What are the limits of deep learning?

https://www.pnas.org/content/116/4/1074

by pizza • 8 years ago

102 64 8 years ago

"NNs that rely on backprop only really work on continuous functions and many interesting things like language[0] are not products of those" is real high on my list of limits.

[0] https://www.linkedin.com/pulse/google-hyping-why-deep-learni...

evrydayhustling6• 8 years ago

State of the art on most NLP tasks is now provided by deep learning, thanks to a host of techniques for dealing with both categorical and structurally complex data, such as LSTMs. Here's a great resource for keeping up with development in that area: https://github.com/sebastianruder/NLP-progress

yorwba3• 8 years ago

That argument seems to be based on the idea that because written language uses a discrete encoding, it's impossible to use continuous functions. But as soon as you replace deterministic discrete values by continuous probabilities of those values, you're back in the space of continuous functions. Even non-NN approaches to NLP use that representation, because optimizing continuous parameters is something we can do very well. Discrete optimization is much harder.

YeGoblynQueenne10• 8 years ago

And how well do all those techniques work when it comes to representing meaning, rather than structure?

evrydayhustling6• 8 years ago

Yet another piece that cherrypicks AI failures in order to make vague thesis that "deep learning is not enough". I can't decide if it makes the article more or less honest that the "Beyond Deep Learning" section is essentially a catalog of... recent developments in Deep Learning. The need to (selectively) pidgeonhole and then judge (on impossible to operationalize criteria) a fast-evolving field is purely rhetorical and, worse, aimed at folks who don't understand the way the field is progressing.

I've got nothing against raising platforms for alternative lines of AI research, especially including highlighting the ways they complement current deep learning But this article format is so transparent and tired.

eli_gottlieb3• 8 years ago

If you actually read the papers they cite, you'll notice they bring in probabilistic programming as well as other alternatives to deep learning.

evrydayhustling6• 8 years ago

I didn't say they didn't! Yes, they "bring in" probabilistic programming, graphical representations of symbolic structures, and more. But all of the citations 6-10 are from papers that integrate new approaches into a deep neural model. Most are from self- described deep learning teams, and all would be welcome at conferences and publications within the deep learning community.

Integrating with other inference paradigms has been a major driver in neural network research for 30y... so trying to cast progress like this as some sort of comeuppance that proves the insufficiency of the field is BS.

dmreedy2• 8 years ago

The criteria certainly are impossible to judge currently. Which makes your assertion that the field is 'progressing' strange.

I think there's a lot of merit to remembering the last time we got this excited about a given model for AI. And the time before that. And the time before that.

Go back further, to the time that we were all so excited that we could model the entire world with a little bit of beautiful logic. And the subsequent disappointment.

There is so much nonsense out there, so much hand-waving, mixed signals, false promises, downright naivety, and both malicious and completely inadvertent miscommunication. A little rigor in a field this excitable would be nice, I say.

YeGoblynQueenne10• 8 years ago

Unfortunately, when rigor crawls in, excitement flies out. Rigorous machine learning would be really boring for everyone except its most dedicated followers. It's probably how Geoff Hinton and his students soldiered on with their research through alternating waves of lay hype and disappointment.

RivieraKid2• 8 years ago

Which makes your assertion that the field is 'progressing' strange.

Well, it is progressing. That assertion is correct.

evrydayhustling6• 8 years ago

The progress is well defined. Both academic and industrial deep learning groups aggressively define challenge tasks in vision, NLP, relational learning and more, then document progress on them. There is even an unprecedented release of code and datasets to let others replicate and extend work.

Detractions like your "so much nonsense out there" pick straw man goalposts - whether it's AGI (whatever that really means) or "the craziest thing I heard from IBM marketing". The fact that some people are talking about crazy milestones doesn't mean that there aren't real ones being met.

RivieraKid2• 8 years ago

Yeah, I started typing a comment trying to say something like that but was unable to put it into words.

I've only skimmed through the article but in general, these criticisms of DL are annoying, they rarely bring new or valuable information. It's usually a mix of disproving claims nobody is making (e.g. that DL is close to general AI) or stating something that's obvious.

endymi0n2• 8 years ago

I think the current critique of Deep Learning as a technique is a dead end. Now I'm only a semi-professional in a sea of recent development and research, but if you ask me, the technique itself can take us all the way to General Intelligence and more.

What's missing from this discussion is how we humans don't just use a single, vision based "deep net" in order to identify the banana, we use lots of very different deep nets that run in parallel and cross-validate each other.

We have a vision net like computers that can be fooled with the same tricks (just google "optical illusion").

But a second net does three-dimensional inference of the banana shape from stereo vision and cross-validates that with a toaster. If in doubt, unnoticed parallax movement will be commanded to our body in order to get a better shape estimation.

Another net parses the content. A banana in a jungle or on a table would be highly plausible. A banana in an operating room or on a roller coaster would be implausible.

Another net parses context and timeline. If no banana was there and no banana has been incoming through our continuous estimated 3d space, there is probably no banana there.

There's smell, taste, all kinds of things running in parallel that continuously cross-validate our perception of this being a banana, probably a lot of different models just from vision alone that feed back into the main classification thread.

We'll even happily do looped cross validation if our eyes rest on a scene. Our memory will fetch reference data for all first order predictions and start subconciously processing this into the main prediction. It will predict how a banana feels, how soft it is, how warm. If we see a banana on lava that is not burning, our logic and reasoning will challenge the prediction. All of this happens subconciously, and there's no magic behind.

If you ask me, there are no limits of deep learning. There's just still a limit of imagination in feature extraction and processing.

dmreedy2• 8 years ago

If there's one lesson to take away from our history of studying the brain, the phenomenon of human intelligence, and the halting 'progress' of AI, it's that it always seems as easy as you propose. At first.

tabtab6• 8 years ago

It could be like Turing Equivalency in that deep nets of deep nets may be able to do the job, but are not the most efficient and/or easiest AI for humans to tune and tame.

I'm not saying gluing bunches of deep nets together is bad R&D, only that it may not be the only or best AI game in town. I suppose if we got a wad of deep nets to give General Intelligence (including "common sense"), but it took a big room of servers, that may be a good enough springboard, but I'm not sure putting all our eggs in the deep net basket is the best route to fuller AI just because it's been the most promising up to this point.

It's roughly comparable to spending all R&D perfecting vacuum tubes in the 1940's because they were the best known route back then to general digital electronics. The best starters are not always the best finishers.

cr0sh2• 8 years ago

It's roughly comparable to spending all R&D perfecting vacuum tubes in the 1940's

Vacuum tubes continued to be improved all the way up into the early 1960s, long after transistors had taken hold:

https://en.wikipedia.org/wiki/Compactron

One in particular was used mainly in television sets, all the way up into the early 1970s:

https://en.wikipedia.org/wiki/Nuvistor

...it competed with transistors, and was almost as small as some of its contemporary transistor cousins.

Even today, you can't rule out the vacuum tube - for some things, there isn't any substitutes:

http://www.nutsvolts.com/magazine/article/vacuum-tubes-for-t...

...and new ideas for the old technology are being investigated:

http://www.sciencemag.org/news/2012/05/return-vacuum-tube

vokep• 8 years ago

Agreed. Are you working on anything related to this?

randcraw3• 8 years ago

It's far too early to render a judgment on the long term potential of DL toward building Commander Data, both pro and con. But after some 60 years of AI research we understand the problem space pretty well, and we can see also that DL has shown little or no promise toward serving many of the component processes necessary to deliver General AI/AGI.

(And just because you can federate and combine DL networks like you can NAND gates doesn't mean they are turing equivalent to the human brain. It's fundamental to claims of such turing-AI-equivalency that we have absolutely no idea yet how to do this -- making yet another AI technique likely to suffer the same fate as all preceding AI methods in failing to solve general AI problems, that is, a failure to generalize and scale up beyond serving a specific narrow task, AKA brittleness.)

For instance, Gary Marcus at NYU has addressed the present limits of DL in terms of cognition in several papers and talks:

https://arxiv.org/abs/1801.00631

https://medium.com/@GaryMarcus/the-deepest-problem-with-deep...

Likewise, Doug Hofstadter has been critical of using DL to automate machine translation's semantic nuances:

https://qz.com/1088714/qa-douglas-hofstadter-on-why-ai-is-fa...

At present, it's still unclear how DL could surmount single-task brittleness to address these shortcomings, much less diversify into the use of logic or the modeling of concepts, their dependencies, causality, explain itself, and much more. Unbridled faith in DL's ability to jump these sharks should be tempered by the AI community's past mismeasures of earlier AI tech that we were sure would conquer everest over the years, but in fact fell far short (e.g. perceptrons, expert systems, backprop-driven shallow nets, and oh so much more).

cr0sh2• 8 years ago

One great problem - and I am but a hobbyist at best - is that of back-propagation, which is the foundation of deep learning neural networks.

As far as I know of (and I could be wrong) - there has yet to be any kind of proven method that the human brain uses any form of back-propagation for learning. From what I understand, there isn't a biological equivalent.

Which may be ok, in the long run - after all, we fly using fixed wings and propellers, and not flapping wings (as was tried over and over in the history of flight, with little to no progress and numerous failures). That is, success can be had by not replicating nature exactly.

ANNs do this at one level - neurons in the network are simplified mathematical representations. But the learning portion seems to add more complexity; the backpropagation algorithm is simple in concept, but seemingly more complex than what the human brain and natural neurons do for learning.

We also seem to learn things faster; natural neural networks - brains - regardless of species - only require relatively few examples to learn and generalize from, while deep learning artificial neural networks require thousands and more, of labeled examples.

These two issues - the need for backpropagation for learning, and the need for thousands and thousands of labeled examples exhaustively run through - both have resulted in machines and systems of hardware and software of immense complexity and power usage. We haven't even created the equivalent of a mouse brain, and if we did, it would take the energy output of Niagara Falls to run it.

From what I understand, this is also (in broad terms) Geoffrey Hinton's position as well; that our current systems, as well as they do perform (and do they - I'm still amazed at some of the things I saw my simplified implementation of NVidia's End-to-End model do to drive a simulated car around a track), are probably not the right implementation for going forward, and we should step back and find another way (Hinton's current thing is something called "Capsule Networks" which I haven't looked into too deeply - from what I have seen of them, they still appear overly complex, but again, that is just my hobbyist opinion, and certainly carries no weight here).

As a hobbyist in this field, I don't have any solutions to these issues, but they are something that keeps me thinking: How has nature actually done it? How does is it able to be done with so little energy input (beyond scale differences)? How does it work with so few training examples?

mettamage• 8 years ago

Hobbyist here as well.

As far as I know of (and I could be wrong) - there has yet to be any kind of proven method that the human brain uses any form of back-propagation for learning. From what I understand, there isn't a biological equivalent.

Is there research done on how individual or a small group of neurons learn?

We also seem to learn things faster; natural neural networks - brains - regardless of species - only require relatively few examples to learn and generalize from, while deep learning artificial neural networks require thousands and more, of labeled examples.

When you take the point of your parent's post, then I'd assume having multiple neural nets learning at the same time and seeing things from different perspectives then we might get there as well.

We haven't even created the equivalent of a mouse brain, and if we did, it would take the energy output of Niagara Falls to run it.

Our brain is specialized architecture. Emulating a brain on a Von Neumann machine designed computer is similar to a brain trying to emulate a computer. To me it makes sense that you'd need specialized hardware for this.

I wonder if we could simulate a neural net of a simple creature, like a worm.

visarga• 8 years ago

neurons in the network are simplified mathematical representations.

That is the important difference. A biological neuron is more like a cell in the cell automaton (Conway's game of life). It is a dynamic process, even if the dynamics change by a small factor, it could have a very different behaviour. Artificial neurons on the other hand are much more uniform and lack the time dimension. If you account for dynamics, then learning in the brain is a completely different thing than learning in neural nets. It has been proven that Hebbian learning (local plasticity of synapses) approximates gradient based learning, though in simulation it attains lower accuracies.

Another observation was that biological neural nets don't have a mechanism to communicate gradients back. But a deep learning experiment proved you can have fixed random back connections and the network will still learn. They don't actually need to be symmetrical connections.

The real mystery is about loss functions. Such a loss function would have to be computed in a distributed fashion in the brain. They are the origin of the learning signal. The brain has many such loss functions that have been discovered by evolution. We have not understood loss functions well enough, and understanding them would be a huge leap forward.

Another important observation is that the brain learns from the environment, not a fixed training set. Environments are richer sources, and our current crop of AI agents are playing all sorts of games, but they never have the same complexity as real world. Btw, tomorrow DeepMind will make a Starcraft demo.

scottlocklin2• 8 years ago

I keep saying this to neural net enthusiasts: brains and neural nets don't have anything in common other than some nouns. Not network topology, not even level encoding (bio neural nets are pulse rate encoded -closest "neural net" to this is something like a liquid state machine, which is not easily simulated on a Von Neumann computer). We actually know an awful lot about how things in the visual cortex work. Nothing like a "deep network."

gdubs• 8 years ago

To add to this, I think the critique in this article greatly underestimates how much data a child is exposed to by the point they can identify a cow. Yes, you can teach a two year old what a cow is without training them on 10k cow images — but they have developed deep neural connections by this point in their lives, generalizable on all sorts of problems.

YeGoblynQueenne10• 8 years ago

> We have a vision net like computers that can be fooled with the same tricks (just google "optical illusion").

I don't know of any optical illusions where a bus turned upside down is perceived as a plowshare, or an image of an elephant superimposed on the image of a room causes the other objects in the room to be misidentified.

Human optical illusions reveal the structural bias in the way we see, but that has nothing to do with adversarial examples and overfitting to a few pixels.

twtw• 8 years ago

a bus turned upside down is perceived as a plowshare

If I'm thinking of the same picture you are, it's a schoolbus on its side on a snowy road that is misidentified as a snowplow.

FWIW, that picture does look a lot like a snowplow to my eye. Big yellow rectangle against road cleared of snow? If I glanced at it, I might have said it's a snowplow.

endymi0n2• 8 years ago

I don't know of any optical illusions where a bus turned upside down is perceived as a plowshare, or an image of an elephant superimposed on the image of a room causes the other objects in the room to be misidentified.

I wouldn't be so sure about that actually. Observing my daughters, I'm sometimes extremely surprised about how they identify objects. As you grow, your concious mind probably simply does not notice the continuous cross-validation happening.

Also, at the severe augmentation levels of years of ~50 fps "live video" training data, I can imagine there's another compensation algorithm at work in humans that tries to logically orient any picture into a most common, "upright" fashion before even trying to identify anything.

Verdex3• 8 years ago

I'm not so sure. We have the entire industry of magicians devoted to making some pretty ridiculous illusions. Everyone knows it's fake on a conscious level. But the fact that everyone applauds this thing they know is fake is a pretty good proof that they are very impressed on a subconscious level.

The human mind has a lot more potential inputs AND it also tends to have years or decades of experience to draw on as well. Give the AI a quadcopter to fly around in and a huge database and it's going to go investigate the mislabeled data and figure out that it was misled. Similarly people walk towards mirages looking for an oasis.

hnick2• 8 years ago

Stay awake long enough though and human image processing gets really messy. I personally mis-identify things as people or spiders when scanning a scene. My brain jumps to a conclusion before I have time to double check and refine what I see.

I wonder what the mental model for that is - I've always assumed the brain iterates and refines but short-circuits when we're tired since processing is slower/limited.

sifoobar• 8 years ago

I wish more so called experts and professionals would put serious effort into meditation and/or quantum physics.

There's plenty more to consciousness than statistics and pattern matching, self awareness is a big one; and adding more layers of the same and/or curating better data sets isn't going to solve the problem.

Of course there are limits, it's statistics and pattern matching and has very little to do with intelligence. Not to speak of risks, since it will keep failing in surprising ways and never be able to explain why.

scott_s• 8 years ago

The "Beyond Deep Learning" section of the article talks about this approach:

A more radical possibility is to give up trying to tackle the problem at hand by training just one big network and instead have multiple networks work in tandem. In June 2018, the DeepMind team published an example they call the Generative Query Network architecture (7, http://science.sciencemag.org/content/360/6394/1204?ijkey=01...), which harnesses two different networks to learn its way around complex virtual environments with no human input. One, dubbed the representation network, essentially uses standard image-recognition learning to identify what’s visible to the AI at any given instant. The generation network, meanwhile, learns to take the first network’s output and produce a kind of 3D model of the entire environment—in effect, making predictions about the objects and features the AI doesn’t see. For example, if a table only has three legs visible, the model will include a fourth leg with the same size, shape, and color.

tabtab6• 8 years ago

If we want "dissectable" AI, we will almost certainly have to break them up into modules or semi-independent sub-systems. While one-big-network may technically be able to do the job, it's harder for most humans to study, tune, and tame such a system. We like to know who to sue if bots go kaflooey :-)

alexandernst• 8 years ago

I'm not an IA expert in any way; I just merely started learning some basic stuff, but so far the only thing I see from all the algorithms I have covered so far is that IA, deep learning, ML and so on are just statistical databases or sets, so to call them. They don't learn anything, they just repeat. From that I can say that IA is limited only to the data you can label and then feed it.

cwkoss• 8 years ago

The limit of deep learning is the quantity and accuracy of training data. For some problems, you can't run 10^x trials: you can synthesize training data but it may be of questionable real-world relevance and lead to overfitting of the generation algorithm.

randcraw3• 8 years ago

The other big limit to DL is its inability to learn a new concept from a single training example, AKA one-shot learning.

Given that humans do most of their supervised learning this way underscores that, even though DL nets intriguingly resemble the architecture of the brain, in fact, DL's current learning process clearly departs from the one used by the brain. (Unless the way a baby fiddles with each new object is in fact a form of data augmentation, increasing the visual sample space by changing the object's angles. But I digress...)

Better to drop the "Neural Net as Brain" metaphor before we Procrustes it to death.

randyrand• 8 years ago

is that a limit of DL?

or is that a limit of how we currently do our training?

hnick2• 8 years ago

It also has no imagination, unless I've missed something (I admit I really need to catch up in this area).

Humans are experts at imagining 'what if' scenarios and learning from things that never even happened. "If this was bright red, is it still a cat? Or do all cats have a limited range of colours on a fixed cat-colour-spectrum?" I'm not sure if any DL systems do that?

YeGoblynQueenne10• 8 years ago

> And it could make the networks far less vulnerable to adversarial attacks simply because a system that represents things as objects, as opposed to patterns of pixels, isn’t going to be so easily thrown off by a little noise or an extraneous sticker.

But, representing objects and their relations is how Good, Old-Fashioned AI works and the inability to deal with noisy input was, exactly, where GOFAI systems stumbled. How are relational representations going to suddendly help neural nets (famously robust to noise already) better handle noise?

In any case, if we really want to learn relations there are symbolic learning techniques that can do that very cheaply from very few examples. Are we reinventing the wheel, only with more artificial neurons, now? And what's the point of that?

AstralStorm• 7 years ago

The problem is that neither AI can differentiate between something new and noise. Even probabilistic models have this problem to a lesser degree. So the idea is indeed to have an ANN learn a kind of a semantic network - then the problem is still pruning it and clustering reliably as well as detecting the new.

darawk• 8 years ago

Then there’s the opacity problem. Once a deep-learning system has been trained, it’s not always clear how it’s making its decisions.

This is no less true of human decision making.

otabdeveloper1• 7 years ago

What do you mean 'limits'?? There are no limits to deep learning! Singularity here we come!

(That was sarcasm. Sad that a disclaimer is needed nowadays.)

What are the limits of deep learning?

Related Stories