"With 7.2 million images of facial expressions from 31 countries, scientists used computer algorithms to discover that the human face can combine different muscles in different ways to express itself in 16,384 unique ways. When they sorted all these thousands of expressions into categories, the study authors expected to find several hundred variations of emotions. Instead, they were surprised to find only 35.... They concluded that most facial expressions are understood by all, and about half of these expressions are used to express joyfulness. The study found that happiness, whether expressed in a contented smile, glowing cheeks or crinkly eyes, is the face other humans find most recognizable. Researchers note that there are three expressions for fear, four for surprise and five each for sadness and anger. Disgust has only one expression."
From "Reason To Smile: Happiness Is The Most Dominant Human Expression, Study Finds" (in Study Finds).
November 11, 2019
Subscribe to:
Post Comments (Atom)
24 comments:
16384 is 2^14. It means they have 14 muscles, probably. You can do any combination. The model intrudes on reality.
0 would be resting bitch-face.
Go to 850 words and you can encode Ogden's Basic English
http://www.manythings.org/vocabulary/lists/l/words.php?f=ogden
That's without the poetic annex.
16384 is 2^14.
I noticed that, so it sounded suspicious, but didn't think of
It means they have 14 muscles, probably. You can do any combination. The model intrudes on reality.
I don't think I've ever "scowled". But I'm looking forward to it.
Also, one can smile but be a Damned Villain. It denotes Happiness, but happiness at what?
Yeah, we developed all those happy faces as cons of one kind or another.
All that matters are the most-used emojis.
"I don't think I've ever "scowled". But I'm looking forward to it."
Don't! We were advised not to do that growing up in case our face would freeze like that.
16384 is 2^14. It means they have 14 muscles, probably.
I agree. But encoding each muscle as a binary seems rather unsophisticated.
On the other hand...
Human Faces Might Only Express Four Basic Emotions
"How many faces can you make? Offhand, you might guess ten, or twenty, but researchers now say…it’s really only four"
"There are 42 individual facial muscles in the face."
This - with the 2^14 - reinforces my impression that "studyfinds.com" is on the scammy side.
I've worked for a few years now filtering training labels for a few different AI models.
Most of the work is either using basic pivot tables in Excel, DAX, file conversion and maybe some higher level sentence diagramming and statistics, for my part.
Some models work by trying to map the brain, to some extent, but most don't. They simply have to be complex enough to be functional on good data labels, then go on their own. There are all kinds.
Look for more assistive AI as costs come down, so that, say, instead of having 5 HR employees, you might have 3, and 1 (without a data science background), can interface with a dashboard to curate the company's HR model.
Not complicated statistics. Really. People with IQs above 110 or 120 can work on the lower levels of the industry. If you have 130 or above you can hack your way through. If you have 140 or 150 or above and a data science background, you have many, many more options.
A lot of people, though, can find meaningful work, on a continued learning curve amidst a lot of competition. By far the pound for pound 'smartest' industry I've worked in and by far the most objectively fair in terms of skills and merit. Less who you know and much more what you know.
HR is pound for pound the smartest industry you've worked in? Holy smokes man, where else have you worked?
daskol, if you refer to Chris N's comment above, I agree with you. To me, HR seemed to be a low spot in the organization where mediocrities and lefty hacks with an axe to grind settled and made trouble for others, especially hiring managers.
I don't think I ever met an HR type who struck me as having an IQ of 115 (one SD above white people average, about entry level for engineering) much less 130 or above. Those people usually require more interesting work and more money than HR can provide.
OTOH, engineering always seemed highly meritocratic to me. It was, very strictly so, in the departments and labs that I was in charge of. But I was never interested in climbing the greasy pole.
Um, I work for data scientists and developers who are training AI models which will make the interface simple enough that HR departments, for example, will become more obsolete and redundant than many are now in managing a department wide or individual not to answer company wide policies and questions.
I work with sentence diagramming and query intent language models attached to images as well
Bot. The data scientists and natural language programmers are the people I’m referring to.
OK Chris that makes more sense.
Must have been a bug in my NLP neural network or some poorly labeled training data
Lots more than 14 muscles. There was a cliche once-smile, because it only takes 14 muscles to smile, but 72 to frown.
Learn to code is not bad advice, and I second Chris’ boosterism of data science, even if I have a funny way of showing it.
Smile, though your heart is aching
Smile, even though it’s breaking
When there are clouds in the sky
You’ll get by
Fake smiles are hopefully labeled as such in the data, since they’re after the underlying emotion and not the mere physical expression.
No problem. I needed to earn money, and found a fascinating, challenging (rapidly changing) and competitive environment. If I can do it, others can too.
Post a Comment