|
animist posted:okay, so, first thing you gotta understand is what people mean by "deep neural networks". a neural network is made up of "neurons", which are functions of weighted sums. thats it. here's a "neuron": hey now posting convnets is cheating
|
# ¿ May 31, 2019 19:07 |
|
|
# ¿ May 11, 2024 22:31 |
|
also i feel bad for recognizing some of those from illustrations alone
|
# ¿ May 31, 2019 19:14 |
|
it's my world now fortunately/unfortunately
|
# ¿ May 31, 2019 21:17 |
|
tbh that can be said of pretty much any technology
|
# ¿ Jun 1, 2019 04:52 |
|
eschaton posted:if I were asked to make an “object recognizer,” I wouldn’t train one huge network on a million images of the object I want to recognize and allow steganography to break everything, I’d train a large number of smaller networks on different characteristics to recognize, and use additional separate networks to determine confidence according to recognitions, etc. finally arriving at the one confidence value This is actually basically Fast R-CNN - except object classification was done by SVMs instead of NNs really all the convnet did for that is feature encoding it’s called fast rcnn but it’s pretty slow compared to the approaches that supplanted it
|
# ¿ Jun 1, 2019 13:53 |
|
oh and also fast rcnn was doing that all given that you were already giving it proposals for where objects in the image were
|
# ¿ Jun 1, 2019 13:57 |
|
well resnet is more of a particular convnet architecture, like the densenet illustration at the end of that earlier post, than a particular machine vision objective
|
# ¿ Jun 1, 2019 14:53 |
|
like pre-convnet image classification and object detection (which are overlapping tasks really) was all about using something like sift vectors, which was pretty much taking an image and encoding it into this vector where each dimension represented some particular human chosen function output and you hoped that the vectors were distinct enough that images of different things would encode distinctly but that also the encoding was general/robust enough that changes in pose or scale/distance wouldn’t prevent things from matching up so an objective like object detection was a bunch of sliding various windows of subsections of the image and searching your database of sift vectors to see if anything matched and of course there was research into how to partition and search through that database of sift vectors faster etc
|
# ¿ Jun 1, 2019 15:05 |
|
rchon posted:don't ask me about deploying mask r-cnn models to production systems with no quantifiable validation metrics or model versioning bonus if it’s just the Facebook repo
|
# ¿ Jun 1, 2019 15:42 |
|
suffix posted:should i be blindly copying code i don't understand from random repositories I mean you’re doing machine learning aren’t you?
|
# ¿ Jun 1, 2019 21:09 |
|
all the ethics courses in the world won't matter when each individual is just some alienated contributor to a greater machine, and who can rationalize away their own involvement in anything horrible which may result from their work
|
# ¿ Jun 6, 2019 18:05 |
|
It’s pretty hard to predict if anything you might work on could be weaponized Actually it’s pretty easy: it most likely will be Plenty of people working for google/amazon/whatever probably honestly didn’t think their work would be picked up by the MIC but it was Plenty of people doing research not even funded by one of the ARPAs might catch their interest out of the blue later and suddenly they’re pumping money into it Chemists and life science folks probably didn’t expect chemical and biological weapons to come out of their stuff Hell some of the early nuclear weapons folks probably didn’t realize how insane that would become
|
# ¿ Jun 7, 2019 06:41 |
|
animist posted:lol the other day while something I was working on wasn’t working I joked about having the GPUs just producing heat instead of anything useful
|
# ¿ Jun 10, 2019 02:52 |
|
I mean, the fact that the features picked up aren't necessarily the same ones a human might consciously choose is a pretty well known phenomenon in machine vision; the amusing bit being that people sometimes use this to get huffy about things when people sometimes unconsciously or consciously also do ridiculous things and we have much better sensors in some ways
|
# ¿ Jun 20, 2019 18:20 |
|
Winkle-Daddy posted:oh look, the handy work of Summly, the "AI News Summarizing Product" built by some kids that Marissa gave like a billion dollars to. Working at yahoo under marissa was the fuckin worst. "guys, guys, don't you see how valuable tumblr is???" if by “built by some kids” you mean farmed out building an application around some algorithms licensed from SRI
|
# ¿ Sep 21, 2019 02:00 |
|
you think open sores has issues thanks to a culture of tribal knowledge and gatekeeping, then you discover people’s trash ML repos that are horribly broken out of the box
|
# ¿ Sep 28, 2019 05:58 |
|
quote:Ray Phan, a senior computer-vision and deep-learning engineer at Hover, a 3D software startup based in San Francisco, told The Register that the lectures were confusing and contained mistakes. The online projects were too simplistic to be useful in the real world, and Raval was absent and unsupportive. Ah yes the senior AI engineer that needs to sign up for an intro to AI course
|
# ¿ Sep 29, 2019 05:36 |
|
animist posted:there's a bunch of "black box attacks" that work this way, only seeing inputs and outputs it also helps that transfer learning is common in vision so you can expect whatever you’re dealing with to be some fine tuned head on an imagenet base
|
# ¿ Sep 30, 2019 03:43 |
|
gently caress me for recognizing this
|
# ¿ Oct 1, 2019 18:40 |
|
animist posted:
don’t snipe my new convnet arch
|
# ¿ Nov 18, 2019 04:02 |
|
there are so many new DSLs masquerading as languages I just don’t keep track
|
# ¿ Nov 27, 2019 06:01 |
|
I mean it’s true
|
# ¿ Nov 28, 2019 18:15 |
|
power botton posted:if you want to do math in a dumb dead language use fortran which is exactly why there are python-FORTRAN bindings
|
# ¿ Dec 4, 2019 14:47 |
|
Some of our computational physicists still use FORTRAN And these are people like my age Living FORTRAN physics libraries
|
# ¿ Dec 4, 2019 14:48 |
|
I can’t tell if things are less horny now or different horny
|
# ¿ Dec 13, 2019 22:58 |
|
A NN classifier for fizz buzz but unironically
|
# ¿ Jan 3, 2020 03:15 |
|
statisticians amirite?
|
# ¿ Jan 31, 2020 21:10 |
|
yeah by the Stanford numbering scheme that's a graduate class e:fb quote:Numbering System
|
# ¿ Mar 26, 2020 23:16 |
|
Sagebrush posted:ah well then it's probably exactly what it says on the box. "grad students, go make a thing, if you need more information here's a list of papers to read, i'll be writing grants. if you make anything cool i'm putting my name as first author" don’t give away the sausage factory!!
|
# ¿ Mar 27, 2020 01:25 |
|
I remember when my weeder class in my engineering program was 1st semester, freshman year I still feel like I made the right decision but for the wrong reason
|
# ¿ Mar 27, 2020 06:07 |
|
DELETE CASCADE posted:what kind of hosed up PI doesn't let the main grad student on the project be first author? how else will he ever graduate you're.... you're being facetious right?
|
# ¿ Mar 27, 2020 19:07 |
|
academics is like the mob, credit kicks up to the top and never [graduating | getting tenure | moving on from being a postdoc] shits downhill
|
# ¿ Mar 27, 2020 19:09 |
|
are you suggesting that the tenured faculty at "elite" institutions are anything but the epitome of humility?
|
# ¿ Mar 27, 2020 21:40 |
|
that’s certainly a good formalization of things
|
# ¿ Apr 29, 2020 01:12 |
|
I remember when I did that last one by accident in college at my student job and got in trouble because we had to claw back and reprint a bunch of reports
|
# ¿ May 10, 2020 04:09 |
|
animist posted:also tbh if you're trying to understand data I'd reach for visualizations wayyy before ML because ML will just spit black boxes back at you. that's because I like to understand things tho actually I think you’ll find ml has a lot of trouble with black boxes
|
# ¿ Jun 5, 2020 21:17 |
|
obviously just shove it into tf-idf vectors
|
# ¿ Jun 6, 2020 13:50 |
|
m8 wtf are you doing
|
# ¿ Jun 8, 2020 16:23 |
|
what's your learning rate strategy?
|
# ¿ Jun 8, 2020 17:34 |
|
|
# ¿ May 11, 2024 22:31 |
|
rchon posted:this network has learned something but no animal ever has ultrafilter posted:Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead idk seems good
|
# ¿ Jun 9, 2020 05:13 |