自然言語処理(NLP)
機械に人間の言語を処理させること
...This is just a bunch of
words to explain what
natural language
processing is and is not
meant to be read. All I
want to say is that it’s
all about having
machines do useful
stuff with language...
!
Skip-Thought Vectors [Kiros+ 2015]
skip-gram の文章版(文章→前後の文章)
... I got back home. I could see the cat on the steps. This was strange. …
encoder RNNが文章の単語ベクトル
を一つずつ読み込んでいく
内部状態を文章のベクトルとして使う
2つの decoder RNNがそのベクトルか
ら前後の文章を生成
最大対数尤度
... I got back home. I could see the cat on the steps. This was strange. …
中心の文章が「I could see the cat on the steps.」の場合、
前文decoder が「I got back home.」と
後文decoder が「This was strange.」の出力確率を高くする
目標関数
後の文章 前の文章
Neural Storyteller の仕組み
そしたら、画像→キャプション→恋愛小説 が成り立つ
skip-
thought
空間
画像・文章
共通空間
x
F(x) = x - c + b
F(x)
“Smiling businessmen
walking together”
“...Of course, i had no
intention of letting him go...”
文献
• Y Bengio, R Ducharme, P Vincent, C Jauvin. 2003. A Neural Probabilistic Model. Journal of
Machine Learning Research 3 (2003) 1137-1155.
• J Firth. 1957. A synopsis of linguistic theory 1930-1955. In Studies in Linguistic Analysis
pp. 1-32
• A Frome, G Corrado, J Shlens, S Bengio, J Dean, M Ranzato, T Mikolov. 2013. DeViSE: A
Deep Visual-Semantic Embedding Model
• Z Harris. 1954. Distributional structure. Word, 10(23):146-162
• G Hinton, J McClelland, D Rumelhart. 1986. Distributed Representations. In Parallel
distributed processing: Explorations in the microstructure of cognition, Volume I. Chapter
3, pp. 77-109, Cambridge, MA: MIT Press.
• N Kalchbrenner, E Grefenstette, P Blunsom. A convolutional neural network for modelling
sentences. ACL, 2014.
文献
• R Kiros, R Salakhutdinov, R Zemel. 2014. Unifying Visual-Semantic Embeddings with
Multimodal Neural Language Models.
• R Kiros, Y Zhu, R Salakhutdinov, R Zemel, A Torralba, R Urtasun, S Fidler. 2015. Skip-
Thought Vectors.
• Q Le, T Mikolov. Distributed representations of sentences and documents. ICML, 2014.
• H Lee, R Grosse, R Ranganath, A Ng. 2009. Convolutional Deep Belief Networks for
Scalable Unsupervised Learning of Hierarchical Representations.
• O Levy, Y Goldberg, I Dagan. 2014. Improving Distributional Similarity with Lessons
Learned from Word Embeddings.
• T Mikolov, K Chen, G Corrado, J Dean. 2013. Efficient Estimation of Word Representations
in Vector Space.
文献
• M Norouzi, T Mikolov, S Bengio, Y Singer, J Shlens, A Frome, G Corrado, J Dean. 2013.
Zero-shot Learning by Convex Combination of Semantic Embeddings
• J Pennington, R Socher, C Manning. 2014. GloVe: Global Vectors for Word Representation.
• R Richens. 1956. Preprogramming for Mechanical Translation. Mechanical Translation,
vol.3, no.1, July 1956; pp. 20-25.
• X Rong. 2014. word2vec Parameter Learning Explained.
• R Socher, M Ganjoo, C Manning, A Ng. 2013. Zero-Shot Learning Through Cross-Modal
Transfer
• R Socher, A Perelygin, J Wu, J Chuang, C Manning, A Ng, C Potts. Recursive deep models
for semantic compositionality over a sentiment treebank. In EMNLP, 2013.
文献
• I Sutskever, O Vinyals, Q Le. 2014. Sequence to Sequence Learning with Neural Networks.
• O Vinyals, A toshev, S Bengio, D Erhan. 2014. Show and Tell: A Neural Image Caption
Generator.
• W Zou. 2013. Bilingual Word Embeddings for Phrase-Based Machine Translation.
Editor's Notes
Elman, Jeffrey L.; et al. (1996). "Preface". Rethinking Innateness: A Connectionist Perspective on Development (Neural Network Modeling and Connectionism). A Bradford Book. ISBN 978-0262550307. connectionism (a term introduced by Donald Hebb in 1940s, and the name we adopt here)