JoeBaker2000
103912
1913
67
Feb 16, 2021 6:34 AM
JoeBaker2000
103912
1913
67
mystichead
Yaaaa.... This hits a little too close to home @OP
threedeviations
What about 'Uncle' Bob ?
Zalminen
Our kid used to think any hairy animal was a dog. Including men with large beards.
MrCoolWhip
Still learns faster than humans
Piornet
Thanks, Facebook.
JustQuitSmoking
If the data sample was 1. Maybe.
CaptainThePirate
Why is the baby 55?
967Evil
WHAT DOES MUMMY LOOK like ?
craunchey
My son does this anytime he sees a beer bottle.
FiftyShadesOfMascarpone
And that's why we have to endure this crap.
Thickthoughts
Also how a fetish is born
turdfuerguson
minibois
"Please click on all images featuring 'DAD'"
FlyingButtPliers
Insufficient feature vectors for one shot learning of classification algorithm results in poor performance. Better training set required.
LetumComplexo
I disagree, this is a sample of our training accuracy which suggests our model does not have the complexity to encompass the solution space.
LetumComplexo
Growth of a model with greater complexity must occur, or else additional training data will be a wasted expense.
Rowsus
Daddy's gone and made an ass of himself again!
BatmanOSX
HandoB4Javert
noWhiteHorseHereJustBatturuPrinsu
INeverSayNi
somtin
Cause it's gone daddy gone. The love is gone. It's gone daddy gone. The love is gone
Yeah, it's gone daddy gone, The love is gone away!
NotaCPA
Actually pretty close, this is why context is so important with machine learning.
LetumComplexo
Right? I actually think I’ll use this as a visual aid. Should be good for discussion overfitting/underfitting since this could be either.
NotaCPA
awesome, if you have any other information on machine learning I'd be interested. I took a 6week course early pandemic that wasn't useful. 2
NotaCPA
for practical application but I did learn lots
LetumComplexo
So there are a few key insights to machine learning: know the difference between over and under fitting and how to identify both.
LetumComplexo
Especially know how to remedy both. Know your problem space, both input and output, and understand the problem you’re trying to solve.
Qesoiix
What were you thinking... training with a single sample? ?
Katateochi
Friendly tanks on a clear day and all the pics of enemy tanks on cloudy days. The AI ended up just being able to say if it was a nice day 2)
Katateochi
OUAT, the us military tried to train an AI to distinguish between friend and enemy tanks. But the training data had all the pics of 1)
FlyingButtPliers
It could work if dad was the only item in the sample. You don't know what features it will learn from. This is why AI is often racist/sexist
FlyingButtPliers
For instance an early male/female classifier was likely to say anybody standing in a kitchen was a woman because that was the training set.
Nikolai5
Its why there isn't any I in AI. It's just a complicated program.
The701
So are we.
LetumComplexo
We are substantially more complicated. Our most complex AI can’t even come close to capturing the complexity of a human brain. Yet.
FiftyShadesOfMascarpone
We have wants, needs, likes and dislikes. Goals that motivate us and visions we aim to realize. Current AIs don't even have sexual fetishes.
The701
That we know of. Identifying crosswalks may turn them on.
Nikolai5
It really isn't the same. An AI constructed to analyse pictures isn't going to make any real decisions. There is no intelligence in AI.
Jonataurus
It decided what it thinks the images represent. And there are plenty of other types of ai. Most of it has just 1 type of job though.
LetumComplexo
That’s not how you should word it. Modern image classifiers runs on a mutant form of KNN. It’s not deciding what the image represents,
LetumComplexo
rather it’s deciding what images it was trained on are similar and using those HUMAN DEFINED classes. It has no concept of meaning.