Visualizing Neural Network Internals

preview_player
Показать описание
Visualizing some of the internals of a neural network during training and inference.

Рекомендации по теме
Комментарии
Автор

I bought your NNFS book a couple of years ago. By far the best book on Neural networking, probably in the entire ML domain. Absolute gold.

nktthegreat
Автор

I’d recommend more visualization tutorials like this… fascinating

digital_down
Автор

Very cool - and surprisingly useful when pitching AI to investors who love to see the flashing patterns and force directed graphs (sadly I’m being serious)

asksearchknock
Автор

This is the first visualization I have seen. You are before you time sir, this will be the norm for models very soon. Excellent work!

stackingwins
Автор

I appreciate all your work. These visualizations confirm this that deep learning is a black box

pebre
Автор

Friggin awesome. Nostalgia never misses

busterfranken
Автор

'which better be pytorch'
gotta love it

ferdyg
Автор

I watched your video of GameGAN (GTA-5 running only on a neural network). This is mind blowing! I try to recreate it with Tux Racer game

kotcraftchannelukraine
Автор

Very cool, I'd love to use that for a group project I'm working on right now, is there an ETA for the full code?

BarkofthePhoenix
Автор

Harrison, please inform us if you plan to continue the tutorial series for NNFS. It's a brilliant piece of work. Please do tell us either way. You been super quiet on this for long :P Mighty thanks for your content!

prateekyadav
Автор

Very cool tool for introductory ML courses at unis, etc.

hola-kxgn
Автор

Very good video keep it up .... love from india ❤

dineshasavsani
Автор

Very useful! Thank you for sharing. One question, since we could know which neuron is more important for a specific class. Can we find out what filters (or kernels) in the three layers makes it important to the specific class?

yuanfongsu
Автор

likin' the great old sentdex style ! :p

lucasb.
Автор

Is it possible to visualize the network architecture directly from the code I build?

sguha
Автор

In PyTorch this can be automated using register_forward_hooks. It even works on custom layers I make. Only issue is how do I visualize convolutional layer weights when there are 128 64x2x2 kernels, so I think I show first 9 kernels in a grid and take first 3 channels

carnap
Автор

sentdex asking everybody what is going on
but not actually caring about the answer

piotrzaborszczyk
Автор

Hey dude, I gotta tell you something, I owe you my career.🙏🏻🙏🏻🙏🏻 keep it up plz🙏🏻👏🏼

lolwtf
Автор

hei just a simple question to you all, Is it possible to make an web app like streamlit out of it ?

bhaskarmondal
Автор

the concept of “hidden layers” might be the biggest crime. they are NOT hidden, and they can be introspected.

thesmallestatom