can also get a research grant, which gives you several TPUs for a month. But that does not mean that you can easily deploy TPUs in your own infrastructure, unless you use Google Cloud and suck up the costs (which may not be possible in academia).
This summary is wrong.
It's true that they use parameter sharing to reduce the number of parameters.
But for any given "class" of model (base/large/xlarge) ALBERT has the same size hidden layer and the same number of layers.
If you try to compare by model size (measured by number of parameters) then ALBERT xxlarge (235M parameters) has less layers than BERT large (334M parameters) - 12 vs 24 - a larger hidden layer (4096 vs 1024) and smaller embedding size (128 vs 1024).
Additionally, in section 4.9 they compare more layers and find "The difference between 12-layer and 24-layerALBERT-xxlarge configurations in terms of downstream accuracy is negligible, with the Avg score being the same. We conclude that, when sharing all cross-layer parameters (ALBERT-style), there is no need for models deeper than a 12-layer configuration"
Huggingface did just that: https://medium.com/huggingface/distilbert-8cf3380435b5
This is an amazing result.
RACE has a performance ceiling of 94.5% set by inaccuracies in the data. Mechanical Turk performance is 73.3%.
It's a hard, reading comprehension test where you can't just extract spans from the text and match against answers. Section 2 of https://www.aclweb.org/anthology/D17-1082.pdf has a sample.
I'd contend that, for the most part, it doesn't matter. It's a bit like the whole ML vs AGI debate ("but ML is just curve fitting, it's not real intelligence"). The more pertinent question for human society is the impact it has - positive or negative. ML, with all its real or perceived weaknesses, is having a significant impact on the economy specifically and society generally.
It'll be little consolation for white collar workers who lose their jobs that the bot replacing them isn't "properly intelligent". Equally, few people using Siri to control their room temperature or satnav will care that the underlying "intelligence" isn't as clever as we like to think we are.
Maybe current approaches will prove to have a cliff-edge limitation like previous AI approaches did. That will be interesting from a scientific progress perspective. But even in its current state, contemporary ML has plently scope to bring about massive changes in society (and already is). We should be careful not to miss that in criticising current limitations.
Word. I think we’re actually at a level now that we’ll soon start questioning how intelligent people really are, and how much of human intelligence is just an uncanny ability to hide incompetence/lack of deeper comprehension.
(Of course we’re a hell of a long way from A.I. with deep comprehension, and may remain so for hundreds of years. It’s impossible to predict that kind of quantum leap IMHO.)
> Humans Who Are Not Concentrating Are Not General Intelligences
https://srconstantin.wordpress.com/2019/02/25/humans-who-are...
Going back to Turing, the argument was for true, human creativity. The claim was that there is no theoretical reason a machine cannot write a compelling sonnet.
After spending the better part of a century on that problem, we have made essentially zero progress. We still believe that there is no theoretical reason a machine cannot write a compelling sonnet. We still have zero models for how that could actually work.
If you are a non-technical person who has been reading popular reporting about ML, you might well have been given the impression that something like GPT2 reflects progress on the sonnet problem. Some very technical people seem to believe this too? Which seems like an issue, because there’s just no evidence for it.
Maybe a larger/deeper/more recurrent ML approach will magically solve the problem in the next twenty years.
And maybe the first machine built in the 20th century that could work out symbolic logic faster than all of the human computers in the world would have magically solved it.
There was no systematic model for the problem, so there was no reason to conclude one way or another, just as there isn’t any today.
ML is a powerful metaprogramming technique, probably the most productive one developed yet. And that matters.
It’s just still not at all what we’ve been proposing to the public for a hundred years. To the best of our understanding, it’s not even meaningfully closer. And that matters too, even if Siri still works fine.
I dunno, I heard music composed with neural nets that is above what the average human could achieve [1]. Not on par with the greatest composers, but over average human level.
In the same line of thought, I have seen models do symbolic math better than automatic solvers, generate paintings better than average humans could paint, even translate better than average second language learners.
I would rate current level in AI at 50% of human intelligence on average, and most of that was accomplished in the recent years.
[1] Google Music Transformer - https://magenta.tensorflow.org/music-transformer
That's not true. The Turing test is that one can't tell the difference between a human and a machine intelligence by communicating with it. That's it.
The claim was that there is no theoretical reason a machine cannot write a compelling sonnet.
And that's absolutely not true. I can't write a compelling sonnet.
If you are a non-technical person who has been reading popular reporting about ML, you might well have been given the impression that something like GPT2 reflects progress on the sonnet problem. Some very technical people seem to believe this too? Which seems like an issue, because there’s just no evidence for it.
I work in the field of NLP and I believe it does reflect progress, and I think there is evidence for it.
The gods are they who came to earth
And set the seas ablaze with gold.
There is a breeze upon the sea,
A sea of summer in its folds,
A salt, enchanted breeze that mocks
The scents of life, from far away
Comes slumbrous, sad, and quaint, and quaint.
The mother of the gods, that day,
With mortal feet and sweet voice speaks,
And smiles, and speaks to men: "My Sweet,
I shall not weary of thy pain."
For the youth, who, long ago,
Came up the long and winding way
Beneath my father's roof, in sorrow—
Sorrow that I would not bless
With his very tears. Oh,
My son the sorrowing,
Sorrow's child. God keep thy head,
Where it is dim with age,
Gentle in her death!
Both samples from https://www.gwern.net/GPT-2
This is exactly it. Every previously seen AI Winter had in common that funding was cut back. However, Google or other companies in the realm could approach a point where further investment wouldn't make sense to them. Until then there won't be a Winter, maybe Autumn, because smaller players disappear.
I found this sentence particularly intriguing given that John Carmack recently announced that he was switching his main focus to AI.
I've direct evidence of that from my day job (building NLP chat bots for Intercom).
That business value will increase as we NLP progresses, even if we're moving towards a local optimum.
Even if we do get stuck, real products and real revenue powered by NLP will help fund research on successive generations.
Of course theres tons of hype about AI. But theres also a big virtuous cycle which just wasn't present in the setup which created previous AI winters.
The current A.I. B.O.O.M is due to end or it is ending already, but this only means now we are equipped with really powerful approximators previous generation of researchers would not even dream of, that we left us with a really tantalizing question:
What is the right question to ask?
We have undoubtedly proved machine are superior to fit, now we need to make them curious.
The open question is whether AGI is the same as Schmidhuber's optimum, or even lies within Schmidhuber's basin.
[0] Cambridge style debate on the topic at Neurips 2019.
This is an unnecessarily uncharitable view of academia.
"Outside the box thinking" is frequently just ignorance and Dunning-Kruger.
Making progress of NLP benchmarks? Must be a sign that we're moving even closer to an even longer and more bitter AI winter.
ML is an empirical science, or a craft if you want, with useful applications. It's not the ultimate theory of intelligence.
Progress is always that way. It plateaus, then suddenly jumps and then plateaus again.
If you complaint is about the general move away from statistics and deep learning becoming the norm, then there are a pretty decent number of labs who are working on coming up with whatever the next deep learning is. There is probabilistic programming and there models are some models with newer biologically inspired computation structures.
Even inside ML and deep learning, people are trying to come up with ways to better leverage unsupervised learning and building large common sense representations of the world.
There is certainly an oversupply of applied deeplearning practitioners, but there are other approaches being explored in the AI/ML community too.