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
260
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
270
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
2.2k
論文紹介: IRのためのパラメータチューニング / ir-tuning
nmasahiro
0
490
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
30
40k
論文紹介 : Population Based Augmentation: Efficient Learning of Augmentation Policy Schedules
nmasahiro
1
710
広告とAI(とハイパーパラメータ最適化) / Ad with AI
nmasahiro
1
2k
Other Decks in Research
See All in Research
Sat2City:3D City Generation from A Single Satellite Image with Cascaded Latent Diffusion
satai
3
170
[RSJ25] Enhancing VLA Performance in Understanding and Executing Free-form Instructions via Visual Prompt-based Paraphrasing
keio_smilab
PRO
0
150
EcoWikiRS: Learning Ecological Representation of Satellite Images from Weak Supervision with Species Observation and Wikipedia
satai
3
290
[論文紹介] Intuitive Fine-Tuning
ryou0634
0
130
日本語新聞記事を用いた大規模言語モデルの暗記定量化 / LLMC2025
upura
0
270
VectorLLM: Human-like Extraction of Structured Building Contours via Multimodal LLMs
satai
4
360
MIRU2025 チュートリアル講演「ロボット基盤モデルの最前線」
haraduka
15
9.3k
引力・斥力を制御可能なランダム部分集合の確率分布
wasyro
0
270
電通総研の生成AI・エージェントの取り組みエンジニアリング業務向けAI活用事例紹介
isidaitc
1
1.1k
超高速データサイエンス
matsui_528
1
170
Mechanistic Interpretability:解釈可能性研究の新たな潮流
koshiro_aoki
1
490
「どう育てるか」より「どう働きたいか」〜スクラムマスターの最初の一歩〜
hirakawa51
0
960
Featured
See All Featured
Music & Morning Musume
bryan
46
6.9k
Reflections from 52 weeks, 52 projects
jeffersonlam
354
21k
Rebuilding a faster, lazier Slack
samanthasiow
84
9.2k
Fireside Chat
paigeccino
41
3.7k
Building Adaptive Systems
keathley
44
2.8k
Optimising Largest Contentful Paint
csswizardry
37
3.5k
[RailsConf 2023 Opening Keynote] The Magic of Rails
eileencodes
31
9.7k
A Modern Web Designer's Workflow
chriscoyier
697
190k
I Don’t Have Time: Getting Over the Fear to Launch Your Podcast
jcasabona
34
2.5k
Typedesign – Prime Four
hannesfritz
42
2.8k
Unsuck your backbone
ammeep
671
58k
The World Runs on Bad Software
bkeepers
PRO
72
11k
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