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
論文紹介: Sample Reuse via Importance Sampling in I...
Search
Masahiro Nomura
April 09, 2020
Research
0
210
論文紹介: Sample Reuse via Importance Sampling in Information Geometric Optimization / sample_reuse_igo
Masahiro Nomura
April 09, 2020
Tweet
Share
More Decks by Masahiro Nomura
See All by Masahiro Nomura
ランダム欠損データに依存しない推薦システムのバイアス除去 / towards-resolving-propensity-contradiction-in-offline-recommender-learning
nmasahiro
0
190
転移学習によるハイパーパラメータ最適化の高速化 / warm_starting_cma
nmasahiro
0
1.9k
論文紹介: IRのためのパラメータチューニング / ir-tuning
nmasahiro
0
410
機械学習における ハイパーパラメータ最適化の理論と実践 / hpo_theory_practice
nmasahiro
31
32k
論文紹介 : Population Based Augmentation: Efficient Learning of Augmentation Policy Schedules
nmasahiro
1
620
広告とAI(とハイパーパラメータ最適化) / Ad with AI
nmasahiro
1
1.8k
Other Decks in Research
See All in Research
Weekly AI Agents News! 10月号 論文のアーカイブ
masatoto
1
250
MIRU2024_招待講演_RALF_in_CVPR2024
udonda
1
330
文献紹介:A Multidimensional Framework for Evaluating Lexical Semantic Change with Social Science Applications
a1da4
1
220
20241115都市交通決起集会 趣旨説明・熊本事例紹介
trafficbrain
0
230
新規のC言語処理系を実装することによる 組込みシステム研究にもたらす価値 についての考察
zacky1972
0
120
第79回 産総研人工知能セミナー 発表資料
agiats
2
160
日本語医療LLM評価ベンチマークの構築と性能分析
fta98
3
640
さんかくのテスト.pdf
sankaku0724
0
340
Leveraging LLMs for Unsupervised Dense Retriever Ranking (SIGIR 2024)
kampersanda
2
190
Physics of Language Models: Part 3.1, Knowledge Storage and Extraction
sosk
1
950
「並列化時代の乱数生成」
abap34
3
820
研究の進め方 ランダムネスとの付き合い方について
joisino
PRO
55
19k
Featured
See All Featured
The Language of Interfaces
destraynor
154
24k
Fashionably flexible responsive web design (full day workshop)
malarkey
405
65k
Thoughts on Productivity
jonyablonski
67
4.3k
The Success of Rails: Ensuring Growth for the Next 100 Years
eileencodes
44
6.8k
The Power of CSS Pseudo Elements
geoffreycrofte
73
5.3k
Building Flexible Design Systems
yeseniaperezcruz
327
38k
Bootstrapping a Software Product
garrettdimon
PRO
305
110k
ReactJS: Keep Simple. Everything can be a component!
pedronauck
665
120k
Fantastic passwords and where to find them - at NoRuKo
philnash
50
2.9k
Happy Clients
brianwarren
98
6.7k
Building an army of robots
kneath
302
43k
Become a Pro
speakerdeck
PRO
25
5k
Transcript
. Sample Reuse via Importance Sampling in Information Geometric Optimization
^ILW1NL Y 18 : I ) ) * ( WP LW WPK ^ 1
• . S RL ?L XL [P 8S UWY L
S RP N P 8 MUWS YPU 7LUSLYWP = YPSP YPU • . OP P OP OPW DU OLP 1 PSUYU P = OP U U =O W • . WCP[ ) (- 75 = (+ L]YL K • . w 87= ʼ • . 28: 1 5 h • . 1 5 z~ 2
• 87= • 87= • • 3
87= • i i i • . s p f
s • w ~n • 1 5 5 . ]ac nʼ s • 28: 71 . ]ac nʼ • w z~ 1 5 ? e A K YL 4
87= • 87= (. • M I M 5
87= • zrzf5EM ] F sm • y • z~
s n • z~ 5EB ] F 6
87= • zrzf5EM ] F sm • y • z~
s n • z~ 5EB ] F 7
87= • zrzf5EM ] F sm • y • z~
s n • z~ 5EB ] F 8
87= • zrzf5EM ] F sm • y • z~
s n • z~ 5EB ] F 9
87= • now f5EB ] F w ʼ • ~
• 87= ). • N I S • . • : KP[s • i rv w 10
87= • • d ]ac r • • 5EB ]
F zfc 11
87= • • d ]ac r • • 5EB ]
F zfc • d ~ sm 12
87= • 87= d n~n • PYLW YPU x ~~z
~ f t~n n • ~ ~ 13
8S UWY L S RP N • ( G /
( f r s • zʼn . ] b G N ] ∫ " # %& # '# • i . G r d • n 8 . • A IP XLKr|f G zʼ s n 14
• 28: . ]ac nʼ87= • • 1 5 .
]ac nʼ87= • z n z k 15
. 28: 16 • = L ] . ( ʼ
s ~(s • :L KP N= LX . r (s zʼ s ~(s
. 1 5 17 • 5RRP XUPK ?UXL IWU .
• 1 RL^ ? XYWPNP . • ? XYWPNP s1 RL^ s n
. 1 5 18
• 87= ʼ i i • 28: 1 5 •
8S UWY L S RP N • 28: 1 5 z~ • ? XYWPNP zr ʼsf • • 19