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
2019_G検定対策_数学講座01_ディープラーニングでの基礎計算問題/20190125_JD...
Search
ITO Akihiro
January 25, 2019
Technology
0
7
2019_G検定対策_数学講座01_ディープラーニングでの基礎計算問題/20190125_JDLA_G_Math_1
G検定対策社内数学講座
--
ディープラーニングでの基礎計算問題
パーセプトロンとCNNと活性化関数
ITO Akihiro
January 25, 2019
Tweet
Share
More Decks by ITO Akihiro
See All by ITO Akihiro
エンジニア目線でのテスラ
akit37
0
48
「重鎮問題」について(軽めに)
akit37
0
61
Software + Hardware = Fun++
akit37
0
28
基本的に "リモートしかない" ワーク/20231128_KBS_LT
akit37
1
22
3つの先端技術が コミュニティ軸で融合した話。/20230615_CMCMeetup
akit37
0
19
Bootleg_越境してみたときのアウェイ感。/20230328_CMCMeetup
akit37
0
25
始まりは2017年のG検定。/20221026_AITable
akit37
0
16
kintone知能化計画/20220902_kintone_and_JPStripes
akit37
0
26
外観検査用画像前処理の_コツをコード解説付きで。/20220810_CDLE_LT
akit37
0
17
Other Decks in Technology
See All in Technology
AIのグローバルトレンド 2025 / ai global trend 2025
kyonmm
PRO
1
140
生成AI時代におけるAI・機械学習技術を用いたプロダクト開発の深化と進化 #BetAIDay
layerx
PRO
1
1.2k
Kiroでインフラ要件定義~テスト を実施してみた
nagisa53
3
370
UDDのススメ - 拡張版 -
maguroalternative
1
580
はじめての転職講座/The Guide of First Career Change
kwappa
5
4.1k
Eval-Centric AI: Agent 開発におけるベストプラクティスの探求
asei
0
130
20250807_Kiroと私の反省会
riz3f7
0
230
LLMで構造化出力の成功率をグンと上げる方法
keisuketakiguchi
0
880
【新卒研修資料】数理最適化 / Mathematical Optimization
brainpadpr
27
13k
OPENLOGI Company Profile for engineer
hr01
1
38k
人に寄り添うAIエージェントとアーキテクチャ #BetAIDay
layerx
PRO
10
2.3k
마라톤 끝의 단거리 스퍼트: 2025년의 AI
inureyes
PRO
1
750
Featured
See All Featured
Build The Right Thing And Hit Your Dates
maggiecrowley
37
2.8k
The World Runs on Bad Software
bkeepers
PRO
70
11k
Helping Users Find Their Own Way: Creating Modern Search Experiences
danielanewman
29
2.8k
Unsuck your backbone
ammeep
671
58k
Imperfection Machines: The Place of Print at Facebook
scottboms
267
13k
Building an army of robots
kneath
306
45k
The Art of Programming - Codeland 2020
erikaheidi
54
13k
No one is an island. Learnings from fostering a developers community.
thoeni
21
3.4k
Let's Do A Bunch of Simple Stuff to Make Websites Faster
chriscoyier
507
140k
The Art of Delivering Value - GDevCon NA Keynote
reverentgeek
15
1.6k
10 Git Anti Patterns You Should be Aware of
lemiorhan
PRO
656
60k
The Cult of Friendly URLs
andyhume
79
6.5k
Transcript
ディープラーニングでの 基礎計算 〜パーセプトロンとCNNと活性化関数〜 Jun. 2019 created by ITO Akihiro
パーセプトロンの概念 入力 重み バイアス 出力 入力:1 or 0 出力:1 or 0
画像認識の概念 1 2 3 4 5 6 7 8 9
10 11 12 31 32 12 21 22 23 3 1 2 ・ ・ ・ 入力層 中間層 出力層 1 0 1 2 3 4 5 6 7 8 9 10 11 12 0 1 0 0 重み w1-1 ... w1-36 重み w2-1 ... w2-6 それぞれのシグモイドニューロンは、 ある特徴に対して発火する
CNN (Convolutional Neural Network)の概念 畳み込み層 フィル タ3 特徴マップ 3 プーリング層
出力層 ユニット3 特徴を抽出 特徴をさらに濃縮
フィルタを順に当て 画像を単純化し、 特徴を抽出する。 1 2 3 4 5 6 7
8 9 10 11 12 13 14 15 16 17 18 19 20 21 ... ... 121 元画像 フィルタ1 フィルタ2 斜め方向に反応 縦方向に反応
例: 入力画像 (H,W)=(11,11) フィルタサイズ(Fh,Fw)=(3,3) パディング (P)=0 ストライド (S)=2 ⇒ 出力画像 (Oh,Ow)=(5,5) 1 0 0
1 0 1 3 1 2 1 2 2 3 2 3 2 2 2 3 2 1 2 0 1 1 0 0 0 0 0 0 2 1 2 0 2 2 3 2 2 1 2 2 2 1 0 1 0 1 0 公式
活性化関数 ステップ シグモイド tanh (ハイパボリックタンジェント) ReLU ReLU6 0 1 0
0 0 0 6 0 0 1 0 -1 1 0