![\"Dr.](\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2019\/12\/Research_Podcast_Phil-Bachman_Site_10_2019_1400x788-1024x576.png\")
<\/p>\n<\/p>\n
Deep learning methodologies like supervised learning have been very successful in training machines to make predictions about the world. But because they\u2019re so dependent upon large amounts of human-annotated data, they\u2019ve been difficult to scale. Dr. Phil Bachman<\/a>, a researcher at MSR Montreal<\/a>, would like to change that, and he\u2019s working to train machines to collect, sort and label their own data, so people don\u2019t have to.<\/p>\n Today, Dr. Bachman gives us an overview of the machine learning landscape and tells us why it\u2019s been so difficult to sort through noise and get to useful information. He also talks about his ongoing work on Deep InfoMax<\/a>, a novel approach to self-supervised learning, and reveals what a conversation about ML classification problems has to do with Harrison Ford\u2019s face.<\/p>\n Phil Bachman: Training a machine to look at a large amount of unannotated data and point to specific examples and say, well, I think if a human comes in and tells me exactly what that thing is, I\u2019ll learn a lot about the problem that I\u2019m trying to solve<\/span>.<\/span> <\/span>S<\/span>o this general notion of carefully selecting which of those examples you want to spend the money or spend the time to get a human to go in and provide the annotations for those examples, that\u2019s this idea of active learning.<\/span> <\/span><\/p>\n Host: <\/span><\/b>You\u2019re listening to the Microsoft Research Podcast, a show that brings you closer to the cutting-edge of technology research and the scientists behind it. I\u2019m your host, Gretchen Huizinga.<\/span><\/b> <\/span><\/p>\n Host: <\/span><\/b>Deep learning methodologies<\/span><\/b> like supervised learning have been very successful in training machines to make predictions about the world<\/span><\/b>.<\/span><\/b> <\/span><\/b>B<\/span><\/b>ut because they\u2019re so dependent upon large amounts of human-annotated data, they\u2019ve been difficult to scale. Dr. Phil Bachman, a researcher at MSR Montreal, would like to change that, and he\u2019s working to train machines to collect, sort and label their own data, so people don\u2019t have to.<\/span><\/b> <\/span><\/p>\n Today, Dr. Bachman gives us an overview of the machine learning landscape and tells us why it\u2019s been so difficult to sort through noise and get to useful information. He also talks about his ongoing work on Deep <\/span><\/b>InfoMax<\/span><\/b>, a novel approach to self-supervised learning, and reveals what a conversation about ML classification problems has to do with Harrison Ford\u2019s face.<\/span><\/b> <\/span><\/b>That and much more on this episode of the Microsoft Research Podcast.<\/span><\/b> <\/span><\/p>\n (music plays)<\/span><\/i><\/b> <\/span><\/p>\n Host: Phil Bachman, welcome to the podcast<\/span><\/b>!<\/span><\/b> <\/span><\/p>\n Phil Bachman: Hi. Thanks for having me.<\/span> <\/span><\/p>\n Host: So as a researcher at MSR Montreal, you\u2019ve got a lot going on. Let\u2019s start macro and then get micro. And we\u2019ll start with a little phrase that I like in your bio that says you want to understand the ways in which actionable information can be distilled from raw data. Unpack it for us. What big problem or problems are you working on? What gets you up in the morning?<\/span><\/b> <\/span><\/p>\n Phil Bachman: So I\u2019d say the key here is to sort of understand the distinction between information in general and, let\u2019s say, information that might be useful. <\/span>So for example, if images are coming from the camera that you are using to pilot a self-driving car, then low-level sensor noise probably doesn\u2019t provide you useful information<\/span>\u2026<\/span> <\/span><\/p>\n Host: Hmm.<\/span><\/b> <\/span><\/p>\n Phil Bachman: <\/span>\u2026<\/span>for deciding whether to stop the car or whether to turn or make other sorts of decisions that are useful for driving.<\/span> <\/span>So, what I\u2019m interested in, sort of this phrase, actionable information here, it\u2019s referring specifically to trying to focus on getting our models to capture the information content in the data that we\u2019re looking at that is actually going to be useful in the future for making some sorts of decisions. So if we\u2019re training a model that\u2019s processing the video data that\u2019s being used to drive this car, then perhaps we don\u2019t want to waste the effort of the model on trying to represent this low-level information about small variations in pixel intensity. And we\u2019d rather have the model focus its capacity for representing information on the information that corresponds to sort of higher-level structure in the image, so things like the presence or absence of a pedestrian or another car in front of it. So that\u2019s kind of what I mean with this phrase, actionable information. So this distillation from raw data is on doing learning from data that hasn\u2019t been manually curated or that doesn\u2019t have a lot of information injected into it by a human who\u2019s doing the data collection process. So going back to the self-driving car example, I\u2019d like to have a system where we could allow the computer just to watch thousands of hours of video that\u2019s captured from a bunch of cars driving around. Then what I want to be able to do is have a system that\u2019s just watching all of that video and doesn\u2019t require that much input from a person who\u2019s pointing to the video and saying specifically what\u2019s going to be interesting or useful in the future. So this information that\u2019s going to be useful for performing the types of tasks that we want our model to do eventually.<\/span> <\/span><\/p>\n Host: B<\/span><\/b>efore we get specific, give us a short historical <\/span><\/b>tour <\/span><\/b>of the deep-learning methodologies as a level set<\/span><\/b>,<\/span><\/b> and then tell us why we need a methodology for learning representations from unlabeled data.<\/span><\/b> <\/span><\/p>\n Phil Bachman: Okay. So in the context of machine learning, people often break it down into three categories. So there will be supervised learning, unsupervised learning and reinforcement learning\u2026<\/span> <\/span><\/p>\n Host: <\/span><\/b>Mmm<\/span><\/b>-hmm.<\/span><\/b> <\/span><\/p>\n Phil Bachman: \u2026and it\u2019s not always clear what the distinction between the methods are. But supervised learning is sort of what\u2019s had the most immediate success and what\u2019s driving a lot of the deep learning power technologies that are being used for doing things like speech recognition in phones or doing automated question answering for chat bots and stuff like that. So supervised learning refers to kind of a subset of the techniques that people apply when they have access to a large amount of data and they have a specific type of action that they want a model to perform when it processes that data. And what they do is<\/span>,<\/span> they get a person to go and label all the data and say, okay, well this is the input to the model at this point in time. And given this input, this is what the model should output. So yo<\/span>u\u2019<\/span>re putting a lot of constraints on what the model is doing and constructing those constraints manually by having a person looking at a set of a million images and, for each image, they say, oh, this is a cat, this is a dog, this is a person, this is a car. So after having done that for thousands of hours, you now have a large data set where you have a bunch of different images and each of those images has an associated tag. And so now the kind <\/span>of techniques that we work with and the optimization methods that we use for training our models, are very effective at fitting really large powerful models to large amounts of this sort of annotated data.<\/span> <\/span><\/b>So that\u2019s kind of the traditional supervised learning. But the major downside there is that the process of providing all of those annotations can be very expensive. So that process of supervised learning has a lot of issues with scalability. What we\u2019d like to do<\/span>,<\/span> ideally<\/span>,<\/span> is make use of a lot of that and figure out what kinds of information is actionable. So finding the information that seems like it will be useful for making decisions. So that\u2019s getting into a contrast between supervised learning and unsupervised learning. And then there\u2019s also reinforcement learning which is a slightly different set of techniques where you actually allow a model to go out and <\/span>kind of perform experiments or <\/span>try to do things and then<\/span> somehow<\/span> it receives feedback about the things that it\u2019s doing that says, oh, what you just did, that was a good thing or that was a bad thing.<\/span> <\/span><\/p>\n Host: Hmm.<\/span><\/b> <\/span><\/p>\n Phil Bachman: And <\/span>that<\/span> i<\/span>t<\/span> learns by kind of a process of trial and error. So that\u2019s a general idea of reinforcement learning.<\/span> <\/span><\/p>\n Host: Hmm. Okay<\/span><\/b>. W<\/span><\/b>e mentioned two flavors of this, unsupervised and then self-supervised. Is that another differentiation there?<\/span><\/b> <\/span><\/p>\n Phil Bachman: So the self-supervised learning, it\u2019s not a completely different thing, but it\u2019s a sort of subset of those types of techniques. So, the general idea behind self-supervised learning is that we try to design procedures that will generate little supervised learning problems for a model to solve, where the process of generating those little supervised learning problems is kind of automatic. And the hope here is that the kind of procedurally-generated supervised learning problems that our little algorithm is generating, based on the unlabeled data, will force the model to capture some useful information about the structure of that data that will allow it to answer more, sort of, human-oriented questions easier in the future. So just to clarify this concept of procedurally generating supervised learning problems, one really simple example would be that you could try to train a model to have some understanding of the statistical structure of visual data by showing a model a bunch of images, but what you do is you take each image and you split it into a left half and a right half. So now what you do is you take your model, and all the model is allowed to see if the left half of the image\u2026<\/span> <\/span><\/p>\n Host: Hmm.<\/span><\/b> <\/span><\/p>\n Phil Bachman: \u2026and then you have another model that sort of tries to form a representation of the right half of the image. And so the model that looked at the left half of the image, you present it with representations of the right halves of, like, let\u2019s say, ten thousand images, one of which is the image that it looked at. So, it\u2019s kind of got like a partner that it\u2019s looking for in this big bag of encoded right halves of images. And the job of the encoder that\u2019s processing the left half of the image is to be able to look in that bag and pick out the right half that actually corresponds to the <\/span>image that it originally came from. So in this case, we\u2019re taking something that looks like unsupervised learning, but instead, here, what we\u2019re doing, is treating it more like a supervised learning problem. So the model that looks at the left half of the image, its task is to solve something that looks like just a simple classification problem. <\/span>And then making this like a one thousand<\/span>–<\/span>way classification problem.<\/span> <\/span><\/p>\n Host: The other thing that comes to my mind is, there\u2019s this weird thing on the internet where like, Harrison Ford\u2026 <\/span><\/b>y<\/span><\/b>ou see half of his face and the other half of his face and they are completely different. Like if you put each halves together, they wouldn\u2019t look like Harrison Ford, but together with the different halves, they look like him. <\/span><\/b>So that would really trick the machine<\/span><\/b>,<\/span><\/b> I would think.<\/span><\/b> <\/span><\/p>\n Phil Bachman: Actually, I wouldn\u2019t be so confident about that<\/span>!<\/span> <\/span><\/p>\n Host: Really?<\/span><\/b> <\/span><\/p>\n Phil Bachman: Yeah. <\/span>The question that you<\/span>\u2019<\/span>re sort of training the machine to answer is, which of these possible things do you think is most likely associated with the thing that you<\/span>\u2019<\/span>re currently looking at? So unless there was somebody else\u2019s right face half, that looked significantly more Harrison Ford-<\/span>ish<\/span>, than his own right face half, then the model actually could do pretty reasonably, I\u2019d expect.<\/span> <\/span><\/p>\n Host: That\u2019s hilarious.<\/span><\/b> <\/span><\/p>\n Phil Bachman: So unless you had somebody who\u2026 where it was like this really strict dichotomous separation of the halves of their face, like Two-Face from Batman or something\u2026<\/span> <\/span><\/p>\n Host: Right. That\u2019s another one!<\/span><\/b> <\/span><\/p>\n Phil Bachman: \u2026in which case maybe the model would fail, but\u2026<\/span> <\/span><\/p>\n Host: I love that.<\/span><\/b> <\/span><\/p>\n Phil Bachman: \u2026if it\u2019s within like standard realm of human variability, I think it would be okay. <\/span> <\/span><\/p>\n Host: Well that\u2019s good. So let\u2019s move ahead to the algorithms that we\u2019re talking about here. And you call them learning algorithms, and you\u2019ve described your goal for learning algorithms in some intriguing ways. You want to train machines to go out and fetch data for themselves and actively find out about the world, and you want to get the machine to ask itself interesting questions so it begins to build up its own knowledge base. Tell us about these learning algorithms for active learning and what it takes to turn a machine into an information-seeking missile?<\/span><\/b> <\/span><\/p>\n Phil Bachman: Yeah, so this kind of overall objective there that you\u2019ve described is targeted at kind of expanding the scope of which parts of the problems that we\u2019re currently trying to solve<\/span>,<\/span> are solved by the machine rather than by a person who is acting as a shepherd for the machine<\/span>,<\/span> or as a teacher or something along those lines<\/span>.<\/span> So right now, the machine learning component of most systems is <\/span>a <\/span>very important part of the system, but there\u2019s a whole lot of human effort that surrounds the production and use of something like a practical image classifier or a practical machine translation system. So that\u2019s one part of the effort that\u2019s required for getting an automated system out there in the world. So part of the process is just the initial decision, like the thing that we want to do is machine translation, here\u2019s a way of formalizing that problem and specifying it such that we can go out and now perform another part of the process \u2013 so this other part of process is a data collection.<\/span> <\/span><\/p>\n Host: Hmm.<\/span><\/b> <\/span><\/p>\n Phil Bachman: So you\u2019d have to go out and you\u2019d have to explicitly collect a lot of data that is relevant to the task that you are trying to solve. And then you have to take that data and <\/span>you maybe have to <\/span>have somebody curate it to make that data more directly useful or more immediately useful for the kinds of algorithms that we tend to use right now. So a lot of the work that I want to do is about trying to reduce the amount of human effort that\u2019s required on those two fronts and trying to get as much of those two parts of the problem automated and built into the models that we\u2019re training so that we don\u2019t have to go out and manually annotate all the data.<\/span>Related:<\/h3>\n
\n
\nTranscript<\/h3>\n