Posts by monty75 296 posts • joined 24 Mar 2010
It's not my definition, its Merriam-Webster's.
The bias is in the training. The GAN learns from the training data what a face looks like. If it sees 90% white faces it will favour white skin in its definition of "face". This isn't a new revelation - it's a well-known phenomenon as the journal article referred to in the Reg article states. It's basically a manifestation of the old maxim "garbage in, garbage out"
"To declare yourself infected, you need to enter a special code from a healthcare provider after testing positive"
So it solves the trolling problem but introduces a bottleneck around the testing capacity which is part of the problem it's supposed to solve. What about the large number of people who contract the disease but have relatively mild symptoms requiring some bed rest and recouperation? Are we expecting everyone who has a bit of a cough to make their way to a testing centre "just in case"?
"We are developing a tool to help triage individuals for COVID-19 testing."
"Dr Deborah Birks, White House coronavirus response coordinator, presented a chart indicating that the alleged website will allow people to log-in through some undisclosed mechanism and check their symptoms."
All of which sounds an awful lot like the one the NHS has been running for weeks now. I bet they managed with considerably fewer than 1700 devs.
They're playing fast and loose with the statistics. They say 70% of wanted people who walked past the camera were matched. How do they know that? That would presume that they know exactly how many wanted people walked by and were able to identify them all by some other means. It doesn't take account of how many people who were on the watchlist sauntered past completely unnoticed by computer and by plod (the "unknown unknowns" in Rumsfeld-speak). What they actually mean is that *at least* 30% of wanted people were not spotted.
As for the 1 in a 1000 false positives, we're meant to take that as meaning it's right 99.9% of the time but it doesn't consider the number of false negatives (wanted people who are not identified) and as I said above that presumes that we know how many wanted people are actually in the crowd to start with.
Then there's the fact that they are talking as if all people are uniformly likely to be picked out in error. As regular Reg readers will know, facial recognition is notoriously bad at identifying non-white people so while the overall false positive rate might be 1 in 1000 that could be something like 1 in 100,000 white people but 1 in 100 or even 1 in 10 black people (depending on the makeup of the crowd). Being stopped every tenth time you step out the house could get really annoying really quickly.
Finally, we are meant to just assume that everyone on the list is there because there is some genuine need for the police to stop them. They don't tell us anything about how accurate and up to date their data is. Sure, they may identify 7 out of 10 people they're looking for but if those people aren't actually wanted by police then the efficacy of the facial recognition system is greatly diminished. GIGO.
Addendum: finally finally, there's no comparison given for this facial recognition system against other methods such as, you know, giving coppers a bunch of mugshots or even just randomly stopping people and fingerprinting them.
My phone and laptop have automatically connected themselves to Eduroam networks at other unis all over the U.K. and also in Germany. It’s not always welcome as the phone sometimes decides to stop using 4G when it can see a weak Eduroam hotspot at some educational establishment or another a mile or so away.
Has Boris Johnson published that report on Russian interference, the one he blocked before the election but promised would be released afterwards? I must have missed it. Maybe Dominic Cummings' russian handlers decided best to keep it under wraps.
It was approved for release almost immediately after the election - odd that isn't it? We still have to wait for the technicality of the appropriate committee being reconstituted before we get to see what has been allowed out unredacted.
There's quite a lot of options. OpenHAB is my personal preference but there's also HomeAssistant and Domoticz. If you want voice control there's Mycroft. Node Red is pretty easy to use for visually programming control logic.
The problem is finding the IoT hardware that works without talking to its own cloud services. My smart plugs have been reflashed with the open source Tasmota firmware but the manufacturer has subsequently patched the hole that enabled that so newly purchased devices would be locked in to their cloud.
I've been getting electronic prescriptions for a couple of years now. When I go the pharmacy they print it out for me to complete the form on the back. All this has achieved is to move the paper from GP to pharmacy. I mean, it's quicker and less effort for me but it's not paperless.
The Register - Independent news and views for the tech community. Part of Situation Publishing
Biting the hand that feeds IT © 1998–2020
