Profile

Cover photo
Greg Linden
4,309 followers|2,515,839 views
AboutPosts+1's

Stream

 
Amusing idea here, can you see what someone is seeing from the outside when they are using Google Glass (or some other wearable computation device)? Turns out you can, though not very well, but well enough that the assumption that it's private and people can't snoop on what you are seeing is probably wrong.
Abstract. Transparent near-eye displays are shipping now for augmented reality applications. In addition to these applications, they promise a private display safe from shoulder surfing. Multiple researchers in the security and HCI communities have proposed systems building on the assumption ...
1
Add a comment...

Greg Linden

Shared publicly  - 
 
Great snippet in this new book on the scientific process: "A scientific approach to the world is fundamentally optimistic and empowering ... The scientific way of looking at the world is founded on the idea that the world is comprehensible ... Scientific thinking turns 'I don't know' into 'I don't know... yet'. The process of science lets you answer just about any question you might ask ... In a very deep way, that's as optimistic as you can get."
2
Greg Linden's profile photo
4 comments
 
For reference, here is a direct link to the snippet in the text of the book:
http://goo.gl/qEE6ac
Add a comment...

Greg Linden

Shared publicly  - 
 
Fascinating analysis of Silk Road, diving into the history of these kinds of markets, and comparing it to other markets where there is anonymity, lack of trust, and lack of information. Some excerpts: "Buyers do not know if they can trust their sellers ... If a seller cheats them, they cannot easily retaliate, since they do not know who the seller is ... Game theory suggests that without the possibility of retaliation, no buyers will enter into business in the first place, since they have every expectation that they will be cheated. There will, in short, be no market ... Introduced an automated rating system ... escrow system ... policing ... getting banned ... increasing levels of bureaucracy and rule‑enforcement ... Scammers gamed the system by establishing themselves as apparently reliable drug dealers, making a large number of near-simultaneous sales, demanding that customers finalise the payment before they got the goods and then disappearing with the money ...  Found himself building a micro-state ... eventually, [even used] the threat of violence against the most dangerous rule‑breakers."
The Silk Road might have started as a libertarian experiment, but it was doomed to end as a fiefdom run by pirate kings
2
Jérôme Flipo's profile photo
Add a comment...

Greg Linden

Shared publicly  - 
 
There's a couple things about Jeff Dean's recent talk (http://goo.gl/hVfIdC) that I wanted to emphasize. First, I absolutely love the way these huge networks and their learning algorithms are robust enough that they can not only handle dropout (randomly dropping nodes, Geoffrey Hinton's crazy but remarkably effective idea) but also the learning or values having extra noise from precision errors or race conditions. That kind of noise tolerance is something that's been missing from AI, but is a core part of biological intelligence, and it's exciting to see it in real world, deployed, state-of-the-art speech and object recognition systems.

Second, I want to reiterate how remarkable it is that these systems use raw data (raw pixels and raw waveforms) as input and have to build their own representations. That's a little like taking a person who's never heard a sound before, not to mention ever hearing language before, and trying to have them learn how to transcribe English speech. It's super hard, crazy difficult, as in almost impossible to believe they are solving that problem that way. Even our own brains appear to have structure that predisposes them to be able to perform at tasks like language understanding. The idea that these networks are able to succeed at this impossibly hard task -- no preprocessing of data, no existing structure to the network, everything from the representation to how to use the representation is learned -- is remarkable.
10
1
Jack Hebert's profile photoGreg Linden's profile photoTau-Mu Yi's profile photoHong Chen's profile photo
5 comments
 
Those are good points of comparison, how well do humans perform, how well do other machine learning methods perform.
Add a comment...
 
Good news: "While she was at Google, Lee became one of the most outspoken corporate lawyers on the problem of patent trolls"
Lee selected after tech companies denounced a candidate from Johnson & Johnson.
12
Add a comment...

Greg Linden

Shared publicly  - 
 
Article out of Netflix on a dull but extremely important problem every large website has, high accuracy near real-time detection of anomalies that should cause more attention (such as paging your human): "Outlier detection can be a pain point for all data driven companies, especially as data volumes grow ... Need for automated anomaly detection tools ensuring data quality and identifying suspicious anomalies."

Nice tidbit here: "We initially tested techniques like moving averages with standard deviations and time series/regression models (ARIMAX) but found that these simpler methods were not robust enough in high cardinality data. The algorithm we finally settled on uses Robust Principal Component Analysis (RPCA) to detect anomalies."
Outlier detection can be a pain point for all data driven companies, especially as data volumes grow. At Netflix we have multiple datasets growing by 10B+ record/day and so there's a need for automated anomaly detection tools ensuring data quality and identifying suspicious anomalies.
2
1
Saravanan Thirumuruganathan's profile photo
Add a comment...

Greg Linden

Shared publicly  - 
 
Nice. Every browser should do this.
Last year, Chrome started to block downloads for applications that "make unexpected changes to your computer — for instance, switching your homepage or other browser settings to ones you don't want". Now Chrome shows a warnin...
4
Add a comment...
Have them in circles
4,309 people
Chris Holloway's profile photo
Nirmalya Rajpandit's profile photo
Harold Mangum's profile photo
Ilya Godunov's profile photo
Michelle Joyce's profile photo
Neeraj Sharma's profile photo
light led's profile photo
Mekwan Bhavesh's profile photo
Thomas Marteau's profile photo

Greg Linden

Shared publicly  - 
 
This seems at least worthy of debate. From Steve Jobs: "You're being vain, you want them to like you ... It's really demeaning to think that, in this deep desire to be liked, you've compromised giving clear, unambiguous feedback."

On a personal note, I tried this over my career, and came away feeling like I was naive and foolish to have done so. Early in my career, I assumed what Steve Jobs said was true, but then I discovered that not only do people think you're a stunning asshole when you're this blunt even if your only goal is to help quickly, but also a large percentage of people ignore and reject all feedback given in this way. So, my view is that Steve Jobs gave some downright awful advice here, but I think it's worthy of mention and discussion anyway.
"You don't care about how they feel! You're being vain, you want them to like you."
8
Jaime Yap's profile photoGreg Linden's profile photoScott Hess's profile photoTau-Mu Yi's profile photo
11 comments
 
I agree with the above.

Steve Jobs was a once-in-a-lifetime genius as a technology entrepreneur. Michael Jordan (heretofore referred to as MJ) was a once-in-a-lifetime genius as a basketball player. MJ would have been a terrible coach and is a terrible owner (by any objective criteria). Steve Jobs would have been a terrible product manager with advice like the above.
Add a comment...

Greg Linden

Shared publicly  - 
 
Nice piece on the state-of-the-art in battery research. Here's an excerpt on the goals: "Right now, batteries cost about $300 per each kiloWatt-hour of capacity. For the two largest use cases (electric vehicles and on-grid storage), we need that figure to drop to about $100 per kW-hr in order for the technology to compete with fossil-fuel-powered cars and generating facilities. For the grid, where the batteries are stationary, it doesn't matter how much they weigh. But for a more effective electric vehicle, we'd like to see the energy density rise from its present 200 W-hr/kg to about 600 W-hr/kg."

There's also this interesting tidbit: "Started looking into lithium-sulfur batteries, which have a theoretical capacity of 2,500 W-hr/kg ... Intermediates called polysulfides that can leak away from the electrode and undergo reactions elsewhere in the battery ... The challenge for her group is getting the polysulfides to react before they can drift off."
Stuffing lithium into a material causes it to expand; can we control it?
24
4
Jérôme Flipo's profile phototimothy dawes's profile photoRenaud Janson's profile photoJohn Sartoris's profile photo
2 comments
 
The technology will have to evolve to survive.

The next step will be to make all this energy storage safe, and not make hundreds of thousands of handy little mobile bombs for terrorists to use for their own twisted means.
Add a comment...

Greg Linden

Shared publicly  - 
 
Jeff Dean of Google will be talking at UW CS in 10 minutes (3:30pm today). Live streaming is available. The talk is titled: "Large-Scale Deep Learning For Building Intelligent Computer Systems"
News & Events · News · Talks & Events · CSE Video · About CSE · People · Faculty · Post-Docs · Staff · Students · Visitors · Education · Courses · Course Videos · Calendars · Paul G. Allen Center · CSE Outreach · Broadening Participation · K-12: DawgBytes · ABET Accreditation ...
9
1
Greg Linden's profile photorif a. saurous's profile photoAleatha Parker-Wood's profile photo
2 comments
 
+Greg Linden: I didn't see the talk. I don't work on Jeff's team, but I do work closely with that infrastructure, so I'm going to offer a bit of context.

I think the idea that "race conditions when apply gradient descent in parallel" don't matter is a bit too strong; I would instead say that we have found a region of parameter space where certain combinations of numbers of replicas and learning rate give good results. The theory doesn't say what that range is, and if you move outside it, you're gonna have a bad time.

Also, models based on pixels are pretty much the whole story in images nowadays, which is a huge shift, but for audio, models based on raw waveforms are a little ways off --- you can sometimes get competitive performance with them compared to spectrogram based features, but the networks are rarely way better like they are with pixels, and they're often bigger/slower. I do think the day when using raw samples for audio is common is coming soon.
Add a comment...

Greg Linden

Shared publicly  - 
 
Worthwhile vision of the future in this video out of Microsoft. Ars' take on it is also a good read: "Everything is a touch screen ... Desks, glass walls, and even foldable, flexible sheets all become screens without any obvious power supply or electronics ... User interfaces are all extremely contextual, adapting perfectly and instantly to match the task at hand ... somehow they manage to consistently show the right information ... It has an elegance that real software, that has to deal with real users and real usage, will always lack ... While it looks cool as hell, it's about a million miles away from ever becoming real."

Ars is right that the AI required to guess exactly what information people want to see (when they might not even know themselves), if even possible, is nowhere near 5-10 years away. And this UI, as shown, requires that AI.

I'd add that the video only shows people consuming information, never producing it. And there's a good reason for that, as it's not clear how this UI, as cool as it is, allows input. Presumably, if asked, the authors would hand wave about voice recognition for input, but that's hard to do and hard to get the UX right. This story is missing a big part of how computing tools will look if it completely leaves out using computing devices for production of information.

Criticisms aside, it's a fun vision of the future. Definitely worth watching and seeing what you think.
Redmond shows us a world in which technology works so much better than it ever does.
1
Ade Oshineye's profile photoGreg Linden's profile photo
3 comments
 
+Ade Oshineye Thanks, that's an excellent article. I particularly liked this part: "Real artists ship, dabblers create concept products ... Pretenders don't quite understand that design is born of constraints ... Constraints have a wonderful way of focusing the mind on the fundamentals, whereas concept products can often have the opposite effect ... Concept products are ... incomplete promises. Shipping products, by contrast, are brutally honest ... You get what’s delivered ... The balance between fantasy and capability."

The article goes on to talk about the iPhone and how much of the design was constrained by battery life. Nice old piece there, thanks for pointing it out.
Add a comment...

Greg Linden

Shared publicly  - 
 
Apple's smart watch strategy, sell a luxury good, not a device: "How do you make a smartwatch appeal to the fashion set? By bringing it into their natural habitat: Ridiculously expensive retailers in fashion hubs ... high-end Paris stores like Galeries Lafayette on the Champs Elysses ... Aiming at the same deep-pocketed consumers that keep the Swiss watch industry alive ... Compete with the likes of Rolex and Richemont ... The Apple Watch will start at $350 ...  the steel watch could go for more than $500, while the gold Apple Watch Edition could go for several thousand dollars."
How do you make a smartwatch appeal to the fashion set? By bringing it into their natural habitat: Ridiculously expensive retailers in fashion hubs. Fren
1
Add a comment...
People
Have them in circles
4,309 people
Chris Holloway's profile photo
Nirmalya Rajpandit's profile photo
Harold Mangum's profile photo
Ilya Godunov's profile photo
Michelle Joyce's profile photo
Neeraj Sharma's profile photo
light led's profile photo
Mekwan Bhavesh's profile photo
Thomas Marteau's profile photo
Links
Contributor to
Story
Tagline
internet relic
Introduction
Greg Linden's +1's are the things they like, agree with, or want to recommend.
Saturday Morning Breakfast Cereal by Zach Weinersmith, February 28, 2015...
www.gocomics.com

One of the many great comics you can read for free at GoComics.com! Follow us for giveaways & giggles.

What Google DeepMind Means for A.I. - The New Yorker
www.newyorker.com

Engineers at Google DeepMind have designed a program that can master vintage video games. How will it shape the future of computer learning?

The Netflix Tech Blog: RAD - Outlier Detection on Big Data
techblog.netflix.com

Outlier detection can be a pain point for all data driven companies, especially as data volumes grow. At Netflix we have multiple datasets g

Chrome's Warning for Sites With Unwanted Software
googlesystem.blogspot.com

Last year, Chrome started to block downloads for applications that "make unexpected changes to your computer — for instance, switching your

Schneier on Security: "Surreptitiously Weakening Cryptographic Systems"
www.schneier.com

Abstract: Revelations over the past couple of years highlight the importance of understanding malicious and surreptitious weakening of crypt

Microsoft, Google Beat Humans at Image Recognition | EE Times
www.eetimes.com

The world's first algorithm to classify images more accurately than humans is claimed by Microsoft.

Schneier on Security: Man-in-the-Middle Attacks on Lenovo Computers
www.schneier.com

Man-in-the-Middle Attacks on Lenovo Computers. It's not just national intelligence agencies that break your https security through man-in-th

Schneier on Security: The Obsolescence of Submarines
www.schneier.com

The Obsolescence of Submarines. Interesting article on the submarine arms race between remaining hidden and detection. It seems that it is m

YouTube Radio
googlesystem.blogspot.com

YouTube has a feature called YouTube Mix, which automatically creates a playlist with many videos related to the video that's currently play

We're Closer Than Ever to a Birth Control Pill for Men | WIRED
www.wired.com

For dudes who like to notch their belts or track miscellaneous pants-related data, the fact that men produce 1,500 sperm every second seems

We Know How You Feel - The New Yorker
www.newyorker.com

Face-scanning software is learning to read emotions, and the business world can’t wait. Raffi Khatchadourian on Affectiva.

Saturday Morning Breakfast Cereal by Zach Weinersmith, February 07, 2015...
www.gocomics.com

One of the many great comics you can read for free at GoComics.com! Follow us for giveaways & giggles.

Download Google Earth Pro for Free
googlesystem.blogspot.com

My Maps Pro is available for free and now Google Earth Pro is also free. Google Earth's business version can be downloaded from this page an

More Related Images in Google Image Search
googlesystem.blogspot.com

When you select a result in Google Image Search, you can now see more related images. Google used to display 8 related images, but now there

Schneier on Security: Operating a Fake Bank
www.schneier.com

Operating a Fake Bank. Here's a story of a fake bank in China -- a real bank, not an online bank -- that stole $32m from depositors over a y