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
Char-rnn aurkezpena
Search
Manex Agirrezabal
March 14, 2016
Research
0
63
Char-rnn aurkezpena
Manex Agirrezabal
March 14, 2016
Tweet
Share
More Decks by Manex Agirrezabal
See All by Manex Agirrezabal
The Flipped Classroom model for teaching Conditional Random Fields in an NLP course
manexagirrezabal
0
23
NLP for poetry generation and analysis
manexagirrezabal
0
55
Institut seminar 2020
manexagirrezabal
0
22
Automatic Scansion of Poetry (KU)
manexagirrezabal
0
400
RANLP talk
manexagirrezabal
0
54
Defense (Final version)
manexagirrezabal
0
61
Poesiaren eskantsio automatikoa: Bi hizkuntzen azterketa
manexagirrezabal
0
59
CodeFEST literature presentation
manexagirrezabal
0
55
Ongoing work (in mid 2016)
manexagirrezabal
0
16
Other Decks in Research
See All in Research
Julia Tokyo #11 トーク: 「Juliaで歩く自動微分」
abap34
1
1.3k
20240209 データを肴に熊本の交通を考える会「車1割削減、渋滞半減、公共交通2倍」をめざし世界に学ぼう
trafficbrain
0
770
[研究室用] 2038年問題研究の現状報告
ran350
0
290
LiDARセキュリティ最前線
kentaroy47
0
270
生成AIを用いたText to SQLの最前線
masatoto
1
2k
MegaParticles: GPUを利用したStein Particle Filterによる点群6自由度姿勢推定
koide3
1
500
2024-01-23-az
sofievl
1
730
説明可能AI:代表的手法と最近の動向
yuyay
1
580
LLMマルチエージェントを俯瞰する
masatoto
26
15k
データで診て考える合志市の渋滞と公共交通 ~めざせ 車1割削減、渋滞半減、公共交通2倍~
trafficbrain
0
460
第12回全日本コンピュータビジョン勉強会:画像の自己教師あり学習における大規模データセット
naok615
0
500
Target trial emulationの概要
shuntaros
2
1.1k
Featured
See All Featured
The Invisible Customer
myddelton
114
12k
The Success of Rails: Ensuring Growth for the Next 100 Years
eileencodes
29
6k
Large-scale JavaScript Application Architecture
addyosmani
503
110k
4 Signs Your Business is Dying
shpigford
175
21k
Stop Working from a Prison Cell
hatefulcrawdad
265
19k
Intergalactic Javascript Robots from Outer Space
tanoku
266
26k
ReactJS: Keep Simple. Everything can be a component!
pedronauck
658
120k
Happy Clients
brianwarren
91
6.4k
[Rails World 2023 - Day 1 Closing Keynote] - The Magic of Rails
eileencodes
1
1.3k
Building Adaptive Systems
keathley
29
1.8k
Building Better People: How to give real-time feedback that sticks.
wjessup
353
18k
Fontdeck: Realign not Redesign
paulrobertlloyd
75
4.9k
Transcript
Poesiaren metrika DL bidez Manex Agirrezabal https://github.com/manexagirrezabal/char-rnn/
Proba ezberdinak TensorFlow: Sequence-to-sequence models https://www.tensorflow.org/versions/master/tutorials/seq2seq/index.html Torch: char-rnn (Andrew Karpathy)
https://github.com/karpathy/char-rnn/
Char-rnn http://karpathy.github.io/2015/05/21/rnn-effectiveness/ Karaktere mailako hizkuntz-ereduak sortzeko balio du. Sarrera gisa
testu hutsa.
Char-rnn Gure beharretarako moldatu behar: to swell the gourd and
plump the ha zel shells - ' - ' - ' - ' - ' wo man much missed how you call to me call to me ' - - ' - - ' - - ' - -
Char-rnn Dataset-a testu soil gisa: To_= swell_+ the_= gourd_+ and_=
plump_+ the_= ha_+ zel_= shells_+ To_= swell_+ the_= gourd_+ and_= plump_+ the_= hazel_+= shells_+ Wo_+ man_= much_= missed_+ how_= you_= call_+ to_= me_= call_+ to_= me_= Woman_+= much_= missed_+ how_= you_= call_+ to_= me_= call_+ to_= me_=
Char-rnn (training) $ th train.lua Parametroak: Model: [RNN, LSTM edo
GRU] rnn_size: LSTMaren (zelda) barruko tamaina num_layers: LSTMaren kapa kopurua seq_length: sekuentzian ikasteko karaktere kopurua
Char-rnn (prediction) $ th sample(mod).lua Parametroak: Model: eredu entrenatua Primetext:
sarrera testua (_ karakterearekin amaituta)
Char-rnn (prediction) Python programa bat (callSampleMod.py) aurreko programari deitzeko pausuz
pausu: $ th sampleMod.lua model M1 primetext “to_” = $ th sampleMod.lua model M1 primetext “to_= swell_” + $ th sampleMod.lua model M1 primetext “to_= swell_+ the_” = ...
Char-rnn (prediction) Arazoa: Hasieran, informazio gutxi duenez, batzuetan hanka sartzen
(+ propagatzen) du predikzioan. Adibidez, “to_” sarrerarekin Horrentzako soluzioa, predikzioa bi aldetara egitea.
Char-rnn (FW) Parametroak optimizatu nahi ditugu (seq_length, batch_size, rnn_size, ...)
Embedding-ak erabili nahi ditugu, baina gure hipotesia da ez dutela asko lagunduko.