文献紹介: Similarity-Based Reconstruction Loss for Meaning Representation

文献紹介: Similarity-Based Reconstruction Loss for Meaning Representation

2019/05/28の文献紹介で発表

F637b583c221c132af26c91cb3dba0ca?s=128

Yumeto Inaoka

May 26, 2019
Tweet

Transcript

  1. Similarity-Based Reconstruction Loss for Meaning Representation

  2. Literature 2

  3. Abstract • • • 3

  4. Introduction • • 4

  5. Related Work • • • • 5

  6. Related Work • • 6

  7. Auto-Encoder •ℒ , • • • • 7

  8. Weighted similarity loss •ℒ = − σ =1 sim ,

    • • • : • • sim() • 8
  9. Weighted cross-entropy loss •ℒ = − σ =1 sim ,

    log( ) • • 9
  10. Soft label loss •ℒ = − σ =1 ∗log •

    ∗ = ൞ sim , σ =1 sim(,) , ∈ top N 0 , ∉ top N • • 10
  11. True-label encoding 11

  12. Tasks & Datasets • • • 12

  13. Results 13

  14. Results 14

  15. Additional Experiments • • 15

  16. Results • • 16

  17. Results 17

  18. Results 18

  19. Results 19

  20. Discussion • • 20

  21. Conclusion • • • • 21