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
On Adversarial Example for Character-Level Neu...
Search
onizuka laboratory
October 17, 2018
Research
0
100
On Adversarial Example for Character-Level Neural Machine Translation
弊研究室で行なったCOLING2018読み会の発表資料です。
onizuka laboratory
October 17, 2018
Tweet
Share
More Decks by onizuka laboratory
See All by onizuka laboratory
Phrase-Based & Neural Unsupervised Machine Translation
onilab
0
110
Tell-and-Answer: Towards Explainable Visual Question Answering using Attributes and Captions
onilab
0
71
Card-660: A Reliable Evaluation Framework for Rare Word Representation Models
onilab
0
33
A Word-Complexity Lexicon and A Neural Readability Ranking Model for Lexical Simplification
onilab
0
120
Integrating Transformer and Paraphrase Rules for Sentence Simplification
onilab
0
59
An Auto-Encoder Matching Model for Learning Utterance-Level Semantic Dependency in Dialogue Generation
onilab
0
55
Generating More Interesting Responses in Neural Conversation Models with Distributional Constraints
onilab
0
100
Modeling Multi-turn Conversation with Deep Utterance Aggregation
onilab
0
95
Learning Semantic Sentence Embeddings using Pair-wise Discriminator
onilab
0
120
Other Decks in Research
See All in Research
When Submarine Cables Go Dark: Examining the Web Services Resilience Amid Global Internet Disruptions
irvin
0
300
Combinatorial Search with Generators
kei18
0
770
Vision and LanguageからのEmbodied AIとAI for Science
yushiku
PRO
1
530
最適決定木を用いた処方的価格最適化
mickey_kubo
4
1.9k
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
240
最適化と機械学習による問題解決
mickey_kubo
0
170
MetaEarth: A Generative Foundation Model for Global-Scale Remote Sensing Image Generation
satai
4
190
【緊急警告】日本の未来設計図 ~沈没か、再生か。国民と断行するラストチャンス~
yuutakasan
0
150
言語モデルの地図:確率分布と情報幾何による類似性の可視化
shimosan
5
1.4k
ストレス計測方法の確立に向けたマルチモーダルデータの活用
yurikomium
0
1.5k
なめらかなシステムと運用維持の終わらぬ未来 / dicomo2025_coherently_fittable_system
monochromegane
0
2.8k
AI エージェントを活用した研究再現性の自動定量評価 / scisci2025
upura
1
150
Featured
See All Featured
ReactJS: Keep Simple. Everything can be a component!
pedronauck
667
120k
Why Our Code Smells
bkeepers
PRO
339
57k
Refactoring Trust on Your Teams (GOTO; Chicago 2020)
rmw
34
3.1k
Java REST API Framework Comparison - PWX 2021
mraible
33
8.8k
A Modern Web Designer's Workflow
chriscoyier
696
190k
RailsConf & Balkan Ruby 2019: The Past, Present, and Future of Rails at GitHub
eileencodes
139
34k
BBQ
matthewcrist
89
9.8k
10 Git Anti Patterns You Should be Aware of
lemiorhan
PRO
656
61k
Become a Pro
speakerdeck
PRO
29
5.5k
Put a Button on it: Removing Barriers to Going Fast.
kastner
60
4k
Templates, Plugins, & Blocks: Oh My! Creating the theme that thinks of everything
marktimemedia
31
2.5k
Practical Tips for Bootstrapping Information Extraction Pipelines
honnibal
PRO
23
1.4k
Transcript
A C ( A ) , A ( - A
D • L • White-box adversarial example • U • T O E • M NMT N O • EJ 2
• - - MadA x M wmt boecN - M
• N p vsM h vsM h - i • - T Mk M Tn r • Mg - ad l 3
m 4 n ) 1M T F 7777777 2 0
) 1 2 k o i m h g F a k e ( i ct b AM rd
& & - & B & - & B I
O 5 & &
- 6 hx hy #(x, y) 4 ' d a
b W • - f e • - i e • e • e
! 7 x = $%% , … , $%( ;
$*% , … , $*( ; … $+% , … , $+( ! , - - ⃗ !/01 = 0; … ; 0, … , 0, … , −1,0, … , 1,0 0 , … , 0 / ; 0, … , - ! ∇6789 : x, y = ∇< :(x, y)? @ ⃗ !/01 = A: 1 A$/0 − A:(B) A$/0 ! -
)( & ) )( & 8 )( & A A&(
) & & & & TC , )(C A ! "($, &' ) A A & ) )( & &( ) & & & & TC ! !′ −"($, &'+ ) A A
- 5 9 5-2 e i O n O %
mehG i O B omG g Mm mac G n O 0 5 - % l d
hnr 10 D C-41.2 i C a a VWa 2
5 2 25 9 3 56 0 1 D Nt 5 5 e ckN e D L RNN r E gb LTN sISoLN F
5 E6C E 9 EE68 11 EE68 0 B 1
D CE . E 4F6B F E 7 68 F E 7 68 F E 7 68 F E 7 68 03 . -4 o sw 2 5 ec A D AE hIdkb S U r S n F E 7A ci Ig 2 5 x a S SDF6Bb a F E 7 68 2 5bl aR a A6 AC E 9 ci Ig nb Lp t
12 ! = #$ #% … #' … #( #'
C !) = #$ #% … UNK … #( C success !123 = 4 1, if BLEU(!, !123 ) BLEU(!, !) ) ≥ ? 0, otherwise A
- 13
14 !" = $% $& … $() … $*
15
16
17 • c T M aN Me N M gM
• b N h b N h M e N d M gM • - - i M gM