fast.ai Lesson 1: Deep Learning 2018

10 free hours

run on AWS

click this one

click on new machine

pick a region

choose linux ubuntu 16

  

250GB is preferred

ctrl shift v

to paste your password

zoom in

takes an hour or so.

when it finishs running, you need to reboot your paperspace computer

double click the last line to copy the url

then you can  go and past it

change local host into the true IP address

or hold on shift and press enter

no, don‘t follow this suggestion.

ML was invented by this man.

a function that can sovle all problems

gpu is about 10 times faster than cpu

gpu is also cheaper

real time translation

use DL to help combine human experties and what computers are good at

setosa.io/ev/image-kernels/

draw a picture of what neural networks has learnt

just 3 layers is enouch to get some pretty rich behaviors

hit tab, and get a list of methods

hit shift tab tells you the arguments to the method

shift tab twice or three times and get a documentation window

look at the source code

press H

stop the machine and see the connection is closed

原文地址：https://www.cnblogs.com/ecoflex/p/8995592.html

时间： 2024-10-02 12:06:59

fast.ai Lesson 1: Deep Learning 2018的相关文章

fast.ai Lesson 2: Deep Learning 2018

data augumentation 原文地址:https://www.cnblogs.com/ecoflex/p/8995745.html

CoRR 2018 | Horovod: Fast and Easy Distributed Deep Learning in Tensorflow

将深度学习模型的训练从单GPU扩展到多GPU主要面临以下问题:(1)训练框架必须支持GPU间的通信,(2)用户必须更改大量代码以使用多GPU进行训练.为了克服这些问题,本文提出了Horovod,它通过Ring Allreduce实现高效的GPU间通信,而且仅仅更改少量代码就可以实现多GPU训练. TensorFlow中提供了一些分布式训练的API,这些API适用于不同的环境.这就导致用户往往不知道如何更改代码以进行分布式训练,而且debug也很困难.再者,TensorFlow的分布式训练性能与理

CMU Deep Learning 2018 by Bhiksha Raj 学习记录(17)

NN is pretty bad at learning this pattern. green dots are referred to the first layer. blue -> second layer red -> third layer https://github.com/cmudeeplearning11785/machine_learning_gpu/blob/master/Dockerfile www.dockerhub.com https://github.com/w

CMU Deep Learning 2018 by Bhiksha Raj 学习记录(1)

Recitation 2 numpy operations array index x = np.arange(10) ** 2 # x:[ 0 1 4 9 16 25 36 49 64 81] print(x[::-1]) # all reversed print(x[8:1:-1]) # reversed slice print(x[::2]) # every other print(x[:]) # no-op (but useful syntax when dealing with n-d

CMU Deep Learning 2018 by Bhiksha Raj 学习记录(7)

https://zhuanlan.zhihu.com/p/22810533 L2 -> Regression Problems KL -> Classification Problems http://deeplearning.cs.cmu.edu/slides/lec8.stochastic_gradient.pdf 原文地址:https://www.cnblogs.com/ecoflex/p/8886201.html

CMU Deep Learning 2018 by Bhiksha Raj 学习记录(10)

http://deeplearning.cs.cmu.edu/slides/lec11.recurrent.pdf I think the subscripts in this lecture is quite confusing, and even incorrect sometimes. Jacobian Matrix 原文地址:https://www.cnblogs.com/ecoflex/p/8904117.html

CMU Deep Learning 2018 by Bhiksha Raj 学习记录(18)

https://www.youtube.com/watch?v=oiNFCbD_4Tk <eos> end of section a common trick image to seq problem mistake might happen. look at the fig(1,0) could lose info by only taking the red box into consideration. the new framework still has a problem soft

CMU Deep Learning 2018 by Bhiksha Raj 学习记录(20) Lecture 20: Hopfield Networks 1

symmetric version: called Hopfield Net      if y <-- minus y, energy won't change! the second last line's first "-" is not the minus sign https://zhuanlan.zhihu.com/p/21539285 原文地址:https://www.cnblogs.com/ecoflex/p/8971351.html

（转）WHY DEEP LEARNING IS SUDDENLY CHANGING YOUR LIFE

Main Menu Fortune.com E-mail Tweet Facebook Linkedin Share icons By Roger Parloff Illustration by Justin Metz SEPTEMBER 28, 2016, 5:00 PM EDT WHY DEEP LEARNING IS SUDDENLY CHANGING YOUR LIFE Decades-old discoveries are now electrifying the computing