Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Gradient Descent Easy
Search
soulchild
July 23, 2014
Science
0
76
Gradient Descent Easy
Easy/brief version of Gradient Descent from Artificial Intelligence Lecture
soulchild
July 23, 2014
Tweet
Share
More Decks by soulchild
See All by soulchild
Similarities between macOS and iOS development
soulchild
0
110
N Tier Architecture for MMORPG
soulchild
0
86
Other Decks in Science
See All in Science
ultraArmをモニター提供してもらった話
miura55
0
120
Xpenologyなるアングラプロジェクト周りについて語るやつ
sushi514
0
670
スポーツメトリクス設計に対比較法を使いまくる / Sports metrics design using pairwise comparison method (spoana#14)
konakalab
1
730
はじめての「相関と因果とエビデンス」入門:“動機づけられた推論” に抗うために
takehikoihayashi
13
4.6k
統計的因果探索の方法
sshimizu2006
0
880
Yasuke
drawsbygba
0
610
マルチモーダルモデルと自動運転 車載モデルのコスト・スループット・レイテンシ / LLM in Production Meetup #2 20231023
yuyamaguchi
1
1.1k
東大・松尾研主催 LLM Summer 2023 コンペ解法 (11位 – 20位枠での優秀賞)
hayataka88
0
180
AI(人工知能)の過去・現在・未来 —AIは人間を超えるのか—
tagtag
0
230
The Universe, How it works
arafkarsh
0
110
Running llama.cpp on the CPU
ianozsvald
0
220
B-Cubed: Leveraging analysis-ready biodiversity datasets and cloud computing for timely and actionable biodiversity monitoring
peterdesmet
0
170
Featured
See All Featured
[Rails World 2023 - Day 1 Closing Keynote] - The Magic of Rails
eileencodes
8
1.3k
No one is an island. Learnings from fostering a developers community.
thoeni
16
2.1k
GraphQLとの向き合い方2022年版
quramy
33
12k
RailsConf 2023
tenderlove
8
560
Designing on Purpose - Digital PM Summit 2013
jponch
111
6.5k
Building Better People: How to give real-time feedback that sticks.
wjessup
356
18k
Designing Experiences People Love
moore
136
23k
Unsuck your backbone
ammeep
663
57k
Typedesign – Prime Four
hannesfritz
36
2.1k
Web development in the modern age
philhawksworth
203
10k
Learning to Love Humans: Emotional Interface Design
aarron
267
39k
Faster Mobile Websites
deanohume
300
30k
Transcript
Artificial Intelligence Gradient Descent soulchild
Gradient Descent Let computer find the minimum point in a
given graph or equation
Gradient Descent y 0 10 20 30 40 x -6
-4 -2 0 2 4 6 y = x2
Gradient Descent y 0 10 20 30 40 x -6
-4 -2 0 2 4 6 y = x2 Minimum point
Gradient Descent y 0 10 20 30 40 x -6
-4 -2 0 2 4 6 y = x2 Minimum point How does a computer know that this is a minimum point?
Gradient Descent Ans : By brute-forcing the derivative until a
value equal or near to 0 is found y = x2 dy dx = 2x Then guess x by starting from, eg: -6 to 6
Gradient Descent y 0 10 20 30 40 x -6
-4 -2 0 2 4 6 y = x2
Gradient Descent y 0 10 20 30 40 x -6
-4 -2 0 2 4 6 y = x2 dy dx = -12
Gradient Descent y 0 10 20 30 40 x -6
-4 -2 0 2 4 6 y = x2 dy dx = -12 dy dx = —8
Gradient Descent y 0 10 20 30 40 x -6
-4 -2 0 2 4 6 y = x2 dy dx = -12 dy dx = —8 dy dx = -4
Gradient Descent y 0 10 20 30 40 x -6
-4 -2 0 2 4 6 y = x2 dy dx = -12 dy dx = —8 dy dx = -4 dy dx = 0
Gradient Descent y 0 10 20 30 40 x -6
-4 -2 0 2 4 6 y = x2 dy dx = -12 dy dx = —8 dy dx = -4 dy dx = 0 Minimum point found, stop
Gradient Descent A better way to brute force
Gradient Descent y 0 10 20 30 40 x -6
-4 -2 0 2 4 6 y = x2 In short, it works like this
Gradient Descent y 0 10 20 30 40 x -6
-4 -2 0 2 4 6 y = x2 In short, it works like this
Gradient Descent y 0 10 20 30 40 x -6
-4 -2 0 2 4 6 y = x2 In short, it works like this
Gradient Descent y 0 10 20 30 40 x -6
-4 -2 0 2 4 6 y = x2 In short, it works like this
Gradient Descent How to set a good learning rate (α)
?
Gradient Descent When to stop searching ?
Gradient Descent When to stop searching ? Set a maximum
number of iteration
Gradient Descent When to stop searching ? Set a maximum
number of iteration dy dx < n When n can be 0.1, 0.01,.. etc
Gradient Descent Weakness of Gradient Descent 0 3 6 9
12 -8 -6 -4 -2 0 2 4
Gradient Descent Weakness of Gradient Descent 0 3 6 9
12 -8 -6 -4 -2 0 2 4 let say start from here
Gradient Descent Weakness of Gradient Descent 0 3 6 9
12 -8 -6 -4 -2 0 2 4 let say start from here dy dx = 0
Gradient Descent Weakness of Gradient Descent 0 3 6 9
12 -8 -6 -4 -2 0 2 4 let say start from here dy dx = 0 then computer stop finding
Gradient Descent Weakness of Gradient Descent 0 3 6 9
12 -8 -6 -4 -2 0 2 4 let say start from here dy dx = 0 then computer stop finding What about this?! smaller than previous point wor
Gradient Descent Weakness of Gradient Descent Gradient Descent may stuck
in a local minima thus can’t find the global minima
Gradient Descent Q&A