Interview with Yann Lecun on what AI can do | Hacker News Reader

Interview with Yann Lecun on what AI can do

newscientist.com

59 pointssconxu21 comments

Yann gave a talk at my university last week. It's really astounding how far deep learning has come from performing classification and regression to now doing things like reasoning and generation. Does anyone else find this wild? These are tasks that historically required a tremendous amount of domain-specific engineering to perform well. And recurrent memory networks just power right on through them.

I understand that learned, dense vector representations are powerful blocks to operate on, but hot damn the trend of general systems that require less domain-specific engineering is an exciting one that blows my mind.

I definitely find it wild. Recent developments make me very optimistic that I might see useful domestic robots within my lifetime, maybe even something like Data.

It definitely seems way less far-fetched now that we have machine learning techniques for semantic segmentation, object and face recognition, reasoning, voice recognition and synthesis and even translation, extracting useful meaning from text/speech.

You know, I see many different big names in the ML field shooting for a personal 'assistant' or something like that. For whatever reason I have always thought what would be preferable would be an advocate. A being that would always have your back and be watching out for your best self and for you in general, preventing you from getting screwed by someone trying to make an information asymmetry play or trying to access your data. Somehow I worry that most big companies aren't going to be trying to create something that puts the user first. I'd love to be wrong.

I too hope that there would some Linux equivalent in the personal assistant space. It simply runs on my machine/mobile with my data, and I am totally in control of it. Currently this seems little intractable for availability of data and personal computing power but I hope these will be a lesser concern another decade in future.

Ideally there'd be a standardised api for personal assistants, so you can move between different services or swap the whole thing out for an open source alternative.

He was right: For the progress he is seeking now, neural networks stand to play at most a peripheral role. Instead, he needs some new and quite different ideas.

That the networks are neural might mean that they could simulate a worm; that he got more out of them, as, say, non-linear fitting functions, is cute but not much progress for the next challenges he has in mind.

FYI this was submitted a few days ago and no one seemed to care:

https://news.ycombinator.com/item?id=10521398

Sometimes the front page is too busy or something doesn't take. Don't worry about it - they're just internet points. No point linking to it unless there are comments that might also be interesting to readers of these ones, right?

i could careless - i was under the impression HN had URL dup algorithms to make sure you cant double submitted within a few months of the OP

I think dang said recently it was something like 8 hours. Beyond that you can resubmit. I've had cases where I've submitted something and it just takes me to the thread where someone else posted it and it automatically gives them an upvote from me.

"Unsupervised learning"?

Well, setting aside the jargon, one approach is just a statistical hypothesis test, and a lot is known about that.

Unsupervised feature extraction (eg in visual tasks with backprog) goes well beyond what would conventionally be called "a statistical hypothesis test".

For example, look at the image of the features learnt by the AlexNet CNN[1]. In a more conventional "statistical hypothesis test" these would be manually defined by hand. IMHO a system that automatically extracts those features itself deserves to be called more than "just a statistical hypothesis test"

[1] "Example filters learned by Krizhevsky et al" http://cs231n.github.io/convolutional-networks/

Clearly my remark was

> ... one approach is just a statistical hypothesis test ...

So we have no disagreement.

E.g., is this JPG an image of a human face? So, for some variables, extract data on those variables from known (supervised?) human faces and do an hypothesis test with null hypothesis that the JPG is of a human face.

Or, are these human faces like those human faces (unsupervised)? Then get data on the variables and do a two sample test with null hypothesis that the two collections of faces are the same, i.e., from the same distribution.

Well yeah.

But that is like saying that IBM Watson is just a statistical hypothesis test: "Is this answer or this answer correct?"

That is absolutely true, and completely useless.

The hard part isn't telling the difference once you've done the feature extraction, it's doing the feature extraction automatically itself.

(Actually that undersells it too, because you have so many combinations of features to evaluate that making that part work is non-trivial too. But at least that part is just calculus)

We are being too brief.

Here is a relatively non-standard approach to hypothesis testing:

https://news.ycombinator.com/item?id=10484602

So, that was looking for bias. The training was to say what people were the women.

So, could use that to say if these people were like those people. So, could get some people recognition.

Or, go to a valley with a lot of rocks and for each of 1000 rocks measure 10 quantities. Then here is one more rock; did it come from that valley? So, doing valley rock recognition.

Here are 1000 faces from North Korea and 500 from South Korea. Measure 10 quantities on each and ask if the faces are different (this would be a multi-dimensional two sample test, distribution-free).

Here's a more general picture: Take some data. Ask if the data has property X. Make a null hypothesis that the data does have property X, and with that assumption take a function of the data and calculate some probability that would be high if the data has property X. Then see what the probability is for the data do have. It that probability is low, then reject the null hypothesis and conclude that the data does not have property X. So, have now recognized that the data is not X. It may from some further knowledge that conclude that the data has property Y. This might help a self driving car decide if the stuff in the road is just some harmless paper and can drive over it or if it is dangerous and must avoid it, if the ice about to step on is solid, if that knob is to be pulled on or twisted, ..., a lot of questions in AI.

I'm being brief. Questions?

> Take some data. Ask if the data has property X.

Unsupervised learning would ask, "does this bag of data have any properties?" or "what are the most significant properties in this bag of data?" or "are there any partitions of this bag of data that seem to be more similar than others?"

If you know what you're looking for in advance then there are more direct ways to search for it (like hypothesis testing). Unsupervised learning is what you get when you relax the requirement to know in advance what it is you want to test, and relax the requirement that you can tell once you find it.

Ah, we're getting a C in communications.

> "does this bag of data have any properties?"

WOW! That's one heck of a hard question! There can be a LOT of properties! No properties might mean random. So, one case is old random number testing! People settled on the Fourier test -- essentially hypothesis testing.

You seem to be saying that hypothesis testing requires knowing something too specific about what is being looked for. Well, in some of the elementary stuff.

But there's the old distribution free two sample tests: Joe teaches calculus to 30 students, and Mary teaches the same subject to 25 students. Is their different performances significant?

Specific? We are not saying much about what is different, but we do get estimates of Type I error, that is, saying that there is a big difference when there is not. So, how? Put all 30 + 25 student scores (right, something specific, unidimensional, but see below) in a pot, stir briskly, pull out 30 and 25, that the difference in their averages, do this 1000 times, get an empirical distribution of the difference under the hypothesis that there is no difference, then see where the actual difference is in the empirical distribution and estimate false alarm rate.

For measuring several quantities not just one, can look for a significant difference even if have no idea what the heck the cause is. So, are not being specific about a cause.

Generally in making AI decisions, what to have some sense of being correct. A first cut approach is hypothesis testing.

A second cut would be, say, looking for a diagnosis. So, have some data and 20 candidate diagnoses. So, do hypothesis tests on the appropriate data for each of the 20, maybe different parts of the data for each of the 20, and pick the diagnosis based on the Type I estimates.

Hypothesis testing has a cute and powerful idea: Make an hypothesis of no effect and use that to calculate some probabilities. There's got to be a good role for that in AI.

> That's one heck of a hard question!

Exactly. That's the problem. It's not just a question of deciding whether the data is random or has some particular property, but trying to answer the question of whether there is _any_ signal buried in the data -- and then deciding whether that signal is meaningful.

For the most part, unsupervised learning is more like data pre-processing than an alternative to hypothesis testing. General frequentist statistical techniques (including hypothesis testing, goodness of fit, etc.) are an integral part of evaluating the results of unsupervised learning.

From a paper I published, go to a server farm and for three months collect data on each of 10 variables 100 times a second.

Now, in real time, given values on each of those 10 variables, ask if the farm is sick or healthy? Get known, adjustable false alarm rate and a relatively high detection rate.

So, it's a multi-dimensional, distribution-free hypothesis test.

Don't know what patterns or properties are looking for and don't get a diagnosis, but still, if get a detection at a low false alarm rate, then have good evidence the farm is sick.

So, this hypothesis test is not the same as you have described.

So, is this unsupervised learning?

Unsupervised learning is broad set of techniques that seek to infer structure from data, which is pretty fundamentally different from hypothesis testing. Where a hypothesis test asks "is <assertion> true about the structure in this data?", unsupervised techniques ask "is there any structure in this data?"

We're not communicating well. I said a little more in

https://news.ycombinator.com/item?id=10543615

just above.

Reply on news.ycombinator.com