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
NNにふれてみて
Search
bakunyo
February 14, 2017
Technology
0
940
NNにふれてみて
開発合宿@鎌倉で発表した際に使用した資料です
bakunyo
February 14, 2017
Tweet
Share
More Decks by bakunyo
See All by bakunyo
合宿成果@日本色
bakunyo
0
120
GitHub Actions CI/CD Update
bakunyo
0
150
AMP 完全に理解した
bakunyo
0
270
RoRoRoRoR
bakunyo
0
820
Other Decks in Technology
See All in Technology
ユーザーのプロフィールデータを活用した推薦精度向上の取り組み
yudai00
0
370
評価の納得感を2段階高める「構造化フィードバック」
aloerina
1
170
Autonomous Database サービス・アップデート (FY25)
oracle4engineer
PRO
2
770
技術職じゃない私がVibe Codingで感じた、AGIが身近になる未来
blueb
0
130
産業機械をElixirで制御する
kikuyuta
0
170
キャディでのApache Iceberg, Trino採用事例 -Apache Iceberg and Trino Usecase in CADDi--
caddi_eng
0
130
Kotlinで学ぶ 代数的データ型
ysknsid25
5
1.1k
堅牢な認証基盤の実現 TypeScriptで代数的データ型を活用する
kakehashi
PRO
2
220
What's new in OpenShift 4.19
redhatlivestreaming
1
230
In Praise of "Normal" Engineers (LDX3)
charity
2
910
Tenstorrent 開発者プログラム
tenstorrent_japan
0
310
Securing your Lambda 101
chillzprezi
0
280
Featured
See All Featured
Building Better People: How to give real-time feedback that sticks.
wjessup
367
19k
4 Signs Your Business is Dying
shpigford
184
22k
Let's Do A Bunch of Simple Stuff to Make Websites Faster
chriscoyier
507
140k
Dealing with People You Can't Stand - Big Design 2015
cassininazir
367
26k
10 Git Anti Patterns You Should be Aware of
lemiorhan
PRO
657
60k
Imperfection Machines: The Place of Print at Facebook
scottboms
267
13k
Optimising Largest Contentful Paint
csswizardry
37
3.3k
Done Done
chrislema
184
16k
ReactJS: Keep Simple. Everything can be a component!
pedronauck
667
120k
[RailsConf 2023] Rails as a piece of cake
palkan
55
5.6k
Build The Right Thing And Hit Your Dates
maggiecrowley
36
2.7k
The Language of Interfaces
destraynor
158
25k
Transcript
NNʹ;ΕͯΈͯ 2017/02/12 @ ݹయ෦߹॓ Izuta Hiroyuki
Γ͔ͨͬͨ͜ͱ • Ξχϝը૾ྨث • Deep LearningͰϥϒϥΠϒʂΩϟϥΛࣝผ͢Δ - http://christina.hatenablog.com/entry/2015/01/23/212541 • TensorFlowͰΞχϝΏΔΏΓͷ੍࡞ձࣾΛࣝผ͢Δ
- http://kivantium.hateblo.jp/entry/2015/11/18/233834
ͬͨ͜ͱ • DeepLearningຊͷ3~4ষ • χϡʔϥϧωοτϫʔΫͰ࿈ཱํఔࣜΛղ͘ • TensorFlowΛͪΐͬͱ৮ͬͯΈΔ • CIFAR-10ͷը૾ྨ
1࣍ؔ y = 0.3x + 0.1 ॳظ: W(eight) = -1
~ 1ͷϥϯμϜͳ b(ias) = 1 ଛࣦؔ: ̎ޡࠩ ޯ߱Լ๏
1࣍ؔ y = 0.3x + 0.1 ʹۙͮ͘ ʹۙͮ͘
࿈ཱํఔࣜ • ͔ͨ͠܅1݁ࠗࣜʹग़੮ͨ͜͠ͱ͕ແ͘ɺ͝ॕّͷ ૬͕Θ͔Βͳ͍ • طࠗͷ༑ਓ3ਓʹฉ͍ͯɺͦΕͧΕ͝ॕّͷ߹ܭ͕͍͘Β ͔֬ೝͨ͠ • ͝ॕّͷύλʔϯྨɾ༑ਓɾਓͷ3ύλʔϯ ͰɺͦΕͧΕ͍͘Βʁ
͝ॕّ ྨ ਓ ༑ਓ ਓ ਓ ਓ ߹ܭ ສԁ
͝ॕّ
CIFAR-10Ͱը૾ྨ
CIFAR-10 • 10छྨͷը૾ྨʹΘΕΔσʔληοτ
ਪϞσϧ: ଟͷCNN • CNNʢConvolutional Neural Networkʣ ΈࠐΈχϡʔϥϧωοτϫʔΫ • ΈࠐΈɾϓʔϦϯάɾશ݁߹͔ΒͳΔ •
ΈࠐΈ: ಛඍྔͷநग़ • ϓʔϦϯά: Ԡෆมੑͷ֫ಘ
TensorBoard • TensorFlow ඪ४ఴͷՄࢹԽπʔϧ • tensorboard --logdir={logσΟϨΫτϦ}
Demo
CNNͷ్த Accurancy: 0.2 -> 0.5 ͘Β͍·Ͱ্ঢ͢Δͷݟͯதஅ
ϋϚͬͨͱ͜Ζ
TensorFlowͰHello, World • ͍͖ͳΓίέΔɾɾɾ • AttributeError: module 'tensorflow' has no
attribute ‘constant' • import tensorflow as tf • ϑΝΠϧ໊Λ tensorflow.py ʹͯͨ͠ͷ͕ݪҼ • tfͰͳ͘Pythonͷ • http://stackoverflow.com/questions/37383812/tensorflow- module-object-has-no-attribute-placeholder
Depreciated Methods • มͷॳظԽ (چ) tf.initialize_all_variables() (৽) tf.global_variables_initializer() • TensorBoardͷೖྗ
(چ) tf.train.SummaryWriter() (৽) tf.summary.FileWriter() • άϥϑఆٛ (چ) sess.graph_def (৽) sess.graph
(༨ஊ) DocumentʹDepreciated͕͋ͬͨͷͰ PRૹΖ͏ͱͨ͠Βɾɾɾ
(༨ஊ) DocumentʹDepreciated͕͋ͬͨͷͰ PRૹΖ͏ͱͨ͠Βɾɾɾ
(༨ஊ) DocumentʹDepreciated͕͋ͬͨͷͰ PRૹΖ͏ͱͨ͠Βɾɾɾ
ײ • Γ͔ͨͬͨ͜ͱ·ͰḷΓண͚ͳ͔ͬͨ • PythonTensorFlowෆ׳Εײ • ଟ࣍ݩྻɺreshapeͱ͔Ͱਖ਼֬ʹΠϝʔδ ม͍ͨ͠
ࠓޙ • ·ͣຊΛ࠷ޙ·ͰಡΉ • ྲྀΕ͕ΠϝʔδͰ͖ͯ͘Δͱָ͍͠ • ը૾ྨΓ͍ͨɺ͚Ͳը૾ͷऩूͱϥϕϦ ϯά͕໘ͦ͏
͓ΘΓ