You might think it’s redundant with Instagram Stories, or just don’t want to see high school friends’ boring lives, but ephemeral Snapchat-style Stories now have 500 million daily users across Facebook and Messenger. WhatsApp’s Stories feature Status has 500 million dailies too, and Instagram hit that milestone three months ago. That’s impressive because it means one-third of Facebook’s 1.56 billion daily users are posting or watching Stories each day, up from zero when Facebook launched the feature two years ago.
For reference, Stories inventor Snapchat has just 190 million total daily users.
Facebook isn’t just using Stories to keep people engaged, but to squeeze more cash out of them. Today COO Sheryl Sandberg announced that 3 million advertisers have now bought Stories ads across Facebook’s family of apps. I’d expect Facebook to launch a Stories Ad Network soon so other apps can show Facebook’s vertical video ads and get a cut of the revenue.
Facebook’s aggressive move to clone Snapchat Stories not just in Instagram but everywhere might have pissed users off at first, but many of them have come around. If you give people a place to put their face at their top of their friends’ phones, they’ll fill it. And if someone dangles a window into the lives of people you know and people you wish you did, you’ll open that window regularly.
You might think it’s redundant with Instagram Stories, or just don’t want to see high school friends’ boring lives, but ephemeral Snapchat-style Stories now have 500 million daily users across Facebook and Messenger. WhatsApp’s Stories feature Status has 500 million dailies too, and Instagram hit that milestone three months ago. That’s impressive because it means one-third of Facebook’s 1.56 billion daily users are posting or watching Stories each day, up from zero when Facebook launched the feature two years ago.
For reference, Stories inventor Snapchat has just 190 million total daily users.
Facebook isn’t just using Stories to keep people engaged, but to squeeze more cash out of them. Today COO Sheryl Sandberg announced that 3 million advertisers have now bought Stories ads across Facebook’s family of apps. I’d expect Facebook to launch a Stories Ad Network soon so other apps can show Facebook’s vertical video ads and get a cut of the revenue.
Facebook’s aggressive move to clone Snapchat Stories not just in Instagram but everywhere might have pissed users off at first, but many of them have come around. If you give people a place to put their face at their top of their friends’ phones, they’ll fill it. And if someone dangles a window into the lives of people you know and people you wish you did, you’ll open that window regularly.
The Galaxy Fold has been the most polarizing product I can recall having reviewed. Everyone who saw it wanted to play with the long-promised smartphone paradigm shift. The results, on the other hand, were far more mixed.
If nothing else, the Fold has a remarkably high Q-Rating. Each person who saw me using the product had at least a vague idea of what it was all about. I honestly can’t remember the last time I’ve had that reaction with a non-iPhone device. That’s great from brand perspective. It means a lot of people are curious and potentially open to the notion that the Samsung Galaxy Fold is the future.
Of course, it also means there are a lot of people looking on if you fail.
In some ways, this past week with the Samsung Galaxy Fold has been an extremely public beta. A handful of samples were given out to reviewers. Most worked fine (mine included), but at least three failed. It’s what we in the industry call a “PR nightmare.” Or at least it would be for most companies.
Samsung’s weathered larger storms — most notably with the Galaxy Note 7 a few years back. Of course, that device made it much further along, ultimately resulting in two large-scale recalls. The nature of the two issues was also vastly different. A malfunctioning screen doesn’t put the user at bodily risk like an exploding battery. The optics on these things don’t get much worse than having your smartphone banned from planes.
As of this writing, the Fold is still set to go on sale, most likely this year. To be perfectly frank, the April 26 release date seemed overly optimistic well before the first reports of malfunctioning units. It’s never a great sign when a device is announced in February and is only made available for review a few weeks ahead of launch. It’s kind of like when a studio doesn’t let reviewers watch a film before release. It doesn’t necessarily mean it’s bad, but it’s something to keep an eye on.
That’s the thing. The Galaxy Fold is the kind of device you want badly to succeed. You want it to be great and you want Samsung to sell a billion because it’s a genuinely exciting product after a decade of phones that look mostly the same. There’s also the fact that Samsung has essentially been hyping this thing for eight years, since it debuted a flexible display at CES 2011.
In spite of that, however, the home stretch feels rushed. Samsung no doubt saw the writing on the wall, as companies like Huawei readied their own foldable. And while Royole beat the fold to market, Samsung still had a very good shot at the claim of first commercially viable foldable on the market, with a decade of Galaxy devices under its belt and hand-in-hand work with the Google team to create an Android UX that makes sense on a pair of very different screens.
[Source: iFixit]
But this iFixit teardown speaks volumes. “Alarmingly” isn’t the kind of word you want/expect to hear about a company like Samsung, but there it is, followed directly by “fragile” — itself repeated five times over the course of the write-up. iFixit’s findings match up pretty closely with Samsung’s own reports:
A fragile display means knocking it the wrong way can result in disaster.
A gap in the hinges allows dirt and other particles to wedge themselves between the folding mechanism and screen.
Don’t peel off the protective layer. I know it looks like you should, but this is probably the easiest way to wreck your $2,000 phone that doesn’t involve a firearm or blender.
What makes all of this doubly unfortunate is that Samsung has about as much experience as anyone making a rugged phone that works. I feel confident that the company will do just that in future generations, but unless the company can come back with definitive evidence that it’s overhauled the product ahead of launch, this is a difficult product to recommend.
Samsung knew the first-gen Galaxy Fold would be a hard sell, of course. The company was pretty transparent about the fact that the experimental form factor, coupled with the $1,980 price tag, meant the device will only appeal to a small segment of early adopters.
Even so, the company managed to sell out of preorders — though it didn’t say how large that initial run was. Nor are we sure how many users have canceled in the wake of this past week’s events. Certainly no one would blame them for doing so at this point.
But while the apocalyptic shit-posters among us will declare the death of the foldable before it was ever truly born, whatever doesn’t kill Samsung has only made it stronger. And this misfire could ultimately do that for both the company and the category, courtesy of its informal beta testing.
Rewind a mere week or so ago (seriously, it’s only been that long), when we finally got our hands on the Galaxy Fold. I was impressed. And I certainly wasn’t alone. Admittedly, there’s a bit of a glow that first time you see a device that’s seemingly been teased forever. The fact that it exists feels like a kind of victory in and of itself. But the Fold does an admirable job marrying Samsung’s hardware expertise with a new form factor. And more importantly, it’s real and works as advertised — well, mostly, at least.
The truth is, I’ve mostly enjoyed my time with the Galaxy Fold. And indeed, it’s been fun chronicling it on a (nearly) daily basis. There are some things the form factor is great for — like looking at Google Maps or propping it up to watch YouTube videos on the elliptical machine at the gym. There are others when the bulky form factor left me wanting to go back to my regular old smartphone — but those trade-offs are to be expected.
I both like the Fold’s design and understand the criticism. Samsung’s done a good job maintaining the Galaxy line’s iconic design language. The foldable looks right at home alongside the S and Note. That said, the rounded backing adds some bulk to the product. And while open, the device is thinner than an iPhone, when folded, it’s more than double the thickness, owing to a gap between the displays. It’s quite skinny in this mode, however, so it should slip nicely into all but the tightest pants pockets.
In practice, the folding mechanism might be the most impressive part of the product. The inside features several interlocking gears that allow the product to open and shut with ease and let users interact with the device at various states of unfold. I found myself using the device with it open at a 90-degree angle quite a bit, resting in my hand like an open book. The Fold features a pair of magnets on its edges, which let you close it with a satisfying snap. It’s weirdly therapeutic.
Really, the biggest strike against the device from a purely aesthetic standpoint is that it’s not the Mate X. Announced by Huawei a few days after the Fold’s big unveil, the device takes a decidedly more minimalist approach to the category. It’s an elegant design that features less device and more screen, and, honestly, the kind of thing I don’t think most of us expected until at least the second-generation product.
The gulf between the two devices is especially apparent when it comes to the front screen. The front of the screen is around two-fifths bezel, leaving room for a 4.6-inch display with an awkward aspect ratio. The Mate X, meanwhile, features a 6.6-inch front-facing AND 6.4-inch rear-facing display (not to mention the larger eight-inch internal display to the Fold’s 7.3).
There’s reason to recommend the Fold over the Mate X, as well. I can’t speak to the difference in user experience, having only briefly interacted with the Huawei, but the price point is a biggie. The Mate X starts at an even more absurd $2,600, thanks in part to the fact that it will only be available in a 5G version, adding another layer of niche.
That price, mind you, is converted from euros, because 1) The product was announced at MWC in Barcelona and 2) U.S. availability is likely to be a nonstarter again, as the company continues to struggle with U.S. regulators.
Of course, the Fold’s U.S. availability is also in limbo at the moment, albeit for very different reasons.
I ultimately spent little time interacting with the front screen. It’s good for checking notifications and the like, but attempting to type on that skinny screen is close to impossible, with shades of the new Palm device, which implements its own shortcuts to get around those shortcomings. The inside, meanwhile, takes a butterfly keyboard approach, so you can type with both thumbs while holding it open like a book.
There’s also the issue of app optimization. A lot of this can be chalked up to an early version of a first-gen device. But as with every new device, the equation of how much developer time to invest is largely dependent on product adoption. If the Fold and future Fold’s aren’t a success, developers are going to be far less inclined to invest the hours.
This is most painfully obvious when it comes to App Continuity, one of the device’s primary selling points from a software perspective. When working as advertised, it makes a compelling case for the dual screens. Open something on the front and expand your canvas by unfolding the device. Google is among the companies that worked directly with Samsung to optimize apps this way, and it’s particularly handy with Maps. I used it a fair amount on my trip last week to Berkeley (shout out to the fine people at Pegasus Books on Shattuck).
When an app isn’t optimized, Samsung compels you to restart it, or else you get a nasty case of letterbox bars that retain the aspect ratio of the front screen. Continuity isn’t designed to work the other way, either — opening something on the large screen and then transferring to the front. That’s a bit trickier, as shutting the phone is designed to offer a kind of finality to that session, like hitting the power button to put the device to sleep.
I get that, and like many other pieces here, it will be interesting to see how people utilize it. Aside from the obvious hardware concerns, much of the work on the second-generation device will center around learnings from how users interact with this model. I know I surprised myself when I ended up using the 7.3-inch screen to snap photos. It felt silly — like those people who bring iPads to photograph events. But it’s ultimately a much better viewfinder than that measly 4.6-incher.
That’s really just the tip of the iceberg for the inside screen, of course. The size, which is somewhere between phablet and mini tablet, provides ample real estate that can still be held in one hand. It’s a great size for short videos. I’ve watched a lot of YouTube on this thing, though the speakers (a small series of holes on the upper and lower edges) leave a lot to be desired.
And the seam. I found myself uttering the phrase “it could be worse” a lot. Like so much of the general aesthetic (including the odd green-gold color of my Fold’s casing), it’s lighting-dependent. There are plenty of times when you don’t see it all, and other when the glare hits it and makes it look like a line right down the center.
I realized after snapping a couple of photos that it’s particularly apparent in many shots. That probably gives a false impression of its prominence. It sucks that there’s one at all, but it’s not a surprise, given the nature of the design. You mostly don’t notice it, until your finger swipes across it. And even then it’s subtle and totally not a dealbreaker, unlike, say, the massive gap that made the ZTE Axon M look like two phones pasted together.
I love the ability to stand the device up by having it open at a 90-degree angle, so I can watch videos while brushing my teeth. But this orientation blocks the bottom speakers, hampering the already iffy sound. Thankfully, your $1,980 will get you a pair of the excellent Galaxy Buds in box. It’s hard to imagine Apple bundling AirPods with the next iPhone, but I guess stranger things have happened, right?
Multi-Active Window is the other key software piece. It’s something that has been available on other Samsung devices and certainly makes sense here. Open an app, swipe left from the right side of the screen and a tray will open. From there, you can open up to three apps on the display. Once open, the windows feature a small tab at the top that lets you rearrange them.
It’s handy. I used it the most during those times I had a video playing on an exercise machine, so I didn’t have to close out of everything to check emails and Twitter. I’m a gym multi-tasker. I’m sorry, it’s just who I am now.
It worked quite well on the whole, courtesy of robust internals, including 12GB of RAM and a Snapdragon 855. The primary issue I ran into was how some of the apps maintained that half-screen format after I closed out and reopened. I’m sure some people will prefer that, and I’m honestly not sure what the ideal solution is there.
The Fold’s also got a beefy battery on board. Like Huawei’s, it’s split in two — one on either side of the fold. They work out to a beefy 4,380 mAh. That’s just slightly less than Huawei’s 4,500, but again, the Mate X is 5G by default — which means it’s going to burn through mAhs at a faster rate.
Ultimately, the Fold’s greatest strength is Samsung itself. I understand why you probably just did a double take there in the wake of the company’s latest hardware scandal, but the fact is that the company knows how to build phones. The Fold was very much built atop the foundation of the successful Galaxy line, even while it presents a curious little fork in the family tree.
That means a solid and well-thought-out user experience outside of the whole fold thing.
That list includes great cameras with excellent software features and clever tricks like the new Wireless PowerShare, which lets you fold up the phone and charge up those Galaxy Buds or another phone while it’s plugged in. For better or worse, it also includes Bixby. Our model was a European version that didn’t have the full version, but I think I’ve made my thoughts on the smart assistant pretty well known over the last couple of years.
The devoted Bixby button is very much here. And yes, I very much accidentally pressed it a whole bunch. The headphone jack, on the other hand, is conspicuously absent, which is no doubt a big driver behind the decision to include Galaxy Buds. The Fold is an anomaly in a number of ways, but it’s hard to shake the feeling that this might finally represent the beginning of the end for the port on Samsung’s premium devices.
Also absent is the S Pen. The stylus began life on the Note line and has since branched out to other Samsung devices. I suspect the company would have had a tough time squeezing in space for it alongside the dual batteries, and maybe it’s saving something for future generations, but this does feel like the ideal screen size for that accessory.
I’m parting ways with the Fold this week, per Samsung’s instructions. Unlike other products, giving it up won’t feel that tough. There wasn’t a point in the past week when the Fold didn’t feel like overkill. There were, however, times when my iPhone XS screen felt downright tiny after switching back.
In many ways, the foldable phone still feels like the future, and the Fold feels like a stop along the way. There are a lot of first-gen issues that should be/should have been hammered out before mass producing this device. That said, there are certain aspects that can only really be figured out in real-world testing. Take the fact that Samsung subjected the device to 200,000 mechanical open and closes. That’s a lot, and probably more than the life of just about any of these devices, but people don’t open and close like machines. And when it comes to the screen, well, a little dirt is bound to get between the gears, both metaphorically and literally.
As I close this Galaxy Fold a final time, it seems safe to say that the device represents a potentially exciting future for a stagnant smartphone space. But that’s the thing about the future — it’s just not here yet.
A massive penalty hangs over Facebook’s head, but it otherwise had a very strong Q1 earnings report. Facebook reached 2.38 billion monthly users, up 2.5 percent from 2.32 billion in Q4 2018 when it grew 2.2 percent, and it now has 1.56 billion daily active users, up 2.63 percent from 1.52 billion last quarter when it grew 2 percent. Facebook pulled in $15.08 in revenue, up 26 percent year-over-year compared to Refinitiv’s consensus estimates of $14.98 billion in revenue.
Facebook recorded earnings per share of $0.85 compared to estimates of $1.63 EPS. However, that’s because Facebook has set aside $3 billion to cover a potential FTC fine that it’s still resolving. Without that fine, it would have had an EPS of $1.89. Despite the set-aside, Facebook still earned $2.429 billion in profit, though that’s down from $4.988 a year ago and $6.8 billion in Q4 2018.
Facebook’s share price rose 4.89 percent to $191.50 after closing before earnings at $182.58, way up from its recent low of $124.06 in December. Wall Street seems to have already priced in the potential FTC fine. Facebook has agreed to strict oversight of how it handled user privacy in a 2011 deal with the FTC. It promised to not misrepresent its privacy practices or change privacy controls without user permission, and it’s now negotiating the fine for potentially breaking those terms.
Facebook wrote in its earnings release about the FTC fine that:
“In the first quarter of 2019, we reasonably estimated a probable loss and recorded an accrual of $3.0 billion in connection with the inquiry of the FTC into our platform and user data practices, which accrual is included in accrued expenses and other current liabilities on our condensed consolidated balance sheet. We estimate that the range of loss in this matter is $3.0 billion to $5.0 billion. The matter remains unresolved, and there can be no assurance as to the timing or the terms of any final outcome.”
It’s possible Facebook escapes with a lesser fine that would likely still dwarf Google’s $22.5 million penalty for violating an FTC privacy deal. But it also might have to drag down a future quarter of earnings if the fine ranges as high as $5 billion or larger. Though Facebook does have $45.2 billion in cash and securities on hand to pay that fine and make any necessary acquisitions. Facebook’s headcount grew 36% year-over-year to 37,773 as it staffs up its security team, but it still has a 22 percent operating margin.
Facebook has managed to hold on to its 66 percent daily to monthly user ratio, showing people aren’t necessarily using it less despite all the backlash. But Facebook failed to grow past its 186 million daily user count in the US & Canada where it got stuck last quarter, but at least it added 4 million in its lucrative Europe market, plus it had atypically large gains in Asia-Pacific and the Rest Of World regions. As for monetization, Facebook made modest gains in average revenue per user across markets compared to Q3 2018 (excluding the holiday-laden Q4). Europe did especially well, growing ARPU 8.2 percent.
Zooming out, Facebook now has over 2.7 billion total mothly users across its family of Facebook, Messenger, Instagram, and WhatsApp, the same as last quarter. 2.1 billion people use at least one of those apps daily, up from 2 billion last quarter.
This is the first earnings report of a full quarter following Facebook’s worst-ever security breach in September that impacted 50 million users, shaking confidence in the social network’s privacy and security. It’s also the first full quarter in which Facebook sold its own branded hardware — its Portal video chat device that was well received by critics except for the fact that it was made by Facebook.
In March, CEO Mark Zuckerberg announced plans for a massive privacy-centric overhaul of Facebook to turn it from just a townsquare into also a “living room”. That means unifying its messaging apps with a backend that supports end-to-end encryption, and promoting ephemerality in content sharing and communication. That could help deter calls for regulation, make Facebook harder to break up, and help it stay ahead of competitors like Snapchat, but will also be a massive product and engineering undertaking.
Yet the defining story continues to be Facebook’s struggle with claims that its developer platform endangered user privacy and steamrolled competitors in search of growth. The fact that Facebook isn’t losing massive numbers of users after years of sustained scandals is a testament to how deeply it’s woven itself into people’s lives.
A massive penalty hangs over Facebook’s head, but it otherwise had a very strong Q1 earnings report. Facebook reached 2.38 billion monthly users, up 2.5 percent from 2.32 billion in Q4 2018 when it grew 2.2 percent, and it now has 1.56 billion daily active users, up 2.63 percent from 1.52 billion last quarter when it grew 2 percent. Facebook pulled in $15.08 in revenue, up 26 percent year-over-year compared to Refinitiv’s consensus estimates of $14.98 billion in revenue.
Facebook recorded earnings per share of $0.85 compared to estimates of $1.63 EPS. However, that’s because Facebook has set aside $3 billion to cover a potential FTC fine that it’s still resolving. Without that fine, it would have had an EPS of $1.89. Despite the set-aside, Facebook still earned $2.429 billion in profit, though that’s down from $4.988 a year ago and $6.8 billion in Q4 2018.
Facebook’s share price rose 4.89 percent to $191.50 after closing before earnings at $182.58, way up from its recent low of $124.06 in December. Wall Street seems to have already priced in the potential FTC fine. Facebook has agreed to strict oversight of how it handled user privacy in a 2011 deal with the FTC. It promised to not misrepresent its privacy practices or change privacy controls without user permission, and it’s now negotiating the fine for potentially breaking those terms.
Facebook wrote in its earnings release about the FTC fine that:
“In the first quarter of 2019, we reasonably estimated a probable loss and recorded an accrual of $3.0 billion in connection with the inquiry of the FTC into our platform and user data practices, which accrual is included in accrued expenses and other current liabilities on our condensed consolidated balance sheet. We estimate that the range of loss in this matter is $3.0 billion to $5.0 billion. The matter remains unresolved, and there can be no assurance as to the timing or the terms of any final outcome.”
It’s possible Facebook escapes with a lesser fine that would likely still dwarf Google’s $22.5 million penalty for violating an FTC privacy deal. But it also might have to drag down a future quarter of earnings if the fine ranges as high as $5 billion or larger. Though Facebook does have $45.2 billion in cash and securities on hand to pay that fine and make any necessary acquisitions. Facebook’s headcount grew 36% year-over-year to 37,773 as it staffs up its security team, but it still has a 22 percent operating margin.
Facebook has managed to hold on to its 66 percent daily to monthly user ratio, showing people aren’t necessarily using it less despite all the backlash. But Facebook failed to grow past its 186 million daily user count in the US & Canada where it got stuck last quarter, but at least it added 4 million in its lucrative Europe market, plus it had atypically large gains in Asia-Pacific and the Rest Of World regions. As for monetization, Facebook made modest gains in average revenue per user across markets compared to Q3 2018 (excluding the holiday-laden Q4). Europe did especially well, growing ARPU 8.2 percent.
Zooming out, Facebook now has over 2.7 billion total mothly users across its family of Facebook, Messenger, Instagram, and WhatsApp, the same as last quarter. 2.1 billion people use at least one of those apps daily, up from 2 billion last quarter.
This is the first earnings report of a full quarter following Facebook’s worst-ever security breach in September that impacted 50 million users, shaking confidence in the social network’s privacy and security. It’s also the first full quarter in which Facebook sold its own branded hardware — its Portal video chat device that was well received by critics except for the fact that it was made by Facebook.
In March, CEO Mark Zuckerberg announced plans for a massive privacy-centric overhaul of Facebook to turn it from just a townsquare into also a “living room”. That means unifying its messaging apps with a backend that supports end-to-end encryption, and promoting ephemerality in content sharing and communication. That could help deter calls for regulation, make Facebook harder to break up, and help it stay ahead of competitors like Snapchat, but will also be a massive product and engineering undertaking.
Yet the defining story continues to be Facebook’s struggle with claims that its developer platform endangered user privacy and steamrolled competitors in search of growth. The fact that Facebook isn’t losing massive numbers of users after years of sustained scandals is a testament to how deeply it’s woven itself into people’s lives.
Braille is a crucial skill for children with visual impairments to learn, and with these LEGO Braille Bricks kids can learn through hands-on play rather than more rigid methods like Braille readers and printouts. Given the naturally Braille-like structure of LEGO blocks, it’s surprising this wasn’t done decades ago.
The truth is, however, that nothing can be obvious enough when it comes to marginalized populations like people with disabilities. But sometimes all it takes is someone in the right position to say “You know what? That’s a great idea and we’re just going to do it.”
It happened with the BecDot (above) and it seems to have happened at LEGO. Stine Storm led the project, but Morten Blonde, who himself suffers from degenerating vision, helped guide the team with the passion and insight that only comes with personal experience.
In some remarks sent over by LEGO, Blonde describes his drive to help:
When I was contacted by the LEGO Foundation to function as internal consultant on the LEGO Braille Bricks project, and first met with Stine Storm, where she showed me the Braille bricks for the first time, I had a very emotional experience. While Stine talked about the project and the blind children she had visited and introduced to the LEGO Braille Bricks I got goose bumps all over the body. I just knew that I had to work on this project.
I want to help all blind and visually impaired children in the world dare to dream and see that life has so much in store for them. When, some years ago, I was hit by stress and depression over my blind future, I decided one day that life is too precious for me not to enjoy every second of. I would like to help give blind children the desire to embark on challenges, learn to fail, learn to see life as a playground, where anything can come true if you yourself believe that they can come true. That is my greatest ambition with my participation in the LEGO Braille Bricks project
The bricks themselves are very like the originals, specifically the common 2×4 blocks, except they don’t have the full 8 “studs” (so that’s what they’re called). Instead, they have the letters of the Braille alphabet, which happens to fit comfortably in a 2×3 array of studs, with room left on the bottom to put a visual indicator of the letter or symbol for sighted people.
It’s compatible with ordinary LEGO bricks and of course they can be stacked and attached themselves, though not with quite the same versatility as an ordinary block, since some symbols will have fewer studs. You’ll probably want to keep them separate, since they’re more or less identical unless you inspect them individually.
All told the set, which will be provided for free to institutions serving vision-impaired students, will include about 250 pieces: A-Z (with regional variants), the numerals 0-9, basic operators like + and =, and some “inspiration for teaching and interactive games.” Perhaps some specialty pieces for word games and math toys, that sort of thing.
LEGO was already one of the toys that can be enjoyed equally by sighted and vision-impaired children, but this adds a new layer, or I suppose just re-engineers an existing and proven one, to extend and specialize the decades-old toy for a group that already seems already to have taken to it:
“The children’s level of engagement and their interest in being independent and included on equal terms in society is so evident. I am moved to see the impact this product has on developing blind and visually impaired children’s academic confidence and curiosity already in its infant days,” said Blonde.
Danish, Norwegian, English, and Portuguese blocks are being tested now, with German, Spanish and French on track for later this year. The kit should ship in 2020 — if you think your classroom could use these, get in touch with LEGO right away.
Posted by Olivier Bachem, Research Scientist, Google AI Zürich
The ability to understand high-dimensional data, and to distill that knowledge into useful representations in an unsupervised manner, remains a key challenge in deep learning. One approach to solving these challenges is through disentangled representations, models that capture the independent features of a given scene in such a way that if one feature changes, the others remain unaffected. If done successfully, a machine learning system that is designed to navigate the real world, such as a self driving car or a robot, can disentangle the different factors and properties of objects and their surroundings, enabling the generalization of knowledge to previously unobserved situations. While, unsupervised disentanglement methods have already been used for curiosity driven exploration, abstract reasoning, visual concept learning and domain adaptation for reinforcement learning, recent progress in the field makes it difficult to know how well different approaches work and the extent of their limitations.
In "Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations" (to appear at ICML 2019), we perform a large-scale evaluation on recent unsupervised disentanglement methods, challenging some common assumptions in order to suggest several improvements to future work on disentanglement learning. This evaluation is the result of training more than 12,000 models covering most prominent methods and evaluation metrics in a reproducible large-scale experimental study on seven different data sets. Importantly, we have also released both the code used in this study as well as more than 10,000 pretrained disentanglement models. The resulting library, disentanglement_lib, allows researchers to bootstrap their own research in this field and to easily replicate and verify our empirical results.
Understanding Disentanglement
To better understand the ground-truth properties of an image that can be encoded in a disentangled representation, first consider the ground-truth factors of the data set Shapes3D. In this toy model, shown in the figure below, each panel represents one factor that could be encoded into a vector representation of the image. The model shown is defined by the shape of the object in the middle of the image, its size, the rotation of the camera and the color of the floor, the wall and the object.
Visualization of the ground-truth factors of the Shapes3D data set: Floor color (upper left), wall color (upper middle), object color (upper right), object size (bottom left), object shape (bottom middle), and camera angle (bottom right).
The goal of disentangled representations is to build models that can capture these explanatory factors in a vector. The figure below presents a model with a 10-dimensional representation vector. Each of the 10 panels visualizes what information is captured in one of the 10 different coordinates of the representation. From the top right and the top middle panel we see that the model has successfully disentangled floor color, while the two bottom left panels indicate that object color and size are still entangled.
Visualization of the latent dimensions learned by a FactorVAE model (see below). The ground-truth factors wall and floor color as well as rotation of the camera are disentangled (see top right, top center and bottom center panels), while the ground-truth factors object shape, size and color are entangled (see top left and the two bottom left images).
Key Results of this Reproducible Large-scale Study
While the research community has proposed a variety of unsupervised approaches to learn disentangled representations based on variational autoencoders and has devised different metrics to quantify their level of disentanglement, to our knowledge no large-scale empirical study has evaluated these approaches in a unified manner. We propose a fair, reproducible experimental protocol to benchmark the state of unsupervised disentanglement learning by implementing six different state-of-the-art models (BetaVAE, AnnealedVAE, FactorVAE, DIP-VAE I/II and Beta-TCVAE) and six disentanglement metrics (BetaVAE score, FactorVAE score, MIG, SAP, Modularity and DCI Disentanglement). In total, we train and evaluate 12,800 such models on seven data sets. Key findings of our study include:
We do not find any empirical evidence that the considered models can be used to reliably learn disentangled representations in an unsupervised way, since random seeds and hyperparameters seem to matter more than the model choice. In other words, even if one trains a large number of models and some of them are disentangled, these disentangled representations seemingly cannot be identified without access to ground-truth labels. Furthermore, good hyperparameter values do not appear to consistently transfer across the data sets in our study. These results are consistent with the theorem we present in the paper, which states that the unsupervised learning of disentangled representations is impossible without inductive biases on both the data set and the models (i.e., one has to make assumptions about the data set and incorporate those assumptions into the model).
For the considered models and data sets, we cannot validate the assumption that disentanglement is useful for downstream tasks, e.g., that with disentangled representations it is possible to learn with fewer labeled observations.
The figure below demonstrates some of these findings. The choice of random seed across different runs has a larger impact on disentanglement scores than the model choice and the strength of regularization (while naively one might expect that more regularization should always lead to more disentanglement). A good run with a bad hyperparameter can easily beat a bad run with a good hyperparameter.
The violin plots show the distribution of FactorVAE scores attained by different models on the Cars3D data set. The left plot shows how the distribution changes as different disentanglement models are considered while the right plot displays the different distributions as the regularization strength in a FactorVAE model is varied. The key observation is that the violin plots substantially overlap which indicates that all methods strongly depend on the random seed.
Based on these results, we make four observations relevant to future research:
Given the theoretical result that the unsupervised learning of disentangled representations without inductive biases is impossible, future work should clearly describe the imposed inductive biases and the role of both implicit and explicit supervision.
Finding good inductive biases for unsupervised model selection that work across multiple data sets persists as a key open problem.
The concrete practical benefits of enforcing a specific notion of disentanglement of the learned representations should be demonstrated. Promising directions include robotics, abstract reasoning and fairness.
Experiments should be conducted in a reproducible experimental setup on a diverse selection of data sets.
Open Sourcing disentanglement_lib
In order for others to verify our results, we have released disentanglement_lib, the library we used to create the experimental study. It contains open-source implementations of the considered disentanglement methods and metrics, a standardized training and evaluation protocol, as well as visualization tools to better understand trained models.
The advantages of this library are three-fold. First, with less than four shell commands disentanglement_lib can be used to reproduce any of the models in our study. Second, researchers may easily modify our study to test additional hypotheses. Third, disentanglement_lib is easily extendible and can be used to bootstrap research into the learning of disentangled representations—it is easy to implement new models and compare them to our reference implementation using a fair, reproducible experimental setup.
Reproducing all the models in our study requires a computational effort of approximately 2.5 GPU years, which can be prohibitive. So, we have also released >10,000 pretrained disentanglement_lib models from our study that can be used together with disentanglement_lib.
We hope that this will accelerate research in this field by allowing other researchers to benchmark their new models against our pretrained models and to test new disentanglement metrics and visualization approaches on a diverse set of models.
Acknowledgments This research was done in collaboration with Francesco Locatello, Mario Lucic, Stefan Bauer, Gunnar Rätsch, Sylvain Gelly and Bernhard Schöpf at Google AI Zürich, ETH Zürich and the Max-Planck Institute for Intelligent Systems. We also wish to thank Josip Djolonga, Ilya Tolstikhin, Michael Tschannen, Sjoerd van Steenkiste, Joan Puigcerver, Marcin Michalski, Marvin Ritter, Irina Higgins and the rest of the Google Brain team for helpful discussions, comments, technical help and code contributions.
Over the course of the last year or so, Google started adding more job search features to its search engine. Today, it’s expanding this program by also making it easier to find work-from-home jobs.
Typically, when you searched for a job like that, chances are you’ll be bombarded with a bunch of low-quality or even potentially malicious sites. Now, when you search for a job in Google Search using a query like ‘customer support jobs,’ you’ll be able to set the location to ‘work from home’ to find remote and telecommute jobs. Google says that a number of job listing sites, including Working Nomads, We Work Remotely and ZipRecruiter already support this feature. To filter these jobs, Google looks at the standard Schema.org markup that most job sites use to push their listings to Google Jobs.
“Through a number of user studies, we learned that most remote jobs were not clearly labeled as such,” Google product manager Jennifer Su writes in today’s announcement. “And our customers have told us it is difficult to accurately detect and return remote opportunities when users enter ‘work from home’ or related queries into their search box. By offering this capability to our customers, we hope to make remote work opportunities more discoverable to job seekers who need them.”
Google already lets job searchers filter by commute times and type of transit, so it makes sense to add this work-from-home option, too, especially now that more employees are looking for this kind of flexibility.
It’s been a month since Huawei unveiled its latest flagship device — the Huawei P30 Pro. I’ve played with the P30 and P30 Pro for a few weeks and I’ve been impressed with the camera system.
The P30 Pro is the successor to the P20 Pro and features improvements across the board. It could have been a truly remarkable phone, but some issues still hold it back compared to more traditional Android phones, such as the Google Pixel 3 or OnePlus 6T.
A flagship device
The P30 Pro is by far the most premium device in the P line. It features a gigantic 6.47-inch OLED display, a small teardrop notch near the top, an integrated fingerprint sensor in the display and a lot of cameras.
Before diving into the camera system, let’s talk about the overall feel of the device. Compared to last year’s P20 Pro, the company removed the fingerprint sensor at the bottom of the screen and made the notch smaller. The fingerprint sensor doesn’t perform as well as a dedicated fingerprint sensor, but it gets the job done.
It has become hard to differentiate smartphones based on design as it looks a lot like the OnePlus 6T or the Samsung Galaxy S10. The display features a 19.5:9 aspect ratio with a 2340×1080 resolution, and it is curved around the edges.
The result is a phone with gentle curves. The industrial design is less angular, even though the top and bottom edges of the device have been flattened. Huawei uses an aluminum frame and a glass with colorful gradients on the back of the device.
Unfortunately, the curved display doesn’t work so well in practice. If you open an app with a unified white background, such as Gmail, you can see some odd-looking shadows near the edges.
Below the surface, the P30 Pro uses a Kirin 980 system-on-a-chip. Huawei’s homemade chip performs well. To be honest, smartphones have been performing well for a few years now. It’s hard to complain about performance anymore.
The phone features a headphone jack, a 40W USB-C charging port and an impressive 4,200 mAh battery. For the first time, Huawei added wireless charging to the P series (up to 15W).
You can also charge another phone or an accessory with reverse wireless charging, just like on the Samsung Galaxy S10. Unfortunately, you have to manually activate the feature in the settings every time you want to use it.
Huawei has also removed the speaker grill at the top of the display. The company now vibrates the screen in order to turn the screen into a tiny speaker for your calls. In my experience, it works well.
While the phone ships with Android Pie, Huawei still puts a lot of software customization with its EMUI user interface. There are a dozen useless Huawei apps that probably make sense in China, but don’t necessarily need to be there if you use Google apps.
For instance, the HiCare app keeps sending me notifications. The onboarding process is also quite confusing as some screens refer to Huawei features while others refer to standard Android features. It definitely won’t be a good experience for non tech-savvy people.
(P30 Pro on the left, P30 on the right)
Four cameras to rule them all
The P20 Pro already had some great camera sensors and paved the way for night photos in recent Android devices. The P30 Pro camera system can be summed up in two words — more and better.
The P30 Pro now features not one, not two, not three but f-o-u-r sensors on the back of the device.
The main camera is a 40 MP 27mm sensor with an f/1.6 aperture and optical image stabilization.
There’s a 20 MP ultra-wide angle lens (16mm) with an f/2.2 aperture.
The 8 MP telephoto lens provides nearly 5x optical zoom compared to the main lens (125mm) with an f/3.4 aperture and optical image stabilization.
There’s a new time-of-flight sensor below the flash of the P30 Pro. The phone projects infrared light and captures the reflection with this new sensor.
It has become a sort of a meme already — yes, the zoom works incredibly well on the P30 Pro. In addition to packing a lot of megapixels in the main sensor, the company added a telephoto lens with a periscope design. The sensor features a mirror to beam the light at a right angle and put more layers of glass in the sensor without making the phone too thick.
The company also combines the main camera sensor with the telephoto sensor to let you capture photos with a 10x zoom with a hybrid digital-optical zoom.
Here’s a photo series with the wide angle lens, the normal lens, a 5x zoom and a 10x zoom:
And it works incredibly well in daylight. Unfortunately, you won’t be able to use the telephoto lens at night as it doesn’t perform as well as the main camera.
In addition to hardware improvements, Huawei has also worked on the algorithms that process your shots. Night mode performs incredibly well. You just have to hold your phone for 8 seconds so that it can capture as much light as possible. Here’s what it looks like in a completely dark room vs. an iPhone X:
Huawei has also improved HDR processing and portrait photos. That new time-of-flight sensor works well when it comes to distinguishing a face from the background for instance.
Once again, Huawei is a bit too heavy-handed with post-processing. If you use your camera with the Master AI setting, colors are too saturated. The grass appears much greener than it is in reality. Skin smoothing with the selfie camera still feels weird too. The phone also aggressively smoothes surfaces on dark shots.
When you pick a smartphone brand, you also pick a certain photography style. I’m not a fan of saturated photos, so Huawei’s bias toward unnatural colors doesn’t work in my favor.
But if you like extremely vivid shots with insanely good sensors the P30 Pro is for you. That array of lenses also opens up a lot of possibilities and gives you more flexibility.
Fine prints
The P30 Pro isn’t available in the U.S. But the company has already covered the streets of major European cities with P30 Pro ads. It costs P30 Pro for €999 ($1,130) for 128GB of storage — there are more expensive options with more storage.
Huawei also unveiled a smaller device — the P30. It’s always interesting to look at the compromises of the more affordable model.
On that front, there’s a lot to like about the P30. For €799 ($900) with 128GB, you get a solid phone. It has a 6.1-inch OLED display and shares a lot of specifications with its bigger version.
The P30 features the same system-on-a-chip, the same teardrop notch, the same fingerprint sensor in the display, the same screen resolution. Surprisingly, the P30 Pro doesn’t have a headphone jack while the P30 has one.
There are some things you won’t find on the P30, such as wireless charging or the curved display. While the edges of the device are slightly curved, the display itself is completely flat. And I think it looks better.
Cameras are slightly worse on the P30, and you won’t be able to zoom in as aggressively. Here’s the full rundown:
A 40 MP main sensor with an f/1.8 aperture and optical image stabilization.
A 16 MP ultra-wide angle lens with an f/2.2 aperture.
An 8 MP telephoto lens that should provide 3x optical zoom.
No time-of-flight sensor.
In the end, it really depends what you’re looking for. The P30 Pro definitely has the best cameras of the P series. But the P30 is also an attractive phone for those looking for a smaller device.
Huawei has once again pushed the limits of what you can pack in a smartphone when it comes to cameras. While iOS and Android are more mature than ever, it’s fascinating to see that hardware improvements are not slowing down.
