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
宝くじ仮説の話を少し / LT Lottery Ticket Hypothesis
Search
びん@SKA’s Web
November 14, 2021
Programming
480
0
Share
宝くじ仮説の話を少し / LT Lottery Ticket Hypothesis
Python機械学習勉強会 in 新潟のLTの資料です。深層学習の「宝くじ仮説」について少しだけ話します。
びん@SKA’s Web
November 14, 2021
More Decks by びん@SKA’s Web
See All by びん@SKA’s Web
n進数とは / What is Base-n Numeral System
skasweb
0
12
.NETでOBS Studio操作してみたけど…… / Operating OBS Studio by .NET
skasweb
0
200
学習しやすいネットワークガチャ? 深層学習の「宝くじ仮説」 【VRアカデミア クリスマスリレー】 / VR Lottery Ticket Hypothesis
skasweb
1
630
自然言語処理の基本 / Basic of Natural Language Processing
skasweb
0
370
AIとクラスター分析について自作アイテムを用いて解説してみた 【VRアカデミア3周年カウントダウン】 / Explaining AI and Cluster Analysis Using My Item
skasweb
0
380
Other Decks in Programming
See All in Programming
ふにゃっとしない名前の付け方 〜哲学で茹で上げる、コシのあるソフトウェア設計〜
shimomura
0
120
実践ハーネスエンジニアリング:ステアリングループを実例から読み解く / Practical Harness Engineering: Understanding Steering Loops Through Real-World Examples
nrslib
5
5.3k
Terraform言語の静的解析 / static analysis of Terraform language
wata727
1
140
Road to RubyKaigi: Play Hard(ware)
makicamel
1
570
SkillsをS3 Filesに置く時のあれこれ
watany
3
1.5k
Augmenting AI with the Power of Jakarta EE
ivargrimstad
0
410
How We Practice Exploratory Testing in Iterative Development( #scrumniigata ) / 反復開発の中で、探索的テストをどう実施しているか
teyamagu
PRO
3
810
サークル参加から学ぶ、小さな事業の回し方
yuzneri
0
180
Structured Concurrency, Scoped Values and Joiners in the JDK 25 26 27
josepaumard
1
150
Are We Really Coding 10× Faster with AI?
kohzas
0
170
🦞OpenClaw works with AWS
licux
1
350
運転動画を検索可能にする〜Cosmos-Embed1とDatabricks Vector Searchで〜/cosmos-embed1-databricks-vector-search
studio_graph
3
890
Featured
See All Featured
Agile that works and the tools we love
rasmusluckow
331
21k
Build your cross-platform service in a week with App Engine
jlugia
234
18k
Stewardship and Sustainability of Urban and Community Forests
pwiseman
0
200
Future Trends and Review - Lecture 12 - Web Technologies (1019888BNR)
signer
PRO
0
3.5k
The Director’s Chair: Orchestrating AI for Truly Effective Learning
tmiket
1
160
What’s in a name? Adding method to the madness
productmarketing
PRO
24
4k
SEO in 2025: How to Prepare for the Future of Search
ipullrank
3
3.4k
Gemini Prompt Engineering: Practical Techniques for Tangible AI Outcomes
mfonobong
2
390
Building a A Zero-Code AI SEO Workflow
portentint
PRO
0
510
Designing for Timeless Needs
cassininazir
1
220
Faster Mobile Websites
deanohume
310
31k
Redefining SEO in the New Era of Traffic Generation
szymonslowik
1
300
Transcript
͍ۙͷͰʮๅ͘͡ԾઆʯͷΛগ͠ 1ZUIPOػցֶशษڧձ JO৽ׁ ৽ׁݝ࢈75VCFS 4,"`T8FC CJO 1
ࢿྉͷϦϯΫͱࢀߟจݙ 2 IUUQTTQFBLFSEFDLDPNTLBTXFCMUMPUUFSZUJDLFUIZQPUIFTJT ࢀߟจݙ ɾIUUQTEFFQTRVBSFKQMPUUFSZUJDLFU ɾIUUQTMBCPSPBJBDUJWJUZDPMVNOFOHJOFFSๅ͘͡Ծઆ ݩจ ɾJonathan Frankle, Michael
Carbin, “The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks”, cs.LG : 1803.03635, ICLR 2019 ɾ5JBOMPOH$IFO +POBUIBO'SBOLMF 4IJZV $IBOH 4JKJB -JV :BOH;IBOH ;IBOHZBOH 8BOH .JDIBFM$BSCJO l5IF-PUUFSZ5JDLFU)ZQPUIFTJTGPS1SFUSBJOFE#&35/FUXPSLTz
5XJUUFS࿈ܞ 3 #pyml_niigata ͰͭͿ͘ͱʜʜ
࢈ۀ 4 ɾ ڊେ ͳϞσϧʢ#&35ɺ5ɺFUDʣ͕ͳͥਫ਼ྑ͍ͷ͔ʁ ɾύϥϝʔλʔ͕ଟ͍΄ͲΨνϟ͕͋ͨΔʂʁ ɾԠ༻͢Δͱ&EHF"*Ͱߴਫ਼ͳϞσϧΛಈ͔ͤΔʁ
ڊେͳϞσϧ 5
ڊେͳϞσϧ 6 ɾ͜ΜͳʹେྔͷύϥϝʔλʔΛ͏·͘ߋ৽͢Δ͜ͱͰ͖Δͷʁ ɾաֶश͠ͳ͍ͷʁ ๅ͘͡Ծઆ
ๅ͘͡Ծઆ 7 ɾχϡʔϥϧωοτϫʔΫॳظͷॏΈ͕ཚ ɾ࠷ॳ͔Β͍͍ײ͡ʹֶशՄೳͳ෦ωοτϫʔΫ͕͋Δ ɾ͋ͨΓ͘͡ωοτϫʔΫΛத৺ʹֶश͞ΕΔ ɾͣΕ͘͡ωοτϫʔΫࢬמΓͯ͠ਫ਼΄΅མͪͳ͍ ɾ#&35Ͱׂʙׂͷলύϥϝʔλʔ͕Մೳʢʁʣ ɾߴʹֶशɺਪ͕Ͱ͖Δ ɾ&EHF"*Ͱಈ͔ͤΔʁ
Ϟσϧͷݮ 8 ʢ·ͩௐ͍ͯ·ͤΜ PS[ ʣ ʢʣʹಈըߘ͠·͢ ʢࠓ͙͢Γ͍ͨΜ͚ͩͲͬͯਓࢀߟจݙݩจΛಡΜͰΈ͍ͯͩ͘͞ʣ
·ͱΊ 9 ɾχϡʔϥϧωοτϫʔΫͷॳظͷॏΈཚ ɾ͋ͨΓωοτϫʔΫͱͣΕωοτϫʔΫ͕͋Δ ɾ࿈ΨνϟͰͨΒͳ͍ʁ ԯ࿈͘Β͍Ε͚ͬ͜͏ͨΔΑʂ ɾ͍ͬͯ͏ͷ͕࠷ۙͷ ڊେϞσϧʢ#&35ɺ5ɺFUDʣ ɾͣΕωοτϫʔΫΛࢬמΓ͢Εগͳ͍ετϨʔδͰߴʹਪͰ͖Δ ɾ͡Ό͋Ͳ͏Δͷʁ
ˠ ·ͨࠓͳʂ
ࢿྉͷϦϯΫͱࢀߟจݙ 10 IUUQTTQFBLFSEFDLDPNTLBTXFC΄͛΄͛ ࢀߟจݙ ɾIUUQTEFFQTRVBSFKQMPUUFSZUJDLFU ɾIUUQTMBCPSPBJBDUJWJUZDPMVNOFOHJOFFSๅ͘͡Ծઆ ݩจ ɾJonathan Frankle, Michael
Carbin, “The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks”, cs.LG : 1803.03635, ICLR 2019 ɾ5JBOMPOH$IFO +POBUIBO'SBOLMF 4IJZV $IBOH 4JKJB -JV :BOH;IBOH ;IBOHZBOH 8BOH .JDIBFM$BSCJO l5IF-PUUFSZ5JDLFU)ZQPUIFTJTGPS1SFUSBJOFE#&35/FUXPSLTz