Menu
Sign In Search Podcasts Charts People & Topics Add Podcast API Pricing
Podcast Image

Lex Fridman Podcast

Yoshua Bengio: Deep Learning

20 Oct 2018

42 min duration
6142 words
2 speakers
20 Oct 2018
Description

Yoshua Bengio, along with Geoffrey Hinton and Yann Lecun, is considered one of the three people most responsible for the advancement of deep learning during the 1990s, 2000s, and now. Cited 139,000 times, he has been integral to some of the biggest breakthroughs in AI over the past 3 decades. Video version is available on YouTube. If you would like to get more information about this podcast go to https://lexfridman.com/ai or connect with @lexfridman on Twitter, LinkedIn, Facebook, or YouTube where you can watch the video versions of these conversations. 

Audio
Featured in this Episode
Topics Discussed
Transcription

Chapter 1: Who is Yoshua Bengio and what are his contributions to deep learning?

0.031 - 22.191 Lex Fridman

Welcome to the Artificial Intelligence Podcast. My name is Lex Friedman. I'm a research scientist at MIT. If you enjoy this podcast, please rate it on iTunes or your podcast provider of choice, or simply connect with me on Twitter and other social networks at Lex Friedman, spelled F-R-I-D. Today is a conversation with Yoshio Bengio.

0

22.171 - 43.15 Lex Fridman

Along with Jeff Hinton and Yann LeCun, he's considered one of the three people most responsible for the advancement of deep learning during the 1990s and the 2000s and now. Cited 139,000 times, he has been integral to some of the biggest breakthroughs in AI over the past three decades.

0

60.123 - 67.753 Lex Fridman

What difference between biological neural networks and artificial neural networks is most mysterious, captivating, and profound for you?

0

69.636 - 84.375 Yoshua Bengio

First of all, there's so much we don't know about biological neural networks. And that's very mysterious and captivating because maybe it holds the key to improving artificial neural networks. One of the things I studied...

0

85.435 - 110.156 Yoshua Bengio

recently, something that we don't know how biological neural networks do, but would be really useful for artificial ones, is the ability to do credit assignment through very long time spans. There are things that we can in principle do with artificial neural nets, but it's not very convenient and it's not biologically plausible.

111.058 - 140.485 Yoshua Bengio

And this mismatch, I think, this kind of mismatch may be an interesting thing to study to, A, understand better how brains might do these things, because we don't have good corresponding theories with artificial neural nets, and B, maybe provide new ideas that we could explore about things that brain do differently and that we could incorporate in artificial neural nets.

140.667 - 147.615 Lex Fridman

So let's break credit assignment up a little bit. It's a beautifully technical term, but it could incorporate so many things.

148.256 - 168.801 Lex Fridman

So is it more on the RNN memory side, thinking like that, or is it something about knowledge, building up common sense knowledge over time, or is it more in the reinforcement learning sense that you're picking up rewards over time for a particular, to achieve a certain kind of goal?

168.781 - 202.117 Yoshua Bengio

I was thinking more about the first two meanings whereby we store all kinds of memories, episodic memories in our brain, which we can access later in order to help us both infer causes of things that we are observing now. and assign credit to decisions or interpretations we came up with a while ago when you know, those memories were stored.

Chapter 2: What are the differences between biological and artificial neural networks?

578.455 - 603 Yoshua Bengio

Sort of almost guiding some aspect of learning. Right, right, right. So I was talking to Rebecca Sachs just an hour ago, And she was talking about lots and lots of evidence from infants seem to clearly pick what interests them in a directed way. And so they're not passive learners.

0

603.381 - 616.153 Yoshua Bengio

They focus their attention on aspects of the world which are most interesting, surprising in a non-trivial way that makes them change their theories of the world.

0

618.476 - 649.771 Lex Fridman

So that's a fascinating view of the future progress, but on a more maybe boring question, do you think going deeper and larger, so do you think just increasing the size of the things that have been increasing a lot in the past few years will also make significant progress? So Some of the representational issues that you mentioned, they're kind of shallow in some sense.

0

651.093 - 653.195 Yoshua Bengio

Understanding prior. In the sense of abstraction.

0

653.636 - 654.718 Lex Fridman

In the sense of abstraction.

654.738 - 665.552 Yoshua Bengio

They're not getting some. I don't think that having more depth in the network in the sense of instead of 100 layers, we have 10,000 is going to solve our problem. You don't think so?

667.314 - 668.596 Lex Fridman

No. Is that obvious to you?

668.836 - 694.204 Yoshua Bengio

Yes. What is clear to me is that engineers and companies and labs and grad students will continue to tune architectures and explore all kinds of tweaks to make the current state of the art ever slightly better. But I don't think that's going to be nearly enough. I think we need some fairly drastic changes in the way that we're considering learning

694.184 - 704.16 Yoshua Bengio

To achieve the goal that these learners actually understand in a deep way the environment in which they are, you know, observing and acting.

Chapter 3: How can credit assignment in neural networks be improved?

1075.981 - 1093.123 Yoshua Bengio

The kind of knowledge about those relationships in a classical AI system is encoded in the rules. Like a rule is just like a little piece of knowledge that says, oh, I have these two, three, four variables that are linked in this interesting way. Then I can say something about one or two of them given a couple of others, right?

0

1093.324 - 1123.03 Yoshua Bengio

In addition to disentangling the elements of the representation, which are like the variables in a rule-based system, you also need to disentangle the the mechanisms that relate those variables to each other. So like the rules. So the rules are neatly separated. Each rule is living on its own. And when I change a rule because I'm learning, it doesn't need to break other rules.

0

1123.43 - 1149.07 Yoshua Bengio

Whereas current neural nets, for example, are very sensitive to what's called catastrophic forgetting, where after I've learned some things and then I learn new things, they can destroy the old things that I had learned, right? If the knowledge was better factorized and separated, disentangled, then you would avoid a lot of that. Now, you can't do this in the sensory domain, but

0

1150.248 - 1173.793 Yoshua Bengio

What do you mean by sensor? Like in pixel space. But my idea is that when you project the data in the right semantic space, it becomes possible to now represent this extra knowledge beyond the transformation from input to representations, which is how representations act on each other and predict the future and so on. in a way that can be neatly disentangled.

0

1173.994 - 1178.782 Yoshua Bengio

So now it's the rules that are disentangled from each other and not just the variables that are disentangled from each other.

1179.623 - 1182.869 Lex Fridman

And you draw a distinction between semantic space and pixel.

1183.169 - 1183.71 Yoshua Bengio

Yes.

1183.73 - 1185.734 Lex Fridman

Does there need to be an architectural difference?

1185.754 - 1212.739 Yoshua Bengio

Well, yeah. So there's the sensory space like pixels where everything is entangled. the the the information like the variables are completely interdependent in very complicated ways and also computation like the the it's not just variables it's also how they are related to each other is is all intertwined but but i i'm hypothesizing that in the right high level representation space both

Chapter 4: What are the limitations of current deep learning architectures?

2171.739 - 2172.681 Yoshua Bengio

Yes, it's my mother tongue.

0

2173.021 - 2187.322 Lex Fridman

It's one of the Romance languages. Do you think passing the Turing test and all the underlying challenges we just mentioned depend on language? Do you think it might be easier in French than it is in English? No. Or is it independent of language?

0

2187.342 - 2204.398 Yoshua Bengio

I think it's independent of language. I would like to build systems that can use the same principles, the same learning mechanisms to learn from human agents, whatever their language is.

0

2205.526 - 2225.4 Lex Fridman

Well, certainly us humans can talk more beautifully and smoothly in poetry. So I'm Russian originally. I know poetry in Russian is maybe easier to convey complex ideas than it is in English. But maybe I'm showing my bias and some people could say that about French.

0

2225.961 - 2236.193 Lex Fridman

But of course, the goal ultimately is our human brain is able to utilize any kind of those languages to use them as tools to convey meaning.

2237.054 - 2249.708 Yoshua Bengio

Yeah, of course, there are differences between languages and maybe some are slightly better at some things. But in the grand scheme of things, where we're trying to understand how the brain works and language and so on, I think these differences are minute.

2251.257 - 2270.346 Lex Fridman

So you've lived perhaps through an AI winter of sorts. Yes. How did you stay warm and continue with your research? Stay warm with friends. With friends. Okay. So it's important to have friends. And what have you learned from the experience?

2271.608 - 2293.95 Yoshua Bengio

Listen to your inner voice. Don't, you know, be trying to just please the crowds and the fashion. And if you have a strong intuition about something that is not contradicted by actual evidence, go for it. I mean, it could be contradicted by people.

2295.312 - 2298.836 Lex Fridman

But not your own instinct of based on everything you've learned.

Comments

There are no comments yet.

Please log in to write the first comment.