recentpopularlog in

jerryking : latency   6

Hey Siri. Why did Apple pay $200m for an AI start-up?
JANUARY 15 2020 | Financial Times | Richard Waters and Patrick McGee in San Francisco.

For Apple, better on-device AI would allow the company’s customers to keep full control of their personal data.

Apple has paid almost $200m for an AI start up, Seattle-based Xnor, that specialises in bringing intelligence to “smart” devices.....Xnor specialises in running complex machine learning models on so-called edge devices — the wide range of gadgets, from smartphones to smart home devices and cars, that operate beyond the reach of the cloud data centres that currently handle most artificial intelligence processing.  Running machine learning on-device, rather than in the cloud, has become one of the most important technology frontiers in the spread of AI. For Apple, better on-device AI would allow the company’s customers to keep full control of their personal data......That has become an important part of the company’s marketing pitch as it tries to distinguish itself from Google and Facebook.

Xnor had developed a way to run large machine learning models without requiring the computing resources and power normally needed for such data-intensive work (e.g. the technology reduces network demands caused by AI aka latency). .....This means that critical applications can continue to run even when they lose a connection to the cloud, such as in driverless cars.
Apple  artificial_intelligence  cloud_computing  connected_devices  data_centers  decentralization  edge  Facebook  Google  latency  M&A  machine_learning  on-device  personal_data  Richard_Waters  Siri  start_ups  Xnor 
10 weeks ago by jerryking
The future of computing is at the edge
June 6, 2018 | FT | by Richard Waters in San Francisco.

With so much data being produced, sending it all to cloud does not make economic sense.

The economics of big data — and the machine learning algorithms that feed on it — have been a gift to the leading cloud computing companies. By drawing data-intensive tasks into their massive, centralised facilities, companies such as Amazon, Microsoft and Google have thrived by bringing down the unit costs of computing.

But artificial intelligence is also starting to feed a very different paradigm of computing. This is one that pushes more data-crunching out to the network “edge” — the name given to the many computing devices that intersect with the real world, from internet-connected cameras and smartwatches to autonomous cars. And it is fuelling a wave of new start-ups which, backers claim, represent the next significant architectural shift in, an early-stage AI software start-up that raised $12m this month, is typical of this new wave. Led by Ali Farhadi, an associate professor at University of Washington, the company develops machine learning algorithms that can be run on extremely low-cost gadgets. Its image recognition software, for instance, can operate on a Raspberry Pi, a tiny computer costing just $5, designed to teach the basics of computer science......That could make it more economical to analyse data on the spot rather than shipping it to the cloud. One possible use: a large number of cheap cameras around the home with the brains to recognise visitors, or tell the difference between a burglar and a cat.

The overwhelming volume of data that will soon be generated by billions of devices such as these upends the logic of data centralisation, according to Mr Farhadi. “We like to say that the cloud is a way to scale AI, but to me it’s a roadblock to AI,” he said. “There is no cloud that can digest this much data.”

“The need for this is being driven by the mass of information being collected at the edge,” added Peter Levine, a partner at Silicon Valley venture capital firm Andreessen Horowitz and investor in a number of “edge” start-ups. “The real expense is going to be shipping all that data back to the cloud to be processed when it doesn’t need to be.”

Other factors add to the attractions of processing data close to where it is collected. Latency — the lag that comes from sending information to a distant data centre and waiting for results to be returned — is debilitating for some applications, such as driverless cars that need to react instantly. And by processing data on the device, rather than sending it to the servers of a large cloud company, privacy is guaranteed.

Tobias Knaup, co-founder of Mesosphere, another US start-up, uses a recent computing truism to sum up the trend: “Data has gravity.”....Nor are the boundaries between cloud and edge distinct. Data collected locally is frequently needed to retrain machine learning algorithms to keep them relevant, a computing-intensive task best handled in the cloud. Companies such as Mesosphere — which raised $125m this month, taking the total to more than $250m — are betting that this will give rise to technologies that move information and applications to where they are best handled, from data centres out to the edge and vice versa...Microsoft unveiled image-recognition software that was capable of running on a local device rather than its own data centres.
Andreessen_Horowitz  artificial_intelligence  centralization  cloud_computing  computer_vision  data_centers  decentralization  edge  future  Industrial_Internet  IT  latency  low-cost  machine_learning  Microsoft  Richard_Waters 
june 2018 by jerryking
Prostate cancer? Relax, and don’t rush your treatment - The Globe and Mail
The Globe and Mail
Published Tuesday, Sep. 20, 2016

a landmark study, published in the New England Journal of Medicine, has provided some stark data on the benefits, risks and necessity of treatment for men with low- or medium-risk prostate cancer (meaning they have a Gleason score of 6-7).

Related: What's the best method of screening for prostate cancer?

The uplifting news is that, a decade after diagnosis, 99 per cent of men with early prostate cancer are still alive. The sort-of-surprising news is that mortality rates don’t really vary depending on type of treatment, or whether a man is treated at all. .....The other element of this story, which is not part of the new research, is about the effectiveness and appropriateness of testing. Another study published recently showed that digital rectal examination is a poor way of detecting prostate cancer and shouldn’t be done because it provides “maximal pain for minimal gain.”

PSA testing, for its part, is one of the most controversial issues in the cancer field. It doesn’t actually detect cancer, but elevated PSA levels trigger biopsies and often lead to a cascade of overtreatment. In Canada, routine PSA tests are not recommended.

What we really need is a test that shows if prostate cancer, once detected, will prove aggressive and deadly or not, and we don’t have that. Prostate cancer kills 4,100 Canadian men a year, but it’s not by doing more and earlier testing and more aggressive treatment that we will necessarily reduce that number. That’s a hard message to digest, and deliver.
cancers  mens'_health  André_Picard  health  PSA  prostate  overtreatment  thinking_deliberatively  reflections  contextual  timeouts  latency 
september 2016 by jerryking
Instant Communication Can Have Bad Consequences — Letters to the editor -
APRIL 9, 2011.
During the American Civil War, Charles Wilkes, a Union naval officer,
broke international law by capturing two Confederate diplomats en route
to Europe on a neutral British ship, the Trent. Adams observed: "When we
so pride ourselves on what we consider the self-evident value of modern
inventions, we may be given pause when we realize that, had there been a
submarine cable in 1861, it is almost certain that England and the
North would have been at war that December. As it was, the slowness of
communication gave both sides time to think, and allowed [Secretary of
State William H.] Seward in America and [Lords] Palmerston and Russell
in England . . . to guide the situation."

"The slowness of communication" is a phrase to savor. Today it is
assumed that speed of communication is an absolute virtue. Combining
speed with a lack of context, electronic media radically undermine
reflection and criticism. We live in a sea of thoughtlessness, informing
ourselves to death.
Communicating_&_Connecting  Peggy_Noonan  power_of_the_pause  letters_to_the_editor  Civil_War  reflections  immediacy  contextual  timeouts  real-time  latency  unintended_consequences  revenge_effects  thinking_deliberatively 
april 2011 by jerryking

Copy this bookmark:

to read