Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Speaker Deck
PRO
Sign in
Sign up
for free
ML Session n°4
Adrien Couque
April 05, 2017
Technology
0
12
ML Session n°4
Adrien Couque
April 05, 2017
Tweet
Share
More Decks by Adrien Couque
See All by Adrien Couque
acq
0
8
acq
0
32
acq
0
9
acq
0
10
acq
0
8
acq
0
11
acq
1
9
acq
0
5
acq
0
140
Other Decks in Technology
See All in Technology
i35_267
0
270
sadayoshitada0919
1
480
keiya01
20
7.9k
harshbothra
1
160
becolomochi
0
160
torisoup
0
310
stakaya
14
8.4k
tsugimot
0
140
k1low
1
500
indigo13love
2
520
sugimomoto
1
380
bells17
0
610
Featured
See All Featured
keithpitt
401
20k
bryan
100
11k
skipperchong
7
670
bkeepers
408
57k
myddelton
109
11k
mza
80
4.1k
chriscoyier
684
180k
keathley
18
650
scottboms
251
11k
malarkey
119
16k
mojombo
358
62k
jacobian
255
20k
Transcript
ML: regularization and neural networks March 2017
Recap
Recap: linear regression
Recap: polynomial regression
Recap: gradient descent
Regularization
Normalization Goal : have all features be equivalent (in size)
Rescaling : Standardization :
Regularization : goal
Regularization Before : Now : Forces values to stay low
Neural networks
Demo: Tensorflow Playground
Biological neuron
Artificial neuron: Perceptron
Teaching a neuron
Artifical neuron : activation function
Most common activation functions Identity Sigmoid Tanh ReLU
Teaching a neuron
Demo: simple network, but the hard way (92%)
Demo: simple network, but the hard way (92%)
Artificial neuron -> Artificial neural network
Artificial neuron -> Artificial neural network
Teaching a neural net
Demo: Tensorflow Playground
Demo: layered network, still the hard way (97%)
Questions? March 2017