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
結合トピックモデル
Search
Kento Nozawa
March 29, 2016
Research
0
1k
結合トピックモデル
2016年3月29日に『トピックモデルによる統計的潜在意味解析』
読書会ファイナル ~佐藤一誠先生スペシャル~のLTで発表しました
Kento Nozawa
March 29, 2016
Tweet
Share
More Decks by Kento Nozawa
See All by Kento Nozawa
Analysis on Negative Sample Size in Contrastive Unsupervised Representation Learning
nzw0301
0
91
[IJCAI-ECAI 2022] Evaluation Methods for Representation Learning: A Survey
nzw0301
0
500
[NeurIPS Japan meetup 2021 talk] Understanding Negative Samples in Instance Discriminative Self-supervised Representation Learning
nzw0301
0
130
[IBIS2021] 対照的自己教師付き表現学習おける負例数の解析
nzw0301
0
130
Understanding Negative Samples in Instance Discriminative Self-supervised Representation Learning
nzw0301
0
400
Introduction of PAC-Bayes and its Application for Contrastive Unsupervised Representation Learning
nzw0301
2
730
NLP Tutorial; word representation learning
nzw0301
0
160
Analyzing Centralities of Embedded Nodes
nzw0301
0
120
Paper Reading: Noise-Contrastive Estimation of Unnormalized Statistical Models, with Applications to Natural Image Statistics
nzw0301
2
1k
Other Decks in Research
See All in Research
継続的な研究費獲得のための考え方
moda0
2
540
自動運転・AIシステムの問題を賢く探す・賢く直す / Smart Search & Repair Techniques for Automated Driving Systems and AI Systems
ishikawafyu
0
140
SSII2024 [OS3] 基盤モデル(オープニング)
ssii
PRO
0
280
高精度、高効率アナログCompute-in-Memory回路に向けて
kentaroy47
2
110
医療分野におけるLLMの現状と応用可能性について
kento1109
5
550
中高生にSFを読んでもらうには
ichiiida
1
830
「人間にAIはどのように辿り着けばよいのか?ー 系統的汎化からの第一歩 ー」@第22回 Language and Robotics研究会
maguro27
0
410
SSII2024 [PD] SSII、次の30年への期待
ssii
PRO
2
1.3k
DroidKaigi CfP分析
yukihiromori
0
110
ヘルプデスクの事例で学ぶAIエージェント
masatoto
11
4.5k
大規模言語モデル (LLM) の技術と最新動向
ikuyamada
30
15k
アジャイルコミュニティが、宗教ポイと云われるのは何故なのか?
fujiihideo
0
260
Featured
See All Featured
Dealing with People You Can't Stand - Big Design 2015
cassininazir
360
22k
Designing the Hi-DPI Web
ddemaree
276
34k
Sharpening the Axe: The Primacy of Toolmaking
bcantrill
26
1.6k
Art, The Web, and Tiny UX
lynnandtonic
291
20k
How to Create Impact in a Changing Tech Landscape [PerfNow 2023]
tammyeverts
34
1.9k
"I'm Feeling Lucky" - Building Great Search Experiences for Today's Users (#IAC19)
danielanewman
224
21k
Raft: Consensus for Rubyists
vanstee
134
6.5k
Writing Fast Ruby
sferik
623
60k
How STYLIGHT went responsive
nonsquared
93
5k
How GitHub (no longer) Works
holman
305
140k
Designing on Purpose - Digital PM Summit 2013
jponch
113
6.6k
How to Think Like a Performance Engineer
csswizardry
4
590
Transcript
݁߹τϐοΫϞσϧ ʰτϐοΫϞσϧʹΑΔ౷ܭతજࡏҙຯղੳʱ ಡॻձϑΝΠφϧ ~ࠤ౻Ұઌੜεϖγϟϧ~ ݈ਓ (@nzw0301) 2016-03-29
ࣗݾհ ͡Ί·ͯ͠ ݈ਓ (@nzw0301) य़͔ΒஜେͰM1 ڵຯ • ػցֶशɼNLPɼάϥϑɼDL
݁߹τϐοΫϞσϧ จॻσʔλͱରԠ͢ΔใΛ߹Θֶͤͯश • ຊޠͱӳޠ • Ϩγϐͱࡐྉ • ୯ޠͱͦͷࢺ ࢀߟɿ௨ৗͷLDAͷάϥϑΟΧϧϞσϧ 3
D N2 N1 K 2 1 w2 i w1 i ✓ ↵ 1 z1 i z2 i 2 D N K wi ✓ ↵ zi
z ͷαϯϓϦϯάࣜ • 3ষp55ʹैͬͯಋग़Մೳ • ৄ͘͠ http://nzw0301.github.io/2016/02/jointTopicModelsEquation • ҎԼͷ͔ࣜΒGibbs SamplingͷࣜΛٻΊΔ
ࢀߟɿ௨ৗͷLDA 4 p(z1 d,i = k|w1 d,i = v, W1 \d,i , W2, Z1 \d,i , Z2, ↵, 1, 2) p(z1 d,i = k|w1 d,i = v, W1 \d,i , Z1 \d,i , ↵, 1)
ࣜมܗͷ݁Ռ • ݁߹τϐοΫϞσϧͷαϯϓϦϯάࣜ • φ௨ৗͷLDAͱಉ͡ • θʹ͍ͭͯɼؚ·ΕΔ߲͕ิॿใͷ͚ͩ૿͑Δ ࢀߟɿ௨ৗͷLDAͷαϯϓϦϯάࣜ 5 n1
k,v,\d,i + v P v0 (n1 k,v0,\d,i + v0 ) n1 d,k,\d,i + n2 d,k + ↵k P k0 (n1 d,k0,\d,i + n2 d,k0 + ↵k0 ) n1 k,v,\d,i + v P v0 (n1 k,v0,\d,i + v0 ) n1 d,k,\d,i + ↵k P k0 (n1 d,k0,\d,i + ↵k0 )
࣮ߦྫ • ର༁ίʔύεΛఆͨ͠؆୯ͳྫ • ݴޠ͕ҧͬͯτϐοΫڞ௨ http://nzw0301.github.io/2016/02/jointTopicModelsEquation ࣮ɿLDAͷαϯϓϧࣜͰ͏౷ܭྔΛྻʹ 6
ࢀߟจݙ • ؠా ۩࣏. τϐοΫϞσϧ. ߨஊࣾ. 2015. (MLPػցֶशϓϩϑΣογϣφϧγϦʔζ). • ࠤ౻
Ұ. τϐοΫϞσϧʹΑΔ౷ܭతજࡏҙຯղੳ. ίϩφࣾ. 2015. (ࣗવݴޠॲཧγ Ϧʔζ, 8). • David Mimno, Hanna M. Wallach, Jason Naradowsky, David A. Smith and Andrew McCallum. 2009. Polylingual Topic Models. in EMNLP. 7