Live data from Hacker News

Learning to See in the Dark (2018)

github.com

41–50 of 179 posts

Re: Learning to See in the Dark (2018)

#41
The problem with techniques like this is that they fundamentally amount to ‘making a plausible guess as to what the image would look like’, since essentially they can’t extract information that is simply not there. There is a Shannon entropy limit here.

Machine learning is really machine-enhanced educated-guesswork, which has its place but also has its limits.

Re: Learning to See in the Dark (2018)

#44

The problem with techniques like this is that they fundamentally amount to ‘making a plausible guess as to what the image would look like’, since essentially they can’t extract information that is simply not there. There is a Shannon entropy limit here. Machine learning is really machine-enhanced educated-guesswork, which has its place but also has its limits.

Counterpoint: The human brain converting a 2D image to a 3D model is educated-guesswork too :)

Re: Learning to See in the Dark (2018)

#45
post #3

Impressive of the American news channel, CNN, to convert images in minus one second.

CNN here is a "Convolutional Neural Network"

The title needs to be changed so brains recognize it as such. It either needs a preceding adjective or letter indicating what type of convolutional network it is.

The other option is spelling it out.

Most people will read CNN as the news channel. Even those familiar with neural networks.

Re: Learning to See in the Dark (2018)

#47

The problem with techniques like this is that they fundamentally amount to ‘making a plausible guess as to what the image would look like’, since essentially they can’t extract information that is simply not there. There is a Shannon entropy limit here. Machine learning is really machine-enhanced educated-guesswork, which has its place but also has its limits.

I'm also not a fan of how the only part actually readable in the (a) original, which is part of the title in the front book, becomes completely whitewashed in (c). Where the model actually had the most information, it completely removed it in the result...

Re: Learning to See in the Dark (2018)

#48
post #15
post #4

I was just wondering a couple days ago why the image from my phone is so grainy, while my eyes+brain can see everything clear in the dark (it wasn't completely dark, of course). This seems to replicate the post-processing we do in our brain (which is also a giant neural network). I wonder if the process is similar?

Very small numbers of photons (1) are required to trigger rhodopsin cycle. So primary receptor itself is very very VERY sensitive.

To be clear, the parent did not fail to include a citation. The parenthetical note is that rod cells are so sensitive that they react to being struck by a single photon.

Re: Learning to See in the Dark (2018)

#49
post #3

Impressive of the American news channel, CNN, to convert images in minus one second.

They should make a CLI tool, it stores your processed images one sec before it is invoked! See also: https://en.m.wikipedia.org/wiki/Thiotimoline The major peculiarity of the chemical is its "endochronicity": it starts dissolving before it makes contact with water.

See also Ted Chiang's What’s Expected of Us, where there's a device with a button and a LED, and the LED always lights up one second before you press the button.

Re: Learning to See in the Dark (2018)

#50
post #4

I was just wondering a couple days ago why the image from my phone is so grainy, while my eyes+brain can see everything clear in the dark (it wasn't completely dark, of course). This seems to replicate the post-processing we do in our brain (which is also a giant neural network). I wonder if the process is similar?

Your brain doesn’t make a 2 dimensional image based entirely on photons entering your eye. You generate a complex physical model of your surroundings based only partially on visual input and rely substantially on memory.

Kind of like this well trained CNN is no longer relying entirely on the raw pixel values, but is statistically inferring a brighter image from the baseline.
Post reply on HN