Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Understanding_Back-Translation_at_scale.pdf

A3ea3bc5dde6ae2dd6eae71da9c418b0?s=47 MARUYAMA
January 02, 2019
0
69

 Understanding_Back-Translation_at_scale.pdf

A3ea3bc5dde6ae2dd6eae71da9c418b0?s=128

MARUYAMA

January 02, 2019
Tweet

More Decks by MARUYAMA

See All by MARUYAMA
vampire.pdf
A3ea3bc5dde6ae2dd6eae71da9c418b0?s=48 tmaru0204
0
69
Misspelling_Oblivious_Word_Embedding.pdf
A3ea3bc5dde6ae2dd6eae71da9c418b0?s=48 tmaru0204
0
52
Simple_Unsupervised_Summarization_by_Contextual_Matching.pdf
A3ea3bc5dde6ae2dd6eae71da9c418b0?s=48 tmaru0204
0
62
Controlling_Text_Complexity_in_Neural_Machine_Translation.pdf
A3ea3bc5dde6ae2dd6eae71da9c418b0?s=48 tmaru0204
0
61
20191028_literature-review.pdf
A3ea3bc5dde6ae2dd6eae71da9c418b0?s=48 tmaru0204
0
63
Hint-Based_Training_for_Non-Autoregressive_Machine_Translation.pdf
A3ea3bc5dde6ae2dd6eae71da9c418b0?s=48 tmaru0204
0
46
Soft_Contextual_Data_Augmentation_for_Neural_Machine_Translation_.pdf
A3ea3bc5dde6ae2dd6eae71da9c418b0?s=48 tmaru0204
0
61
An_Embarrassingly_Simple_Approach_for_Transfer_Learning_from_Pretrained_Language_Models_.pdf
A3ea3bc5dde6ae2dd6eae71da9c418b0?s=48 tmaru0204
0
56
Addressing_Trobulesome_Words_in_Neural_Machine_Translation.pdf
A3ea3bc5dde6ae2dd6eae71da9c418b0?s=48 tmaru0204
0
48

Featured

See All Featured
Intergalactic Javascript Robots from Outer Space
6804f1775cb4babfcc3851298566fbce?s=48 tanoku
261
25k
Agile that works and the tools we love
462dad0dd24c568a32dcdb0ae895ae1b?s=48 rasmusluckow
319
19k
ピンチをチャンスに：未来をつくるプロダクトロードマップ #pmconf2020
E425fe9f170efa0dfaf8ab69d7107418?s=48 aki_i
22
15k
I Don’t Have Time: Getting Over the Fear to Launch Your Podcast
B3ace07412a5178ea778e7eec161fc0a?s=48 jcasabona
12
900
The Language of Interfaces
Fde6c3a50ca518a909ef35c22c0ee0e4?s=48 destraynor
148
20k
Git: the NoSQL Database
20bfe76b3d6105641f879fe45cfc9272?s=48 bkeepers
PRO
415
59k
Creating an realtime collaboration tool: Agile Flush - .NET Oxford
Fd55de4174accaf3b4f030e43a8a70c6?s=48 marcduiker
4
450
Stop Working from a Prison Cell
C1daf20e5c49ff910745198ef9869ac2?s=48 hatefulcrawdad
261
17k
Writing Fast Ruby
1f74b13f1e5c6c69cb5d7fbaabb1e2cb?s=48 sferik
612
57k
Facilitating Awesome Meetings
245cee81a9c424266e5e401d844ea881?s=48 lara
29
3.9k
How to Ace a Technical Interview
2f5463832ccb768ccb4a1ca3607c27ef?s=48 jacobian
265
21k
Design by the Numbers
027d1ebf66cc039a0bd3b55eeadbe75d?s=48 sachag
271
17k

Transcript

  1. -JUFSBUVSFSFWJFX 6OEFSTUBOEJOH#BDL5SBOTMBUJPOBU4DBMF 4FSHFZ&EVOPW .ZMF0UU .JDIBFM"VMJ %BWJE(SBOHJFS &./-1 QBHFTr
 /BHBPLB6OJWFSTJUZPG5FDIOPMPHZ5BLVNJ.BSVZBNB

  2. ֓ཁ ⾣ٯ຋༁Λ࢖ͬͯɺٖࣅతͳύϥϨϧίʔύεΛੜ੒ ⾣ٯ຋༁ͷࡍʹ͸ɺग़ྗଆʹଟ༷ੑΛ࣋ͨͤΔ͜ͱ͕ॏཁ ⾣8.5Ͱ4UBUFPGUIFBSUΛୡ੒ #-&6QU

  3. ⾣ػց຋༁ͷੑೳ͸ɺσʔλαΠζʹେ͖͘Өڹ͞ΕΔ ⾣ύϥϨϧίʔύε͕গͳ͍৔߹ɺͲ͏͢΂͖͔ ⾣େن໛ͳ୯ݴޠίʔύεΛٯ຋༁ͯ͠ɺσʔλΛ֦ு ͸͡Ίʹ

  4. ٯ຋༁ʹΑΔσʔλ֦ு 4PVSDFTJEF 5BSHFUTJEF 1BSBMMFMDPSQVT 1BSBMMFMDPSQVT .POPMJOHVBM $PSQVT #BDLXBSE NPEFM 5SBJOJOH

  5. ٯ຋༁ʹΑΔσʔλ֦ு 4PVSDFTJEF 5BSHFUTJEF 1BSBMMFMDPSQVT 1BSBMMFMDPSQVT .POPMJOHVBM $PSQVT #BDLXBSE NPEFM

  6. ٯ຋༁ʹΑΔσʔλ֦ு 4PVSDFTJEF 5BSHFUTJEF 1BSBMMFMDPSQVT 1BSBMMFMDPSQVT .POPMJOHVBM $PSQVT #BDLXBSE NPEFM 4ZOUIFUJDEBUB

    (FOFSBUJPO

  7. ٯ຋༁ʹΑΔσʔλ֦ு 4ZOUIFUJDEBUB 4PVSDFTJEF 5BSHFUTJEF 1BSBMMFMDPSQVT 1BSBMMFMDPSQVT .POPMJOHVBM $PSQVT 'PSXBSE NPEFM

    5SBJOJOH

  8. ٯ຋༁ʹΑΔσʔλ֦ு 4ZOUIFUJDEBUB 5FTUEBUB 4PVSDFTJEF 5BSHFUTJEF 1BSBMMFMDPSQVT 1BSBMMFMDPSQVT .POPMJOHVBM $PSQVT 'PSXBSE

    NPEFM

  9. ٯ຋༁ʹΑΔσʔλ֦ு 4ZOUIFUJDEBUB 5FTUEBUB 4PVSDFTJEF 5BSHFUTJEF 1BSBMMFMDPSQVT 1BSBMMFMDPSQVT .POPMJOHVBM $PSQVT 3FTVMU

    'PSXBSE NPEFM

  10. ຋༁Ϟσϧ ⾣5SBOTGPSNFSΛར༻ &ODPEFS %FDPEFS %FDPEFS %FDPEFS 4PVSDFTFOUFODF #04 : :

  11. ٖࣅσʔλͷੜ੒ํ๏ ⾣CFBNTFBSDI΍HSFFEZTFBSDIΛར༻͞Ε͍ͯͨ ⾣Ϟσϧͷग़ྗ෼෍ͷ࠷େࣄޙ֬཰ ."1 ʹ஫໨ͨ͠ख๏ ⾣."1ਪఆΛٯ຋༁ʹ༻͍Δͱɺਅͷσʔλ෼෍ͱ͸ҟͳͬͨ ଟ༷ੑͷ௿͍ٖࣅσʔλʹͭͳ͕Δ

  12. ٖࣅσʔλͷੜ੒ํ๏ ⾣ٖࣅσʔλʹଟ༷ੑΛ࣋ͨͤΔͨΊʹɺ࣍ͷख๏ΛఏҊ TBNQMJOH Ϟσϧͷग़ྗ෼෍ʹैͬͯɺαϯϓϦϯά UPQ ग़ྗ֬཰্Ґ୯ޠΛબ୒ɾ࠶ਖ਼نԽ͠ɺαϯϓϦϯά

  13. ٖࣅσʔλͷੜ੒ํ๏ ⾣ٖࣅσʔλʹଟ༷ੑΛ࣋ͨͤΔͨΊʹɺ࣍ͷख๏ΛఏҊ #FBN OPJTF CFBNTFBSDIʹՃ͑ɺ࣍ͷछྨͷϊΠζΛՃ͑Δ   ֬཰Ͱɺग़ྗ୯ޠΛ࡟আ  

    ֬཰Ͱɺग़ྗ୯ޠΛ#-"/,ʹஔ׵   Ұ༷෼෍ʹैͬͯɺ୯ޠΛೖΕସ͑Δ

  14. σʔληοτ ⾣1BSB$SBXMDPSQVT &O%F .TFOUFODFQBJST ⾣OFXTDSBXMEBUB %F .TFOUFODFT ⾣OFXTUFTU

  15. ࣮ݧ݁Ռ ⾣ٯ຋༁ʹΑΔσʔλ֦ுͷޮՌ ɾCJUFYU .  ٖࣅσʔλ ɾCFBN HSFFEZΑΓ΋ TBNQMJOH CFBN

    OPJTFͷํ͕ ޮՌతͳٖࣅσʔλͰ͋Δ

  16. ࣮ݧ݁Ռ ⾣ٯ຋༁ʹΑΔσʔλ֦ுͷޮՌ

  17. ࣮ݧ݁Ռ ⾣ͳͥϊΠζΛՃ͑ͨํ͕ޮՌతͳͷ͔ʁ ɾHSFFEZ CFBNͷσʔλ͸ τϨʔχϯάʹద͍ͯ͠ͳ͍ ɾHSFFEZ CFBNʹΑͬͯੜ੒͞Εͨ σʔλ͸TBNQMJOH OPJTFΑΓ΋؆୯

  18. ࣮ݧ݁Ռ ⾣ͳͥϊΠζΛՃ͑ͨํ͕ޮՌతͳͷ͔ʁ ɾHSFFEZ CFBNͷσʔλ͸ τϨʔχϯάʹద͍ͯ͠ͳ͍ ɾHSFFEZ CFBNʹΑͬͯੜ੒͞Εͨ σʔλ͸TBNQMJOH OPJTFΑΓ΋؆୯

  19. ࣮ݧ݁Ռ ⾣ͳͥϊΠζΛՃ͑ͨํ͕ޮՌతͳͷ͔ʁ

  20. ࣮ݧ݁Ռ ⾣CJUFYUͱٖࣅσʔλͷൺ཰ ɾCJUFYU.TFOUFODFQBJST ɾCJUFYU .  ٖࣅσʔλ .  ˠCJUFYUVQTBNQMFSBUJP

  21. ࣮ݧ݁Ռ ⾣CJUFYUͱٖࣅσʔλͷൺ཰ ɾHSFFEZ CFBN͸ٖࣅσʔλͷൺ཰ʹ Өڹ͞ΕΔ ˠσʔλͷྔͰଟ༷ੑΛΧόʔ ɾTBNQMJOH OPJTF͸ଟ༷ੑ͕͋ΔͨΊ ٖࣅσʔλൺ཰ͷӨڹ͕গͳ͍

  22. ·ͱΊ ⾣ٯ຋༁Λ࢖ͬͯɺٖࣅతͳύϥϨϧίʔύεΛੜ੒ ⾣CFBN HSFFEZΑΓ΋ɺTBNQMJOH΍OPJTFͷΑ͏ʹ σʔλʹଟ༷ੑΛٖ࣋ͨͤͨࣅσʔλͷํ͕ޮՌత ⾣8.5Ͱ4UBUFPGUIFBSUΛୡ੒ #-&6QU