Deep Learning vs Big Data: Who owns what?

In order to learn anything useful, large-scale multi-layer deep neural networks (aka Deep Learning systems) require a large amount of labeled data. There is clearly a need for big data, but only a few places where big visual data is available. Today we'll take a look at one of the most popular sources of big visual data, peek inside a trained neural network, and ask ourselves some data/model ownership questions. The fundamental question to keep in mind is the following, "Are the learned weights of a neural network derivate works of the input images?" In other words, when deep learning touches your data, who owns what?



Background: The Deep Learning "Computer Vision Recipe"
One of today's most successful machine learning techniques is called Deep Learning. The broad interest in Deep Learning is backed by some remarkable results on real-world data interpretation tasks dealing with speech[1], text[2], and images[3]. Deep learning and object recognition techniques have been pioneered by academia (University of Toronto, NYU, Stanford, Berkeley, MIT, CMU, etc), picked up by industry (Google, Facebook, Snapchat, etc), and are now fueling a new generation of startups ready to bring visual intelligence to the masses (Clarifai.com, Metamind.io, Vision.ai, etc). And while it's still not clear where Artificial Intelligence is going, Deep Learning will be a key player.

Related blog post: Deep Learning vs Machine Learning vs Pattern Recognition
Related blog post: Deep Learning vs Probabilistic Graphical Models vs Logic

For visual object recognition tasks, the most popular models are Convolutional Neural Networks (also known as ConvNets or CNNs). They can be trained end-to-end without manual feature engineering, but this requires a large set of training images (sometimes called big data, or big visual data). These large neural networks start out as a Tabula Rasa (or "blank slate") and the full system is trained in an end-to-end fashion using a heavily optimized implementation of Backpropagation (informally called "backprop"). Backprop is nothing but the chain rule you learned in Calculus 101 and today's deep neural networks are trained in almost the same way they were trained in the 1980s. But today's highly-optimized implementations of backprop are GPU-based and can process orders of magnitude more data than was approachable in the pre-internet pre-cloud pre-GPU golden years of Neural Networks. The output of the deep learning training procedure is a set of learned weights for the different layers defined in the model architecture -- millions of floating point numbers representing what was learned from the images. So what's so interesting about the weights? It's the relationship between the weights and the original big data, that will be under scrutiny today.

"Are weights of a trained network based on ImageNet a derived work, a cesspool of millions of copyright claims? What about networks trained to approximate another ImageNet network?"
[This question was asked on HackerNews by kastnerkyle in the comments of A Revolutionary Technique That Changed Machine Vision.]

In the context of computer vision, this question truly piqued my interest, and as we start seeing robots and AI-powered devices enter our homes I expect much more serious versions of this question to arise in the upcoming decade. Let's see how some of these questions are being addressed in 2015.

1. ImageNet: Non-commercial Big Visual Data

Let's first take a look at the most common data source for Deep Learning systems designed to recognize a large number of different objects, namely ImageNet[4]. ImageNet is the de-facto source of big visual data for computer vision researchers working on large scale object recognition and detection. The dataset debuted in a 2009 CVPR paper by Fei-Fei Li's research group and was put in place to replace both PASCAL datasets (which lacked size and variety) and LabelMe datasets (which lacked standardization). ImageNet grew out of Caltech101 (a 2004 dataset focusing on image categorization, also pioneered by Fei-Fei Li) so personally I still think of ImageNet as something like "Stanford10^N". ImageNet has been a key player in organizing the scale of data that was required to push object recognition to its new frontier, the deep learning phase.

ImageNet has over 15 million images in its database as of May 1st, 2015.

Problem: Lots of extremely large datasets are mined from internet images, but these images often come with their own copyright.  This prevents collecting and selling such images, and from a commercial point of view, when creating such a dataset, some care has to be taken.  For research to keep pushing the state-of-the-art on real-world recognition problems, we have to use standard big datasets (representative of what is found in the real-world internet), foster a strong sense of community centered around sharing results, and maintain the copyrights of the original sources.

Solution: ImageNet decided to publicly provide links to the dataset images so that they can be downloaded without having to be hosted on an University-owned server. The ImageNet website only serves the image thumbnails and provides a copyright infringement clause together with instructions where to file a DMCA takedown notice. The dataset organizers provide the entire dataset only after signing a terms of access, prohibiting commercial use. See the ImageNet clause below (taken on May 5th, 2015).

"ImageNet does not own the copyright of the images. ImageNet only provides thumbnails and URLs of images, in a way similar to what image search engines do. In other words, ImageNet compiles an accurate list of web images for each synset of WordNet. For researchers and educators who wish to use the images for non-commercial research and/or educational purposes, we can provide access through our site under certain conditions and terms."

2. Caffe: Unrestricted Use Deep Learning Models

Now that we have a good idea where to download big visual data and an understanding of the terms that apply, let's take a look at the the other end of the spectrum: the output of the Deep Learning training procedure. We'll take a look at Caffe, one of the most popular Deep Learning libraries, which was engineered to handle ImageNet-like data.  Caffe provides an ecosystem for sharing models (the Model Zoo), and is becoming an indispensable tool for today's computer vision researcher. Caffe is developed at the Berkeley Vision and Learning Center (BVLC) and by community contributors -- it is open source.

Slide from DIY Deep Learning for Vision with Caffe

Problem: As a project that started at a University, Caffe's goal is to be the de-facto standard for creating, training, and sharing Deep Learning models. The shared models were initially licensed for non-commercial use, but the problem is that a new wave of startups is using these techniques, so there must be a licensing agreement which allows Universities, large companies, and startups to explore the same set of pretrained models.

Solution: The current model licensing for Caffe is unrestricted use. This is really great for a broad range of hackers, scientists, and engineers.  The models used to be shared with a non-commercial clause. Below is the entire model licensing agreement from the Model License section of Caffe (taken on May 5th, 2015).

"The Caffe models bundled by the BVLC are released for unrestricted use. 

These models are trained on data from the ImageNet project and training data includes internet photos that may be subject to copyright. 

Our present understanding as researchers is that there is no restriction placed on the open release of these learned model weights, since none of the original images are distributed in whole or in part. To the extent that the interpretation arises that weights are derivative works of the original copyright holder and they assert such a copyright, UC Berkeley makes no representations as to what use is allowed other than to consider our present release in the spirit of fair use in the academic mission of the university to disseminate knowledge and tools as broadly as possible without restriction." 

3. Vision.ai: Dataset generation and training in your home 

Deep Learning learns a summary of the input data, but what happens if a different kind of model memorizes bits and pieces of the training data? And more importantly what if there are things inside the memorized bits which you might not want shared with outsiders?  For this case study, we'll look at Vision.ai, and their real-time computer vision server which is designed to simultaneously create a dataset and learn about an object's appearance. Vision.ai software can be applied to real-time training from videos as well as live webcam streams.

Instead of starting with big visual data collected from internet images (like ImageNet), the vision.ai training procedure is based on a person waving an object of interest in front of the webcam. The user bootstraps the learning procedure with an initial bounding box, and the algorithm continues learning hands-free. As the algorithm learns, it is stores a partial history of what it previously saw, effectively creating its own dataset on the fly. Because the vision.ai convolutional neural networks are designed for detection (where an object only occupies a small portion of the image), there is a large amount of background data presented inside the collected dataset. At the end of the training procedure you get both the Caffe-esque bit (the learned weights) and the ImageNet bit (the collected images). So what happens when it's time to share the model?

A user training a cup detector using vision.ai's real-time detector training interface

Problem: Training in your home means that potentially private and sensitive information is contained inside the backgrounds of the collected images. If you train in your home and make the resulting object model public, think twice about what you're sharing. Sharing can also be problematic if you have trained an object detector from a copyrighted video/images and want to share/sell the resulting model.

Solution: When you save a vision.ai model to disk, you get both a compiled model and the full model. The compiled model is the full model sans the images (thus much smaller). This allows you to maintain fully editable models on your local computer, and share the compiled model (essentially only the learned weights), without the chance of anybody else peeking into your living room. Vision.ai's computer vision server called VMX can run both compiled and uncompiled models; however, only uncompiled models can be edited and extended. In addition, vision.ai provides their vision server as a standalone install, so that all of the training images and computations can reside on your local computer. In brief, vision.ai's solution is to allow you to choose whether you want to run the computations in the cloud or locally, and whether you want to distribute full models (with background images) or the compiled models (solely what is required for detection). When it comes to sharing the trained models and/or created datasets, you are free to choose your own licensing agreement.

4. Open Problems for Licensing Memory-based Machine Learning Models

Deep Learning methods aren't the only techniques applicable to object recognition. What if our model was a Nearest-Neighbor classifier using raw RGB pixels? A Nearest Neighbor Classifier is a memory based classifier which memorizes all of the training data -- the model is the training data. It would be contradictory to license the same set of data differently if one day it was viewed as training data and another day as the output of a learning algorithm. I wonder if there is a way to reconcile the kind of restrictive non-commercial licensing behind ImageNet with the unrestricted licensing use strategy of Caffe Deep Learning Models. Is it possible to have one hacker-friendly data/model license agreement to rule them all?

Conclusion

Don't be surprised if neural network upgrades come as part of your future operating system. As we transition from a data economy (sharing images) to a knowledge economy (sharing neural networks), legal/ownership issues will pop up. I hope that the three scenarios I covered today (big visual data, sharing deep learning models, and training in your home) will help you think about the future legal issues that might come up when sharing visual knowledge. When AI starts generating its own art (maybe by re-synthesizing old pictures), legal issues will pop up. And when your competitor starts selling your models and/or data, legal issues will resurface. Don't be surprised if the MIT license vs. GPL license vs. Apache License debate resurges in the context of pre-trained deep learning models. Who knows, maybe AI Law will become the next big thing.

References
[1] Deep Speech: Accurate Speech Recognition with GPU-Accelerated Deep Learning: NVIDIA dev blog post about Baidu's work on speech recognition using Deep Learning. Andrew Ng is working with Baidu on Deep Learning.

[2] Text Understanding from Scratch: Arxiv paper from Facebook about end-to-end training of text understanding systems using ConvNets. Yann Lecun is working with Facebook on Deep Learning.

[3] ImageNet Classification with Deep Convolutional Neural Networks. Seminal 2012 paper from the Neural Information and Processing Systems (NIPS) conference which showed breakthrough performance from a deep neural network. Paper came out of University of Toronto, but now most of these guys are now at Google.  Geoff Hinton is working with Google on Deep Learning.

[4] J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li and L. Fei-Fei, ImageNet: A Large-Scale Hierarchical Image Database. IEEE Computer Vision and Pattern Recognition (CVPR), 2009.

Jia Deng is now assistant professor at Michigan University and he is growing his research group. If you're interested in starting a PhD in deep learning and vision, check out his call for prospective students. This might be a younger version of Andrew Ng.

Richard Socher is the CTO and Co-Founder of MetaMind, and new startup in the Deep Learning space. They are VC-backed and have plenty of room to grow.

Jia Li is now Head of Research at Snapchat, Inc. I can't say much, but take a look at the recent VentureBeat article: Snapchat is quietly building a research team to do deep learning on images, videos. Jia and I overlapped at Google Research back in 2008.

Fei-Fei Li is currently the Director of the Stanford Artificial Intelligence Lab and the Stanford Vision Lab. See the article on Wired: If we want our machines to think, we need to teach them to see. Yann, you have some competition.

Yangqing Jia created the Caffe project during his PhD at UC Berkeley. He is now a research scientist at Google.

Tomasz Malisiewicz is the Co-Founder of Vision.ai, which focuses on real-time training of vision systems -- something which is missing in today's Deep Learning systems. Come say hi at CVPR.

Deep Learning vs Machine Learning vs Pattern Recognition

Lets take a close look at three related terms (Deep Learning vs Machine Learning vs Pattern Recognition), and see how they relate to some of the hottest tech-themes in 2015 (namely Robotics and Artificial Intelligence). In our short journey through jargon, you should acquire a better understanding of how computer vision fits in, as well as gain an intuitive feel for how the machine learning zeitgeist has slowly evolved over time.

Fig 1. Putting a human inside a computer is not Artificial Intelligence

(Photo from WorkFusion Blog)

If you look around, you'll see no shortage of jobs at high-tech startups looking for machine learning experts. While only a fraction of them are looking for Deep Learning experts, I bet most of these startups can benefit from even the most elementary kind of data scientist. So how do you spot a future data-scientist? You learn how they think. 

The three highly-related "learning" buzz words

“Pattern recognition,” “machine learning,” and “deep learning” represent three different schools of thought.  Pattern recognition is the oldest (and as a term is quite outdated). Machine Learning is the most fundamental (one of the hottest areas for startups and research labs as of today, early 2015). And Deep Learning is the new, the big, the bleeding-edge -- we’re not even close to thinking about the post-deep-learning era.  Just take a look at the following Google Trends graph.  You'll see that a) Machine Learning is rising like a true champion, b) Pattern Recognition started as synonymous with Machine Learning, c) Pattern Recognition is dying, and d) Deep Learning is new and rising fast.

1. Pattern Recognition: The birth of smart programs

Pattern recognition was a term popular in the 70s and 80s. The emphasis was on getting a computer program to do something “smart” like recognize the character "3". And it really took a lot of cleverness and intuition to build such a program. Just think of "3" vs "B" and "3" vs "8".  Back in the day, it didn’t really matter how you did it as long as there was no human-in-a-box pretending to be a machine. (See Figure 1)  So if your algorithm would apply some filters to an image, localize some edges, and apply morphological operators, it was definitely of interest to the pattern recognition community.  Optical Character Recognition grew out of this community and it is fair to call “Pattern Recognition” as the “Smart" Signal Processing of the 70s, 80s, and early 90s. Decision trees, heuristics, quadratic discriminant analysis, etc all came out of this era. Pattern Recognition become something CS folks did, and not EE folks.  One of the most popular books from that time period is the infamous invaluable Duda & Hart "Pattern Classification" book and is still a great starting point for young researchers.  But don't get too caught up in the vocabulary, it's a bit dated.

The character "3" partitioned into 16 sub-matrices. Custom rules, custom decisions, and custom "smart" programs used to be all the rage. 

Quiz: The most popular Computer Vision conference is called CVPR and the PR stands for Pattern Recognition.  Can you guess the year of the first CVPR conference?

2. Machine Learning: Smart programs can learn from examples

Sometime in the early 90s people started realizing that a more powerful way to build pattern recognition algorithms is to replace an expert (who probably knows way too much about pixels) with data (which can be mined from cheap laborers).  So you collect a bunch of face images and non-face images, choose an algorithm, and wait for the computations to finish.  This is the spirit of machine learning.  "Machine Learning" emphasizes that the computer program (or machine) must do some work after it is given data.  The Learning step is made explicit.  And believe me, waiting 1 day for your computations to finish scales better than inviting your academic colleagues to your home institution to design some classification rules by hand.

"What is Machine Learning" from Dr Natalia Konstantinova's Blog. The most important part of this diagram are the "Gears" which suggests that crunching/working/computing is an important step in the ML pipeline.

As Machine Learning grew into a major research topic in the mid 2000s, computer scientists began applying these ideas to a wide array of problems.  No longer was it only character recognition, cat vs. dog recognition, and other “recognize a pattern inside an array of pixels” problems.  Researchers started applying Machine Learning to Robotics (reinforcement learning, manipulation, motion planning, grasping), to genome data, as well as to predict financial markets.  Machine Learning was married with Graph Theory under the brand “Graphical Models,” every robotics expert had no choice but to become a Machine Learning Expert, and Machine Learning quickly became one of the most desired and versatile computing skills.  However "Machine Learning" says nothing about the underlying algorithm.  We've seen convex optimization, Kernel-based methods, Support Vector Machines, as well as Boosting have their winning days.  Together with some custom manually engineered features, we had lots of recipes, lots of different schools of thought, and it wasn't entirely clear how a newcomer should select features and algorithms.  But that was all about to change...

Further reading: To learn more about the kinds of features that were used in Computer Vision research see my blog post: From feature descriptors to deep learning: 20 years of computer vision.

3. Deep Learning: one architecture to rule them all

Fast forward to today and what we’re seeing is a large interest in something called Deep Learning. The most popular kinds of Deep Learning models, as they are using in large scale image recognition tasks, are known as Convolutional Neural Nets, or simply ConvNets. 

ConvNet diagram from Torch Tutorial

Deep Learning emphasizes the kind of model you might want to use (e.g., a deep convolutional multi-layer neural network) and that you can use data fill in the missing parameters.  But with deep-learning comes great responsibility.  Because you are starting with a model of the world which has a high dimensionality, you really need a lot of data (big data) and a lot of crunching power (GPUs). Convolutions are used extensively in deep learning (especially computer vision applications), and the architectures are far from shallow.

If you're starting out with Deep Learning, simply brush up on some elementary Linear Algebra and start coding.  I highly recommend Andrej Karpathy's Hacker's guide to Neural Networks. Implementing your own CPU-based backpropagation algorithm on a non-convolution based problem is a good place to start.

There are still lots of unknowns. The theory of why deep learning works is incomplete, and no single guide or book is better than true machine learning experience.  There are lots of reasons why Deep Learning is gaining popularity, but Deep Learning is not going to take over the world.  As long as you continue brushing up on your machine learning skills, your job is safe. But don't be afraid to chop these networks in half, slice 'n dice at will, and build software architectures that work in tandem with your learning algorithm.  The Linux Kernel of tomorrow might run on Caffe (one of the most popular deep learning frameworks), but great products will always need great vision, domain expertise, market development, and most importantly: human creativity.

Other related buzz-words

Big-data is the philosophy of measuring all sorts of things, saving that data, and looking through it for information.  For business, this big-data approach can give you actionable insights.  In the context of learning algorithms, we’ve only started seeing the marriage of big-data and machine learning within the past few years.  Cloud-computing, GPUs, DevOps, and PaaS providers have made large scale computing within reach of the researcher and ambitious "everyday" developer. 

Artificial Intelligence is perhaps the oldest term, the most vague, and the one that was gone through the most ups and downs in the past 50 years. When somebody says they work on Artificial Intelligence, you are either going to want to laugh at them or take out a piece of paper and write down everything they say.

Further reading: My 2011 Blog post Computer Vision is Artificial Intelligence.

Conclusion

Machine Learning is here to stay. Don't think about it as Pattern Recognition vs Machine Learning vs Deep Learning, just realize that each term emphasizes something a little bit different.  But the search continues.  Go ahead and explore. Break something. We will continue building smarter software and our algorithms will continue to learn, but we've only begun to explore the kinds of architectures that can truly rule-them-all.

If you're interested in real-time vision applications of deep learning, namely those suitable for robotic and home automation applications, then you should check out what we've been building at vision.ai. Hopefully in a few days, I'll be able to say a little bit more. :-)

Until next time.

See discussion about this blog post on Hacker News.