SELU Appendix @SELUAppendix

The appendix from beyond the moon. arxiv.org/abs/1706.02515 Arxiv Joined June 2017

Tweets

62
Followers

614
Following

11
Likes

42

SELU Appendix @SELUAppendix

7 years ago

It is always worth reading the appendix

0 1 14 0 1

Tenant of 22 Willoughby @TenantOf22

7 years ago

@quantumVerd @alfcnz The paper got fixed faster than my boiler I tell you

2 4 54 0 0

RUDDER: Return Decomposition for Delayed Rewards: "Exponentially faster than TD, MC, and MC Tree Search. Outperforms rainbow, A3C, DDQN, Distributional DQN, Dueling DDQN, Noisy DQN, and Prioritized DDQN". The 50-page Appendix reminds me of @SELUAppendix. arxiv.org/abs/1806.07857

3 70 227 0 15

Kelly Banman @k_banman

7 years ago

The “Eventual Consistency” You order a beer, and when you assume the bartender forgot you order another. Suddenly you feel intoxicated and then two beers arrive. #awsdrinks

0 10 60 0 0

vicki @vboykis

7 years ago

Please do yourself a favor and check out the #awsdrinks hashtag.

0 29 42 0 0

Yann LeCun @ylecun

7 years ago

Accurate speech recognition is no longer a red hair ring. It can wreck a nice beach with no error.

8 117 564 0 3

Dylan Bourgeois @dtsbourg

8 years ago

First edition of reading notes! Self-Normalizing Neural Networks, by G. Klambauer, T. Unterthiner, A. Mayr, and S. Hochreiter dtsbourg.github.io/thoughts/posts…

3 13 27 0 2

Keunwoo Choi @keunwoochoi

8 years ago

Yet another activation function!

Christian Szegedy @ChrSzegedy

8 years ago

Yet another activation function!

1 29 105 0 4

1 1 2 0 0

Fitzy McFitzyfitz @fitzyfitzyfitzy

8 years ago

@SELUAppendix you should write a thriller called THE ACTIVATION FUNCTION

1 3 6 0 0

SELU Appendix @SELUAppendix

8 years ago

Does your deep learning framework write its own license? Looks like @pjreddie's does #darknet #yolo

Fitzy McFitzyfitz @fitzyfitzyfitzy

8 years ago

Does your deep learning framework write its own license? Looks like @pjreddie's does #darknet #yolo

1 4 5 0 0

Download Image

0 9 11 0 0

Ben Poole @poolio

8 years ago

@octonion they did architecture search over many variants. see the first footnote in the paper:

1 8 25 0 0

Download Image

hardmaru @hardmaru

8 years ago

f(x) = x．sigmoid(x) They call it the Swish, a self-gated activation function. Seems to outperform Relu. arxiv.org/abs/1710.05941

18 138 439 0 0

Fitzy McFitzyfitz @fitzyfitzyfitzy

8 years ago

My deepmoon deep learning framework is coming along github.com/paulfitz/deepm…

1 1 3 0 0

Download Image

Neal Wu @WuNeal

8 years ago

280 characters of @SELUAppendix: 4x 2 + 2xy + 2.7795x − 2y 2 − 0.9269y − 0.000277982 (2x + y) 2 + 5.39467x . This is equivalent to 0 < 4x 2 + 2xy + 2.7795x − 2y 2 − 0.9269y − 0.000277982 (2x + y) 2 + 5.39467x − (280) −8x 3 − 8x 2 y − 10.9554x 2 + 2xy2 + 1.76901xy -