16. Memory Networks 概要
より直接的に記憶をモデル化するアイディア [2]
記憶 m (object list) と4つのモジュールを持つモデル
何か入力を入れると、記憶をもとに出力を返してほし
い、という問題を解く
Joe went to the kitchen. Fred went to the kitchen. Joe picked up
the milk. Joe travelled to the office. Joe left the milk. Joe went to
the bathroom.
Where is the milk now? A: office
23. Basic model
● 前提
○ 根拠情報 (supporting fact) が与えられる
○ モデルの最終的な出力は単語
● 例
Joe went to the kitchen. Fred went to the kitchen. Joe picked up
the milk. Joe travelled to the office. Joe left the milk. Joe went
to the bathroom.
Where is the milk now? A: office
25. Output Feature Map
● output feature map
入力 x に対して、k 個の supporting memory を探索
(basic model では k = 2)
は入力とメモリのスコア関数 (後述)
26. ● output feature map
Joe went to the kitchen. Fred went to the kitchen. Joe picked up the milk.
Joe travelled to the office. Joe left the milk. Joe went to the bathroom.
Where is the milk now?
Output Feature Map
27. ● output feature map
Joe went to the kitchen. Fred went to the kitchen. Joe picked up the milk.
Joe travelled to the office. Joe left the milk. Joe went to the bathroom.
Where is the milk now?
Output Feature Map
28. ● output feature map
Joe went to the kitchen. Fred went to the kitchen. Joe picked up the milk.
Joe travelled to the office. Joe left the milk. Joe went to the bathroom.
Where is the milk now?
Output Feature Map
29. Output Feature Map
● output feature map
入力 x に対して、k 個の supporting memory を探索
(basic model では k = 2)
は入力とメモリーのスコア関数 (後述)
47. Multi-hop
Joe went to the kitchen. Fred went to the kitchen.
Joe picked up the milk. Joe travelled to the office.
Joe left the milk. Joe went to the bathroom.
Where is the milk now? A: office
新たに出来た出力表現
を次の query として同じ処理を
繰り返す
81. 参考文献
[1] A. Sordoni et al., ‘A Hierarchical Recurrent Encoder-Decoder For Generative
Context-Aware Query Suggestion’., arXiv:1507.02221, 2015.
[2] J. Weston et al., ‘Memory Networks’, arXiv:1410.3916, 2014.
[3] S. Sukhbaatar et al., ‘End-To-End Memory Networks’, arXiv:1503.08895, 2015.
[4] A. Kumar et al., ‘Ask Me Anything: Dynamic Memory Networks for Natural
Language Processing’, arXiv:1506.07285, 2015.
[5] A. Graves et al., ‘Neural Turing Machine’, arXiv:1410.5401, 2015.
[6] Łukasz Kaiser and Ilya Sutskever, ‘Neural GPUs Learn Algorithms’,
arXiv:1511.08228, 2015.
[7] M. Dehghani et al., ‘Universal Transformers’, arXiv:1807.03819, 2018.
[8] Wizard of Wikipedia
[9] Noise Contrastive Estimation
[10] Distributed representation of words and phrases and thier compositionality