Issue 3
Editorial
Thoughts From Your Humble Curators
What a week in Machine Learning! Last week we saw Waymo high-profile lawsuit against Uber, as well as perhaps the first API against online trolling from Jigsaw. Both events got a lot of media coverage. Both of these events are featured in our News section, with our analysis on it.
On exciting news: GTX 1080 Ti is here yesterday, and featured in this issue. Its spec is more impressive than the $1.2k Titan X, and only costs $699.
In other news, you might have heard of DeepCoder in the last few weeks, and how it purportedly steals and integrates code from other repos. Well, it’s fake. We feature a piece from Stephen Merity which debunks these hyped news.
One must-see this week perhaps is Kleiner Perkins’ Mike Abbott’s interview with Prof. Fei-Fei Li from Stanford. The discussion on how A.I. startups can compete with the larger incumbents is definitely worth watching.
As always, check out our FB group, our YouTube channel.
And of course, please share this newsletter with friends so they can subscribe to this newsletter.
News
Waymo vs Uber, and the “KFC Bucket”
Waymo’s Lawsuit against Uber is perhaps the biggest news last week. It was widely reported by popular outlets – The Wired piece is perhaps the most well-written. The formal complaint is readable and provided more interesting details. We chose the Waymo Medium piece here because it gives you a concrete technical complaints of why Waymo is unhappy. The short answer: “The KFC Bucket”.
Why is this such a big deal? The KFC Bucket” you see on top of Google’s self-driving car is the LiDAR system. This is the “360-degree eye” for the car – it a set of spinning lasers that maps the car’s environment so it knows what’s around. More importantly, it is a VERY critical component of any self-driving car. Waymo, born from Google’s Self-Driving Project, has invested close to 10 years to refine said technology.
Let’s step back a bit: why is LiDAR so important to self-driving car? Generally, self-driving car relies on LiDAR, radar and camera to collect information of its surrounding (Our opinion: audio signals ought to be part of it too). Out of the three, LiDAR is best at providing accurate 3-D representation of the surrounding of a car through laser emission/reflection and you can get information up to 100 meters of your surrounding. From an A.I.-standpoint, such 3-D representation allows better localization of the vehicle, scene understanding, and in turn allows the vehicle plan its movement correctly. In layman’s terms, if you can’t see well, you can’t drive.
Whether LiDAR is crucial to self-driving has always been a question Part of the problem is the prohibitive cost of the device, back in 2013, some quotes suggest it cost up to $80k to include LiDAR into a vehicle.
Then, what is so special about Waymo’s LiDAR system? There are two parts of the answers. First of all, it is patented by Waymo in “Devices and methods for a rotating LIDAR platform with a shared transmit/receive path”, filed back in 2014. Early this year, report suggest that Waymo was able to cut cost of LiDAR by close to 90%. So what Uber allegedly has is not just an abstract design, but a highly cost-effective production-quality design, which presumably is what that 9GB, 14000 files is about.
Why is all this drama relevant to AIDL? Because autonomous vehicles is one of the most clear-cut and self-contained applications of A.I. that is impactful on many levels. It’ll be driven by both innovation and offensive/defensive legal IP positions. For a $60-billion company like Uber, they can afford to litigate. For smaller companies though, as Bryan Walker Smith, a law professor at the University of South Carolina and an expert in self-driving regulations, said in the IEEE piece,
“Companies will discover that trivial yet essential parts of automated driving have already been patented,” …… “Google’s patent for driving on the left side of the lane when passing a truck comes to mind. These kind of patents could stop startups without a large defensive patent portfolio from even entering the field.”
The last question perhaps is who is Anthony Levandowski? And why was he mentioned so many times? Levandowski is a rock star of self-driving car. He built a self-driving motorcycle back in 2004, worked with Sebastian Thrun in 2007. He then formed two companies, one on mobile mapping using LiDAR, the other is a self-driving Prius. Both were acquired by Google and he worked until early 2016.
From this little description, we know Levandowski is an important figure of Google’s effort in self-driving. The Wired piece also painted him as a rule-breaker:
Levandowski has built a reputation for a cavalier approach to rules in general. In December, he insisted Uber’s autonomous cars didn’t need to apply for a special permit under California law and set them loose in San Francisco. The California DMV disagreed and revoked the vehicles’ registrations.
Judging from the complaints, Waymo has evidence on both Levandowski was searching and copying the files, and the fact that he is using the trade secret on Uber’s design. Levandowski’s departure also lead to many ex-employers left and join his startup Otto, which as you know bought by Uber for $680 million price tag. No wonder Waymo filed such explosive lawsuit. Chris Swecker, a former assistant FBI director, would say “I would be very surprised if there wasn’t a full criminal investigation behind this.”
Alphabet’s Jigsaw
There are many Alphabets subsidiary and two of the buzziest ones are DeepMind which is known for its A.I. expertise and Waymo, a self-driving car company. What is Jigsaw then? Jigsaw first started from Google Idea. You can think of it as a think tank of Google, as its “Vision” page said,
We’re an incubator within Alphabet that builds technology to tackle some of the toughest global security challenges facing the world today—from thwarting online censorship to mitigating the threats from digital attacks to countering violent extremism to protecting people from online harassment.
Jigsaw was on the news because of its latest Perspective API which allegedly can determine the toxicity/civility of a comment. Of course, Perspective get a lot of media coverage – faked news is a very big problem in the States, especially given the current polarized partisan political environment.
We tested out the Perspective API with a sentence like “I am not saying this problem is bad, but some of the ideas are just downright stupid.” and I got a 78% score which is “similar to toxic comment”. This is a tough sentence for a non-deep learning system because you need to have a long n-gram to associate “not” with ‘bad’ and since the sentence is long, the toxicity of “downright stupid” may not weigh enough. We guess the system is deep-learning based, maybe recursive neural network or a compositional model.
The Perspective project, unlike other “Graduated projects” of Jigsaw, is still in the development phase. We might look at more updates in the future.
GTX 1080 Ti
A beast! According to NVidia, GTX 1080 Ti is up to 35% faster than 1080. But more impressively it is also purportedly faster than Titan X Pascal. More importantly there is 11G of GPU memory. This is exciting because most consumer-grade cards conventionally have only 8G on-card memory. This is important because many bigger models require more than 8G. e.g. VGGNet training using torch would take more than that.
At the price tag $699, it might be the best card for DIY deep learning hardware for a long time until the 208X series release.
Blog Posts
Stop saying DeepCoder steals code from StackOverflow
This is a direct refutation on why DeepCoder never steals code, written by Stephen Merity, a Senior Researcher of SalesForce Deep Learning team. I (Arthur) was going to write a fact-checking piece about it. With Merity’s piece, now I can only summarize his, and hopefully I can do a good job.
First thing first, DeepCoder is not stealing any code, not from SO, in fact not from any piece of software, unlike what NewScientist’s piece suggest. (Yes. NewScientist.) What was really experimented in DeepCoder’s paper was actually based on large database with toy programming language, which the authors called it “Domain Specific Language” (DSL).
As DeepCoder is not stealing any code from any software, then of course it is not stealing any jobs as well. Because if you want DeepCoder to work in real-life, you have to collect many segments of real programming code, annotate them, and retrain the system. That’s what the authors tried to avoid because it would make the task too difficult.
Of course, if you look at the paper closely, there is also the issues of how to search for the right program because you can imagine the program space is huge. Again, that’s perhaps why the authors focus on a DSL rather than a real language.
The unnecessary media hype aside, the paper does feature a rather interesting neural network architecture. It makes for a genuinely interesting read. His analysis would also be useful to those who want to understand the original paper.
Open Source
Awesome Object Proposals
If you have ever play with deep-learning-based image processing, you would know that image detection is always hairy. And if you work on image detection, you may also know that object proposal (e.g. EdgeBox) is the basis of most image detection systems. In this rare but important “awesome” list, Cuong Cao Pham curated a list of essential resources for object proposal, include important literatures as well as commonly used database.

scan-net
Scannet, collected by Angela Dai and her colleagues, is a dataset specialized on floor plan and allows instance segmentation of objects such as sofa, desk and bed. The data is in RGB-D format with ease-of-use in mind. It has 2.5 Million views in more than 1500 scans. So this dataset is ideal for indoor scene understanding. Also of interest are the paper and the github.
Pretrained Word-vector for 90 languages using fasttext
One of the fastest implementation of word vector generation is Facebook’s fasttext. So this resource is useful for anyone who doesn’t want to train word vector their own and have some assurance of quality.
Self-driving cars in the browser
17-year old Jan Hünermann has created a very impressive self-driving car demo in a browser. All written in Javascript, this 2-D demo includes more advanced technique such as prioritized experience replay buffers. Unlike the Karparthy’s convnetjs though, training has to be done separately, but still it is an impressive and worths your time to take a look.
Video

Boston Dynamics’ Handle
Here is a very impressive video from Boston Dyanmics. This time a robot named Handle. (original video can be found here). It shows impressive balance which is always very difficult to achieve in humanoid robots. The more interesting part perhaps is the use a leg with wheels, which the researchers claim,
Wheels are efficient on flat surfaces while legs can go almost anywhere: by combining wheels and legs Handle can have the best of both worlds.
Indeed. Handle is able to handle many difficult landscape including stairs and snow ground.
How AI Startups Must Compete with Google
Mike Abbott, Partner from Kleiner Perkins, sits with Prof. Fei Fei Li, Associate Professor in the Computer Science Department at Stanford, and lately the director of Google Cloud.
Both Abbott and Li are interesting speakers. For one, Abbott points out that a lot of founders of startup rush into the space, and many don’t know what they are doing. We cannot agree more.
Prof. Li, on the hand, emphasize on the importance of data, and snowballing of data in a startup. That, to us, again nails what is wrong many A.I. startups. One important role in a ML/DL-based company is a data manager. Not many A.I. startups even think about that role. And not many companies have a robust strategy or pipeline to collect more. To us, A.I. startups equal big data startups.
In any case, the whole interview is a lot of fun. So we don’t want to spoil you with other interesting points mentioned by Prof. Li and Abbott.
Finally, as much as we are fans of what A.I. can do, we are also very skeptical about the bulk of all these so-called A.I. startups.