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
EMNLP2015読み会:Effective Approaches to Attention-...
Search
tkng
October 24, 2015
Research
2
3.9k
EMNLP2015読み会:Effective Approaches to Attention-based Neural Machine Translation
tkng
October 24, 2015
Tweet
Share
More Decks by tkng
See All by tkng
LSTMを用いた自然言語処理について
tkng
3
3.7k
自然言語処理と深層学習の最先端
tkng
16
7.6k
basis-of-optimization.pdf
tkng
1
1.3k
Other Decks in Research
See All in Research
Prithvi-EO-2.0: A Versatile Multi-Temporal Foundation Model for Earth Observation Applications
satai
3
160
The Economics of Platforms 輪読会 第1章
tomonatu8
0
150
博士学位論文予備審査 / Scaling Telemetry Workloads in Cloud Applications: Techniques for Instrumentation, Storage, and Mining
yuukit
1
1.7k
精度を無視しない推薦多様化の評価指標
kuri8ive
1
370
セミコン地域における総合交通戦略
trafficbrain
0
120
CUNY DHI_Lightning Talks_2024
digitalfellow
0
520
Weekly AI Agents News! 11月号 論文のアーカイブ
masatoto
0
300
NeurIPS 2024 参加報告 & 論文紹介 (SACPO, Ctrl-G)
reisato12345
0
350
大規模日本語VLM Asagi-VLMにおける合成データセットの構築とモデル実装
kuehara
5
1k
Leveraging LLMs for Unsupervised Dense Retriever Ranking (SIGIR 2024)
kampersanda
2
310
言語モデルLUKEを経済の知識に特化させたモデル「UBKE-LUKE」について
petter0201
0
240
CoRL2024サーベイ
rpc
2
1.6k
Featured
See All Featured
Easily Structure & Communicate Ideas using Wireframe
afnizarnur
193
16k
Making the Leap to Tech Lead
cromwellryan
133
9.1k
Faster Mobile Websites
deanohume
306
31k
Writing Fast Ruby
sferik
628
61k
Done Done
chrislema
182
16k
Evolution of real-time – Irina Nazarova, EuRuKo, 2024
irinanazarova
6
580
Art, The Web, and Tiny UX
lynnandtonic
298
20k
[Rails World 2023 - Day 1 Closing Keynote] - The Magic of Rails
eileencodes
33
2.1k
[RailsConf 2023] Rails as a piece of cake
palkan
53
5.3k
Building a Scalable Design System with Sketch
lauravandoore
461
33k
Chrome DevTools: State of the Union 2024 - Debugging React & Beyond
addyosmani
4
380
Mobile First: as difficult as doing things right
swwweet
223
9.5k
Transcript
Effective Approaches to Attention-based Neural Machine Translation Authors: Minh-Thang LuongɹHieu
PhamɹChristopher D. Manning ಡΉਓ: ಙӬ೭ ਤશͯ͜ͷจ͔ΒҾ༻ &./-1ಡΈձ
ࣗݾհɿಙӬ೭ • Twitter ID: @tkng • εϚʔτχϡʔεגࣜձࣾͰNLPͬͯ·͢
ࠓͷจʁ • Effective Approaches to Attention-based Neural Machine Translation •
ڈ͙Β͍͔ΒྲྀߦΓ࢝Ίͨseq2seqܥͷख ๏ͷ֦ு
Seq2seq modelͱʁ • Encoder/Decoder modelͱݴ͏ • ༁ݩͷจΛݻఆͷϕΫτϧʹΤϯίʔυ ͯ͠ɺ͔ͦ͜Β༁ޙͷจΛσίʔυ͢Δ • ՄมͷσʔλऔΓѻ͍͕͍͠ͷͰɺ
͑ͯݻఆʹͯ͠͠·͏ͱ͍͏ൃ
Ͳ͏ͬͯݻఆʹΤϯίʔυ ͢Δͷʁ • recurrent neural networkΛ͏ • http://colah.github.io/posts/2015-08-Understanding-LSTMs/ • http://kaishengtai.github.io/static/slides/treelstm-acl2015.pdf
• LSTM = recurrent neural networkͷҰछ
Seq2seqϞσϧͰͷ༁
Seq2seq·ͰͷಓͷΓ (1) • Recurrent Continuous Translation Models (EMNLP2013) • Learning
Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation (EMNLP2014)
Seq2seq·ͰͷಓͷΓ (2) • Sequence to Sequence Learning with Neural Networks
(NIPS2014) • ൺֱతγϯϓϧͳStacked LSTM͕ྑ͍ੑೳΛ ࣔ͢͜ͱ͕࣮ݧͰࣔ͞Εͨ • ϏʔϜαʔνɺٯॱͰͷೖྗɺΞϯαϯϒϧ ͷ3छྨͷ͕ೖ͍ͬͯΔ
Seq2seqϞσϧͷऑ • จʹऑ͍ • ݻ༗໊ࢺ͕ೖΕସΘΔ
AttentionʹΑΔվળ [Bahdanau+ 2015] • DecodeͷࡍͷContextʹEncodeͷࡍͷ֤࣌ࠁ ʹ͓͚ΔӅΕঢ়ଶͷॏΈ͖Λ༻͍Δ • ॏΈࣗମRNNͰܭࢉ͢Δ
ࠓճͷจͷߩݙ • ৽͍͠attention (local attention) ΛఏҊͨ͠ • ༁ݩจʹ͓͍ͯɺҐஔɹ͔ΒલޙD୯ޠ ͷӅΕঢ়ଶͷॏΈ͖ΛऔΔ •
ॏΈͷܭࢉglobal attentionͷ߹ͱಉ༷ • ɹ1ͭͣͭਐΊ͍ͯ͘߹ʢlocal-mʣ ͱɺ͜ΕࣗମRNNʹ͢Δ߹ʢlocal- pʣͷ2ͭΛ࣮ݧ͍ͯ͠Δ pt pt
local attention
local attentionͷҹ • ޠॱ͕ࣅ͍ͯΔݴޠؒͰͷ༁ͳΒɺ໌Β͔ ʹ͜ͷํ͕ྑͦ͞͏ • ӳΈ͍ͨʹޠॱ͕େ͖͘ҧ͏߹ɺ Ґஔɹͷਪఆࣗମ͕͍͠λεΫʹͳͬͪΌ ͍ͦ͏… pt
࣮ݧ݁ՌɿWMT'14
࣮ݧ݁ՌɿWMT'14 • Α͘ݟΔͱɺlocal attentionͰͷੑೳ্ +0.9ϙΠϯτ • ଞͷςΫχοΫͰՔ͍ͰΔϙΠϯτ͕ଟ͍
࣮ݧ݁ՌɿWMT'15
͍͔ͭ͘༁αϯϓϧ
·ͱΊ • Seq2seqϞσϧͷ֦ுͱͯ͠ɺlocal attention ΛఏҊͨ͠ • ఏҊख๏͍͔ͭ͘ͷ࣮ݧʹ͓͍ͯɺState of the artͷੑೳΛୡͨ͠
ײ • Local attentionΛඍ • ྨࣅ͢Δख๏ͱ۩ମతʹͲ͏ҧ͏͔͕໌շʹ ॻ͔Ε͓ͯΓɺಡΈ͔ͬͨ͢ • AttentionΛཧղͰ͖ͯΑ͔ͬͨʢখฒײʣ