The Deep Learning Gold Rush of 2015

In the last few decades, we have witnessed major technological innovations such as personal computers and the internet finally reach the mainstream. And with mobile devices and social networks on the rise, we're now more connected than ever. So what's next? When is it coming? And how will it change our lives? Today I'll tell you that the next big advance is well underway and it's being fueled by a recent technique in the field of Artificial Intelligence known as Deep Learning.

The California Gold Rush of 2015 is all about Deep Learning. 

It's everywhere, you just don't know how to look.

All of today's excitement in Artificial Intelligence and Machine Learning stems from ground-breaking results in speech and visual object recognition using Deep Learning[1]. These algorithms are being applied to all sorts of data, and the learned deep neural networks outperform traditional expert systems carefully designed by scientists and engineers. End-to-end learning of deep representations from raw data is now possible due to a handful of well-performing deep learning recipes (ConvNets, Dropout, ReLUs, LSTM, DQN, ImageNet). But if there's one final takeaway that we can extract from decades of machine learning research, is that for many problems going deep isn't a choice, it's often a requirement.

Most of the apps and services you're already using (AirBnB, Snapchat, Twitch.tv, Uber, Yelp, LinkedIn, etc) are quite data-hungry and before you know it, they're all going to go mega-deep. So whether you need to revitalize your data science team with deep learning or you're starting an AI-from-day-one operation, it's pretty clear that everybody is rushing to get some of this Silicon Valley Gold.

From Titans to Gold Miners: Your atypical Gold Rush

Like all great gold rushes, this movement is led by new faces, which are pouring into Silicon Valley like droves. But these aren't your typical unskilled immigrants willing to pick up a hammer, nor your fresh computer science grads with some app-writing skills. The key deep learning players of today (known as the Titans of Deep Learning) are computer science professors and researchers (seldom born in the USA) leaving their academic posts and bringing their students and ideas straight into Silicon Valley.

"Turn on, Tune in, Dropout" -- Timothy Leary

Recently, Google and Facebook announced that their operations are now being powered by Deep Learning [2,3]. And with most Deep Learning Titans representing the tech giants (Yann LeCun at Facebook Research, Geoffrey Hinton at Google, Andrew Ng at Baidu), Deep Learning is likely to become one of the most sought after tech skills. With Toyota to invest in $1 Billion in Robotics and Artificial Intelligence Research (November 6, 2015), the announcement of YC Research (October 7, 2015), and the new Google Brain Residency Program "Pre-doc" AI jobs (October 26, 2015), Silicon Valley just got a whole lot more interesting.

Silicon Valley re-defines itself, yet again 

To understand why it took so long for Deep Learning to take-off, let's take a brief look at the key technologies which defined Silicon Valley over the last 50 years.  The following timeline gives an overview of where Silicon Valley has been and where it's going.

Figure Adapted from Steve Blank's Secret History of Silicon Valley

1970s: Semiconductors 
The story of the digital-era starts with semiconductors. "Silicon" in "Silicon Valley" originally referred to the silicon chip or integrated circuit innovations as well as the location (close to Stanford) of much tech-related activity. The dominant firm from that time period was Fairchild Semiconductor International and it eventually gave rise to more recognizable companies like Intel. For a more detailed discussion of this birthing era, take a look at Steve Blank's Secret History of Silicon Valley[4].

Read more about Fairchild at TechCrunch's First Trillion-Dollar Startup 

1980s: Personal Computers
Initially computers were quite large and used solely by research labs, government, and big businesses. But it was the personal computer which turned computer programming from a hobby into a vital skill. You no longer needed to be an MIT student to program on one of these badboys. While both Microsoft and Apple were founded in 1975 and 1976, respectively, they persevered due to their pioneering work in graphical user interfaces. This was the birth of the modern user-friendly Operating System. IBM approached Microsoft in 1980, regarding its upcoming personal computer, and from then on Microsoft would be King for a very long time.

See Mac-history's article on Microsoft's relationship with Apple

1990s: Internet
While the nerds at Universities were posting ascii messages on newsgroups in the 90s, service providers in the 1990s like AOL helped make the internet accessible to everyone. Remember getting all those AOL disks in the mail? Buying a chunk of digital real state (your own domain name) became possible and anybody with a dial up connection and some primitive text/HTML skills could start posting online content. With a mission statement like "organize the world's information", it was eventually Google that got the most of out the late 90s dot-com bubble, and remains a very strong player in all things tech.

2000s: Mobile and Social
While the dot-com bubble was about creating an online presence for startups and established companies, the way we use the internet has dramatically changed since 2001. A ton of new social communities have emerged, and due to Facebook we're now stars in our own reality show. Social and advertising have essentially turned the modern internet into a mainstream TV-like experience. The internet is no longer only for the nerds. The kings of this era (Google and Facebook) are also the biggest players in the Deep Learning space, because they have the largest user bases and in-house apps which can benefit most from machine learning.

2010-2015: Deep Learning comes to the party
Spend more than a day in Silicon Valley and you'll hear the popular expression, "Software is eating the world." Rampant spreading of software was only possible once the internet (1990s) AND mobile devices (2000s) became essential parts of our lives. No longer do we physically mail floppy disks, and social media fuels any app that goes viral. What traditional software is missing (or has been missing up until now) is the ability to improve over time from everyday use. If that same software is able to connect to a large Deep Learning system and start improving, then we have a game-changer on our hands. This is already happening with online advertising, digital assistants like Siri, and smart auto-responders like Google's new email auto-reply feature.

The hierarchical award-winning "AlexNet" Deep Learning architecture 

Visualized using MIT's Toolbox for Deep Learning Neuron Visualization

Massive hiring of deep learning experts by the leading tech companies has only begun, but we also should be on the lookout for new ventures built on top of Deep Learning, not just a revitalization of last decade's successes. On this front, keep a close look at the following Deep Learning Cloud Service upstarts: Richard Socher from MetaMind, Matthew Zeiler from Clarifai, and Carlos Guestrin from Dato.

2015-2020: Deep Learning Revitalizes Robotics
Recently it has been shown that Deep Learning can be used to help robots learn tasks involving movement, object manipulation, and decision making[6,7,8,9]. Before Deep Learning, lots of different pieces of robotic software and hardware would have to be developed independently and then hacked together for demo day. Today, you can use one of a handful of "Deep Learning for Robotics recipes" and start watching your robot learn the task you care about.

Robots Learns to Grasp using Deep Learning at Carnegie Mellon University. 

Supersizing Self-supervision: Learning to Grasp from 50K Tries and 700 Robot Hours [6]

With their 2013 acquisition of Boston Dynamics (a hardware play), 2014 acquisition of DeepMind (a software play), and a serious autonomous car play, Google is definitely early to the Robotics party. But the noteworthy bits are happening at the intersection of deep learning and robotics.  I suggest taking a closer look at the Robotics research of Pieter Abbeel of Berkeley, Abhinav Gupta of Carnegie Mellon, and Ashutosh Saxena of Stanford -- all likely stars in the next Deep Learning for Robotics race. As long as Rodney Brooks keeps creating innovative Robotics platforms like Baxter, my expectations for Robotics are off the charts.

Conclusion

Unlike in 1849, the Deep Learning Gold Rush of 2015 is not going to bring some 300,000 gold-seekers in boats to California's mainland. This isn't a bring-your-own-hammer kind of game -- the Titans have already descended from their Ivory Towers and handed us ample mining tools. But it won't hurt to gain some experience with traditional "shallow" machine learning techniques so you can appreciate the power of Deep Learning.

I hope you enjoyed today's read and have a better sense of how Silicon Valley is undergoing a transformation. And remember, today's wave of Deep Learning upstart CEOs have PhDs, but once Deep Learning software becomes more user-friendly (TensorFlow?), maybe you won't have to wait so long to dropout.


References

[1] Krizhevsky, A., Sutskever, I. and Hinton, G. E. ImageNet Classification with Deep Convolutional Neural Networks. In NIPS 2012.
[2] D'Onfro, J. Google is 're-thinking' all of its products to include machine learning. Business Insider. October 22, 2015.
[3] D'Onfro, J. How Facebook will use artificial intelligence to organize insane amounts of data into the perfect News Feed and a personal assistant with superpowers. Business Insider. November 3, 2015.
[4] Blank, S. Secret History of Silicon Valley. 2008.
[5] Donglai Wei, Bolei Zhou, Antonio Torralba William T. Freeman. mNeuron: A Matlab Plugin to Visualize Neurons from Deep Models. 2015.
[6] Lerrel Pinto, Abhinav Gupta. Supersizing Self-supervision: Learning to Graspfrom 50K Tries and 700 Robot Hours. arXiv. 2015.
[7] Sergey Levine, Chelsea Finn, Trevor Darrell, Pieter Abbeel. End-to-End Training of Deep Visuomotor Policies. In RSS 2015.
[8] Mnih, Volodymyr, et al. "Human-level control through deep reinforcement learning." Nature 518.7540 (2015): 529-533.
[9] Ian Lenz, Ross Knepper, and Ashutosh Saxena. DeepMPC: Learning Deep Latent Features for Model Predictive Control.  In Robotics Science and Systems (RSS), 2015

From feature descriptors to deep learning: 20 years of computer vision

We all know that deep convolutional neural networks have produced some stellar results on object detection and recognition benchmarks in the past two years (2012-2014), so you might wonder: what did the earlier object recognition techniques look like? How do the designs of earlier recognition systems relate to the modern multi-layer convolution-based framework?

Let's take a look at some of the big ideas in Computer Vision from the last 20 years.

The rise of the local feature descriptors: ~1995 to ~2000

When SIFT (an acronym for Scale Invariant Feature Transform) was introduced by David Lowe in 1999, the world of computer vision research changed almost overnight. It was robust solution to the problem of comparing image patches. Before SIFT entered the game, people were just using SSD (sum of squared distances) to compare patches and not giving it much thought.

The SIFT recipe: gradient orientations, normalization tricks

SIFT is something called a local feature descriptor -- it is one of those research findings which is the result of one ambitious man hackplaying with pixels for more than a decade.  Lowe and the University of British Columbia got a patent on SIFT and Lowe released a nice compiled binary of his very own SIFT implementation for researchers to use in their work.  SIFT allows a point inside an RGB imagine to be represented robustly by a low dimensional vector.  When you take multiple images of the same physical object while rotating the camera, the SIFT descriptors of corresponding points are very similar in their 128-D space.  At first glance it seems silly that you need to do something as complex as SIFT, but believe me: just because you, a human, can look at two image patches and quickly "understand" that they belong to the same physical point, this is not the same for machines.  SIFT had massive implications for the geometric side of computer vision (stereo, Structure from Motion, etc) and later became the basis for the popular Bag of Words model for object recognition.

Seeing a technique like SIFT dramatically outperform an alternative method like Sum-of-Squared-Distances (SSD) Image Patch Matching firsthand is an important step in every aspiring vision scientist's career. And SIFT isn't just a vector of filter bank responses, the binning and normalization steps are very important. It is also worthwhile noting that while SIFT was initially (in its published form) applied to the output of an interest point detector, later it was found that the interest point detection step was not important in categorization problems.  For categorization, researchers eventually moved towards vector quantized SIFT applied densely across an image.

I should also mention that other descriptors such as Spin Images (see my 2009 blog post on spin images) came out a little bit earlier than SIFT, but because Spin Images were solely applicable to 2.5D data, this feature's impact wasn't as great as that of SIFT. 

The modern dataset (aka the hardening of vision as science): ~2000 to ~2005

Homography estimation, ground-plane estimation, robotic vision, SfM, and all other geometric problems in vision greatly benefited from robust image features such as SIFT.  But towards the end of the 1990s, it was clear that the internet was the next big thing.  Images were going online. Datasets were being created.  And no longer was the current generation solely interested in structure recovery (aka geometric) problems.  This was the beginning of the large-scale dataset era with Caltech-101 slowly gaining popularity and categorization research on the rise. No longer were researchers evaluating their own algorithms on their own in-house datasets -- we now had a more objective and standard way to determine if yours is bigger than mine.  Even though Caltech-101 is considered outdated by 2015 standards, it is fair to think of this dataset as the Grandfather of the more modern ImageNet dataset. Thanks Fei-Fei Li.

Category-based datasets: the infamous Caltech-101 TorralbaArt image

Bins, Grids, and Visual Words (aka Machine Learning meets descriptors): ~2000 to ~2005
After the community shifted towards more ambitious object recognition problems and away from geometry recovery problems, we had a flurry of research in Bag of Words, Spatial Pyramids, Vector Quantization, as well as machine learning tools used in any and all stages of the computer vision pipeline.  Raw SIFT was great for wide-baseline stereo, but it wasn't powerful enough to provide matches between two distinct object instances from the same visual object category.  What was needed was a way to encode the following ideas: object parts can deform relative to each other and some image patches can be missing.  Overall, a much more statistical way to characterize objects was needed.

Visual Words were introduced by Josef Sivic and Andrew Zisserman in approximately 2003 and this was a clever way of taking algorithms from large-scale text matching and applying them to visual content.  A visual dictionary can be obtained by performing unsupervised learning (basically just K-means) on SIFT descriptors which maps these 128-D real-valued vectors into integers (which are cluster center assignments).  A histogram of these visual words is a fairly robust way to represent images.  Variants of the Bag of Words model are still heavily utilized in vision research.

Josef Sivic's "Video Google": Matching Graffiti inside the Run Lola Run video

Another idea which was gaining traction at the time was the idea of using some sort of binning structure for matching objects.  Caltech-101 images mostly contained objects, so these grids were initially placed around entire images, and later on they would be placed around object bounding boxes.  Here is a picture from Kristen Grauman's famous Pyramid Match Kernel paper which introduced a powerful and hierarchical way of integrating spatial information into the image matching process.

Grauman's Pyramid Match Kernel for Improved Image Matching

At some point it was not clear whether researchers should focus on better features, better comparison metrics, or better learning.  In the mid 2000s it wasn't clear if young PhD students should spend more time concocting new descriptors or kernelizing their support vector machines to death.

Object Templates (aka the reign of HOG and DPM): ~2005 to ~2010

At around 2005, a young researcher named Navneet Dalal showed the world just what can be done with his own new badass feature descriptor, HOG.  (It is sometimes written as HoG, but because it is an acronym for “Histogram of Oriented Gradients” it should really be HOG. The confusion must have came from an earlier approach called DoG which stood for Difference of Gaussian, in which case the “o” should definitely be lower case.)

Navneet Dalal's HOG Descriptor

HOG came at the time when everybody was applying spatial binning to bags of words, using multiple layers of learning, and making their systems overly complicated. Dalal’s ingenious descriptor was actually quite simple.  The seminal HOG paper was published in 2005 by Navneet and his PhD advisor, Bill Triggs. Triggs got his fame from earlier work on geometric vision, and Dr. Dalal got his fame from his newly found descriptor.  HOG was initially applied to the problem of pedestrian detection, and one of the reasons it because so popular was that the machine learning tool used on top of HOG was quite simple and well understood, it was the linear Support Vector Machine.

I should point out that in 2008, a follow-up paper on object detection, which introduced a technique called the Deformable Parts-based Model (or DPM as we vision guys call it), helped reinforce the popularity and strength of the HOG technique. I personally jumped on the HOG bandwagon in about 2008.  My first few years as a grad student (2005-2008) I was hackplaying with my own vector quantized filter bank responses, and definitely developed some strong intuition regarding features.  In the end I realized that my own features were only "okay," and because I was applying them to the outputs of image segmentation algorithms they were extremely slow.  Once I started using HOG, it didn’t take me long to realize there was no going back to custom, slow, features.  Once I started using a multiscale feature pyramid with a slightly improved version of HOG introduced by master hackers such as Ramanan and Felzenszwalb, I was processing images at 100x the speed of multiple segmentations + custom features (my earlier work).

The infamous Deformable Part-based Model (for a Person)

DPM was the reigning champ on the PASCAL VOC challenge, and one of the reasons why it became so popular was the excellent MATLAB/C++ implementation by Ramanan and Felzenszwalb.  I still know many researchers who never fully acknowledged what releasing such great code really meant for the fresh generation of incoming PhD students, but at some point it seems like everybody was modifying the DPM codebase for their own CVPR attempts.  Too many incoming students were lacking solid software engineering skills and giving them the DPM code was a surefire way to get some some experiments up and running.  Personally, I never jumped on the parts-based methodology, but I did take apart the DPM codebase several times.  However, when I put it back together, the Exemplar-SVM was the result.

Big data, Convolutional Neural Networks and the promise of Deep Learning: ~2010 to ~2015

Sometime around 2008, it was pretty clear that scientists were getting more and more comfortable with large datasets.  It wasn’t just the rise of “Cloud Computing” and “Big Data,” it was the rise of the data scientists.  Hacking on equations by morning, developing a prototype during lunch, deploying large scale computations in the evening, and integrating the findings into a production system by sunset.  I spent two summers at Google Research, I saw lots of guys who had made their fame as vision hackers.  But they weren’t just writing “academic” papers at Google -- sharding datasets with one hand, compiling results for their managers, writing Borg scripts in their sleep, and piping results into gnuplot (because Jedis don’t need GUIs?). It was pretty clear that big data, and a DevOps mentality was here to stay, and the vision researcher of tomorrow would be quite comfortable with large datasets.  No longer did you need one guy with a mathy PhD, one software engineer, one manager, and one tester.  Plenty of guys who could do all of those jobs.

Deep Learning: 1980s - 2015

2014 was definitely a big year for Deep Learning.  What’s interesting about Deep Learning is that it is a very old technique.  What we're seeing now is essentially the Neural Network 2.0 revolution -- but this time around, there's we're 20 years ahead R&D-wise and our computers are orders of magnitude faster.  And what’s funny is that the same guys that were championing such techniques in the early 90s were the same guys we were laughing at in the late 90s (because clearly convex methods were superior to the magical NN learning-rate knobs). I guess they really had the last laugh because eventually these relentless neural network gurus became the same guys we now all look up to.  Geoffrey Hinton, Yann LeCun, Andrew Ng, and Yeshua Bengio are the 4 Titans of Deep Learning.  By now, just about everybody has jumped ship to become a champion of Deep Learning.

But with Google, Facebook, Baidu, and a multitude of little startups riding the Deep Learning wave, who will rise to the top as the master of artificial intelligence?

Yann's Deep Learning Page

How to today's deep learning systems resemble the recognition systems of yesteryear?

Multiscale convolutional neural networks aren't that much different than the feature-based systems of the past.  The first level neurons in deep learning systems learn to utilize gradients in a way that is similar to hand-crafted features such as SIFT and HOG.  Objects used to be found in a sliding-window fashion, but now it is easier and sexier to think of this operation as convolving an image with a filter. Some of the best detection systems used to use multiple linear SVMs, combined in some ad-hoc way, and now we are essentially using even more of such linear decision boundaries.  Deep learning systems can be thought of a multiple stages of applying linear operators and piping them through a non-linear activation function, but deep learning is more similar to a clever combination of linear SVMs than a memory-ish Kernel-based learning system.

Features these days aren't engineered by hand.  However, architectures of Deep systems are still being designed manually -- and it looks like the experts are the best at this task.  The operations on the inside of both classic and modern recognition systems are still very much the same.  You still need to be clever to play in the game, but now you need a big computer. There's still lot of room for improvement, so I encourage all of you to be creative in your research.

Research-wise, it never hurts to know where we have been before so that we can better plan for our journey ahead.  I hope you enjoyed this brief history lesson and the next time you look for insights in your research, don't be afraid to look back.

To learn more about computer vision techniques:

SIFT article on Wikipedia

Bag of Words article on Wikipedia

HOG article on Wikipedia
Deformable Part-based Model Homepage
Pyramid Match Kernel Homepage
"Video Google" Image Retrieval System

Some Computer Vision datasets:
Caltech-101 Dataset
ImageNet Dataset

To learn about the people mentioned in this article:

Kristen Grauman (creator of Pyramid Match Kernel, Prof at Univ of Texas)
Bill Triggs's (co-creator of HOG, Researcher at INRIA)
Navneet Dalal (co-creator of HOG, now at Google)

Yann LeCun (one of the Titans of Deep Learning, at NYU and Facebook)

Geoffrey Hinton (one of the Titans of Deep Learning, at Univ of Toronto and Google)
Andrew Ng (leading the Deep Learning effort at Baidu, Prof at Stanford)
Yoshua Bengio (one of the Titans of Deep Learning, Prof at U Montreal)

Deva Ramanan (one of the creators of DPM, Prof at UC Irvine)
Pedro Felzenszwalb (one of the creators of DPM, Prof at Brown)
Fei-fei Li (Caltech101 and ImageNet, Prof at Stanford)
Josef Sivic (Video Google and Visual Words, Researcher at INRIA/ENS)
Andrew Zisserman (Geometry-based methods in vision, Prof at Oxford)
Andrew E. Johnson (SPIN Images creator, Researcher at JPL)
Martial Hebert (Geometry-based methods in vision, Prof at CMU)