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
文献紹介: Delete, Retrieve, Generate: A Simple Appr...
Search
Yumeto Inaoka
June 20, 2018
Research
0
170
文献紹介: Delete, Retrieve, Generate: A Simple Approach to Sentiment and Style Transfer
2018/06/20の文献紹介で発表
Yumeto Inaoka
June 20, 2018
Tweet
Share
More Decks by Yumeto Inaoka
See All by Yumeto Inaoka
文献紹介: Quantity doesn’t buy quality syntax with neural language models
yumeto
1
150
文献紹介: Open Domain Web Keyphrase Extraction Beyond Language Modeling
yumeto
0
200
文献紹介: Self-Supervised_Neural_Machine_Translation
yumeto
0
130
文献紹介: Comparing and Developing Tools to Measure the Readability of Domain-Specific Texts
yumeto
0
140
文献紹介: PAWS: Paraphrase Adversaries from Word Scrambling
yumeto
0
120
文献紹介: Beyond BLEU: Training Neural Machine Translation with Semantic Similarity
yumeto
0
240
文献紹介: EditNTS: An Neural Programmer-Interpreter Model for Sentence Simplification through Explicit Editing
yumeto
0
300
文献紹介: Decomposable Neural Paraphrase Generation
yumeto
0
200
文献紹介: Analyzing the Limitations of Cross-lingual Word Embedding Mappings
yumeto
0
200
Other Decks in Research
See All in Research
Data-centric AI勉強会 「ロボットにおけるData-centric AI」
haraduka
0
470
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
1
330
Leveraging LLMs for Unsupervised Dense Retriever Ranking (SIGIR 2024)
kampersanda
2
310
IM2024
mamoruk
0
230
研究を支える拡張性の高い ワークフローツールの提案 / Proposal of highly expandable workflow tools to support research
linyows
0
310
【NLPコロキウム】Stepwise Alignment for Constrained Language Model Policy Optimization (NeurIPS 2024)
akifumi_wachi
3
540
大規模言語モデルを用いたニュースデータのセンチメント判定モデルの開発および実体経済センチメントインデックスの構成
nomamist
1
130
[輪講] Transformer Layers as Painters
nk35jk
4
690
メールからの名刺情報抽出におけるLLM活用 / Use of LLM in extracting business card information from e-mails
sansan_randd
2
430
言語と数理の交差点:テキストの埋め込みと構造のモデル化 (IBIS 2024 チュートリアル)
yukiar
5
1.1k
The many faces of AI and the role of mathematics
gpeyre
1
1.7k
PhD Defence: Considering Temporal and Contextual Information for Lexical Semantic Change Detection
a1da4
0
130
Featured
See All Featured
XXLCSS - How to scale CSS and keep your sanity
sugarenia
248
1.3M
Reflections from 52 weeks, 52 projects
jeffersonlam
348
20k
Practical Tips for Bootstrapping Information Extraction Pipelines
honnibal
PRO
13
1k
Performance Is Good for Brains [We Love Speed 2024]
tammyeverts
7
650
GraphQLとの向き合い方2022年版
quramy
44
14k
Bash Introduction
62gerente
611
210k
A better future with KSS
kneath
238
17k
Typedesign – Prime Four
hannesfritz
41
2.5k
Sharpening the Axe: The Primacy of Toolmaking
bcantrill
40
2k
Templates, Plugins, & Blocks: Oh My! Creating the theme that thinks of everything
marktimemedia
30
2.2k
Measuring & Analyzing Core Web Vitals
bluesmoon
6
260
Helping Users Find Their Own Way: Creating Modern Search Experiences
danielanewman
29
2.5k
Transcript
Delete, Retrieve, Generate: A Simple Approach to Sentiment and Style
Transfer Juncen Li, Robin Jia, He He, Percy Liang. Proceedings of NAACL-HLT 2018, pages 1865–1874, 2018. จݙհ Ԭٕज़Պֶେֶࣗવݴޠॲཧݚڀࣨ ҴԬເਓ
"CTUSBDU wײͳͲͷଐੑΛɺଐੑʹґଘ͠ͳ͍༰Λ อ࣋ͭͭ͠มΛߦ͏λεΫ wֶशʹଐੑͷΈҟͳΔΑ͏ͳจϖΞΛ༻͠ͳ͍ wϑϨʔζΛ%FMFUF 3FUSJFWFͯ͠ɺͦΕΒΛݩʹ ࠷ऴతͳग़ྗΛ(FOFSBUF͢Δ wैདྷख๏ΑΓଟ͘ͷೖྗʹ͓͍ͯจ๏త͔ͭ దͳग़ྗ͕ੜ͞ΕΔ͜ͱΛਓखධՁͰ֬ೝ !2
*OUSPEVDUJPO w ײελΠϧɺ੍࣌ͷΑ͏ͳଐੑΛ੍ޚͰ͖Δ จੜʹؔ৺͕ߴ·͍ͬͯΔ w ௨ৗɺଐੑͷΈҟͳΔύϥϨϧσʔλ༻ Ͱ͖ͣɺଐੑ͕ϥϕϧ͚͞ΕͨจͷΈΛ༻ w ͜Ε·Ͱʹ("/Λ༻͍ͨख๏͕ఏҊ͞Ε͍ͯΔ͕ɺ ग़ྗ͕࣭Ͱ͋Δ͜ͱ͕ਓखධՁͰ໌
!3
*OUSPEVDUJPO w ଐੑʹӨڹΛ༩͑Δ୯ޠ۟ Ҏ֎΄ͱΜͲมߋͤͣ͞ʹ ଐੑมͰ͖Δ߹͕ଟ͍ w ΑΓ୯७Ͱֶश͕؆୯ͳ Ұ࿈ͷγεςϜΛఏҊ !4
"QQSPBDI !5
"QQSPBDI !6
"QQSPBDI !7
"QQSPBDI !8
"QQSPBDI !9
"QQSPBDI !10
"QQSPBDI !11
"QQSPBDI !12
"QQSPBDI !13
"QQSPBDI !14
%FMFUF w ײଐੑͷ߹ɺlQPTJUJWFzͷ࣌ʹݶͬͯΑ͘ग़ݱ ͢ΔOHSBNͱlOFHBUJWFzͷ࣌ʹݶͬͯΑ͘ग़ݱ ͢ΔOHSBNΛଐੑϚʔΧͱͯ͠আ w OHSBN͔ΒଐੑΛྨ͢ΔφΠʔϒϕΠζྨث ʹ͓͚ΔOHSBNͷ͖͕݅֬ࢦఆͷᮢΛ ͑ͨࡍʹଐੑϚʔΧͱ͢Δ !15
3FUSJFWF w ͭͷ୯ޠܥྻͷڑ͕Ұ൪খ͍͞ͷΛऔΓग़͢ w ڑͷܭࢉํ๏ҎԼͷͭΛ࣮ݧ 5'*%'ͰॏΈ͚ͮΒΕͨ୯ޠͷॏͳΓ DPOUFOUFNCFEEJOHTͷϢʔΫϦουڑ ˢEFMFUFޙͷจΛ3//FODPEFSʹೖྗͨ݁͠Ռ
!16
(FOFSBUF %FMFUF0OMZ w %FMFUFޙͷจͱଐੑ͔Β%FMFUFલͷจΛ෮ݩ͢Δ Α͏ʹֶशΛߦ͏ !17
(FOFSBUF %FMFUF"OE3FUSJFWF w ී௨ʹֶशͤ͞ΔͱɺจଐੑϚʔΧΛ݀ຒΊ͢Δ ͚ͩͷֶशʹͳͬͯ͠·͏ ˠεϜʔδϯά͕ߦΘΕͣྲྀெʹͳΒͳ͍ w ଐੑϚʔΧ֬తʹϊΠζΛՃ͑Δ ˡฤूڑ͕ͰಉଐੑͷผϚʔΧஔ͖͑Δ !18
&YQFSJNFOUT w :FMQϨϏϡʔɺ"NB[POϨϏϡʔͷײΛస w ը૾ΩϟϓγϣϯΛΑΓϩϚϯνοΫ͔ϢʔϞϥε ʹͳΔΑ͏มߋ w ैདྷख๏ɺ)VNBO3FGFSFODFɺఏҊ͢Δͭͷ γεςϜΛൺֱ w
)VNBO3FGFSFODF.5VSLͰऩू !19
%BUBTFUT w Ωϟϓγϣϯͷςετηοτࣄ࣮ͷΈͰ͋ΔͨΊ ଐੑϚʔΧͷআͳ͘ɺૠೖͷΈ !20
&YQFSJNFOUBM%FUBJMT w EJNFOTJPOBMXPSEWFDUPST w TJOHMFMBZFS(36XJUIIJEEFOVOJUT w NBYPVUBDUJWBUJPOGVODUJPO w "EBEFMUBXJUIBNJOJCBUDITJ[FPG w
CFBNTFBSDIXJUIBCFBNTJ[FPG !21
)VNBO&WBMVBUJPO w .5VSLͰޏͬͨϫʔΧʔ͕γεςϜͷग़ྗΛධՁ w ஈ֊ͷϦοΧʔτईͰจ๏ੑɺଐੑɺ ҙຯͷอ࣋ΛධՁ w ·ͨͱධՁ͞Εͨ߹ʹग़ྗޭͱݟ၏͢ w ແ࡞ҝʹநग़ͨ͠αϯϓϧΛධՁ
ʢ֤ଐੑ͝ͱʹαϯϓϧʣ !22
)VNBO&WBMVBUJPO !23
$PODMVTJPO w ςΩετଐੑมʹ͓͍ͯैདྷͷ("/ʹΑΔख๏ ΑΓߴੑೳͳख๏ΛఏҊ w จͷଐੑʹӨڹΛ༩͑Δ۟ہॴతͰ͋Δ͜ͱ͕ ޮՌΛେ͖͍ͯ͘͠Δ w কདྷతʹOHSBNΑΓҰൠతͳଐੑͷ֓೦Λ։ൃ ͢Δͱ༗ӹ͕ͩɺΑΓؼೲతόΠΞεΛ͏
!24