物質計測における機械学習応用と知識発見/Machine learning application and knowledge discovery in materials characterization

第4回 統計・機械学習若手シンポジウムでの講演スライドです。未公開とした部分については後日追加します。
https://sites.google.com/view/statsml-symposium19/

Transcript

1. ෺࣭ܭଌʹ͓͚ΔػցֶशԠ༻ͱ஌ࣝൃݟ
   

   ୈ4ճ౷ܭɾػցֶशएखγϯϙδ΢Ϝ
   ໊ݹ԰޻ۀେֶ
   ླ໦༤ଠ
   ૯߹ݚڀେֶӃେֶߴΤωϧΪʔՃ଎ثݚڀػߏɹ෺࣭ߏ଄Պֶݚڀॴ
   
   
   

   View Slide

2. About me
   • ླ໦༤ଠ ʢYuta SUZUKIʣ https://resnant.github.io
   • Research Interest
   • ࡐྉܭଌʹ޲͚ͨػցֶशͷԠ༻
   • ओʹɺXઢΛ༻͍ͨແػ෺࣭ͷܭଌʢ࣓ੴɺి஑ͳͲʣ
   • ػցֶशΛ༻͍ͨ஌ࣝൃݟ
   • Education
   • ૯߹ݚڀେֶӃେֶɹߴΤωϧΪʔՃ଎ثՊֶݚڀՊ
   • খ໺ݚڀࣨ (PhD, 2019.3 - present)
   • ౦ژཧՊେֶɹجૅ޻ֶݚڀՊ
   • খ࢚ݚڀࣨ (BS & MS in Engineering)
   • Collaborators
   • ೔໺ӳҳ ઌੜʢ౷ܭ਺ཧݚڀॴʣ
   • ڇٱ঵޹ ઌੜʢΦϜϩϯαΠχοΫΤοΫεʣ
   • Funded by JST ACT-I & JSPS DC1

   
   

   View Slide

3. ,&,ߴΤωϧΪʔՃ଎ثݚڀػߏ
   
   • ೔ຊ࠷େͷ෺ཧͷݚڀॴʢ෺ཧతʹ΋େ͖͍ʣɿ500ਓҎ্ͷݚڀऀ
   • Ӊ஦ɺૉཻࢠ͔Βࡐྉɺੜ෺·Ͱ෯޿͍ݚڀ͕͞Ε͍ͯΔ
   • զʑ͸Xઢʢ์ࣹޫʣΛ࢖ͬͯ෺࣭Λௐ΂ΔݚڀʹऔΓ૊ΜͰ͍·͢
   • ์ࣹޫɿ΄΅ޫ଎Ͱӡಈ͢Δిࢠ͔Β์ࣹ͞ΕΔXઢ
   ,&,
   

   View Slide

4. ࣮ݧ૷ஔͳͲ
   
   

   View Slide

5. l.BUFSJBMT*OGPSNBUJDTzʢ#JP $IFNʹଓ͍ͯʣ
   w .BUFSJBMT*OGPSNBUJDT
   w ౷ܭతਪ࿦ʢػցֶशͳͲʣ΍৘ใॲཧٕज़ͷԠ༻ʹΑΓ

   ޮ཰తͳࡐྉ։ൃΛ໨ࢦ֓͢೦
   w ೥ʹถࠃ͕.(*ϓϩδΣΫτΛ։࢝ͯ͠Ҏདྷɺٸܹʹൃల
   
   Perkins, K. et al. The 2019 materials by design roadmap. Journal of Physics D: Applied Physics 52, 013001 (2018).
   

   View Slide

6. .BUFSJBMT*OGPSNBUJDTͷجຊతํ๏࿦
   γϛϡϨʔγϣϯͰσʔληοτ࡞੒
   ػցֶशͰ෺ੑ༧ଌϞσϧΛߏங
   ϋΠεϧʔϓοτ࣮ݧͰݕূ
   ಘͨ஌ࣝΛʹϑΟʔυόοΫ
   
   Ren, F. et al. Science Advances 4, eaaq1566 (2018).
   

   View Slide

7. .BUFSJBMT*OGPSNBUJDTͷجຊతํ๏࿦
   γϛϡϨʔγϣϯͰσʔληοτ࡞੒
   ػցֶशͰ෺ੑ༧ଌϞσϧΛߏங
   ϋΠεϧʔϓοτ࣮ݧͰݕূ
   ಘͨ஌ࣝΛʹϑΟʔυόοΫ
   
   Ren, F. et al. Science Advances 4, eaaq1566 (2018).
   w ࡐྉ߹੒
   w ଌఆ
   w ஌ࣝநग़ʢσʔλղੳʣ
   զʑ͸ɺ

   ࣮ݧʢಛʹଌఆͱղੳʣͷޮ཰ԽΛ௨ͯ͡

   .*ͷݚڀʹऔΓ૊ΜͰ͍·͢ɻ
   

   View Slide

8. .BUFSJBMT*OGPSNBUJDTݚڀಈ޲
   
   • ࠃ಺֎ͷϓϩδΣΫτ
   • Materials Genome Initiative ʢถࠃ 2011ʙʣ
   • NOMADʢEU 2015ʙʣɹ΄͔εΠεɺதࠃɺؖࠃͳͲ
   • MI2Iʢຊࠃ 2015ʙʣ
   • CREST ʰֵ৽ࡐྉ։ൃʱྖҬʢ2017ʙʣ
   • ͖͕͚͞ ʰ৘ใܭଌʱྖҬ ʢ2016ʙʣ
   • δϟʔφϧɺࠃࡍձٞ
   • NeurIPS Workshop (2017, 2018):

   Machine Learning for Molecules and Materials
   • npj Computational Materialsʢ2015ʙ, IF = 9.651ʣˡMIͷઐ໳ࢽ
   • ͦͷଞ
   • Kaggle NOMAD ίϯϖʢಁ໌ಋిମͷ෺ੑ༧ଌίϯςετ, 2018೥ʣ
   • 800νʔϜҎ্͕ࢀՃɺ༏উऀ͸೔ຊਓ
   • N-gramʢࣗવݴޠॲཧ޲͚ಛ௃ྔʣΛ࢖͍ಛ௃ྔΛ࡞ͬͯ༏উ
   • ࠷ۙ͸Խ߹෺ͷ࣓ؾ૬ޓ࡞༻ͷ༧ଌίϯϖ΋
   Sutton, C. et al. Crowd-sourcing materials-science challenges with the NOMAD 2018 Kaggle competition. 

   npj Comput. Mater. 5, 111 (2019).
   

   View Slide

9. 1. σʔλ͕εϞʔϧɾߴ͍औಘίετ
   2. WhyʁΛ໰ΘΕΔέʔε͕ଟ͍
   3. ໰୊ઃఆͷ೉͠͞
   
   ࡐྉՊֶʹ͓͚Δػցֶशͷࠔ೉
   ʢࡐྉܭଌͷཱ৔͔Βʣ
   

   View Slide

10. 1. σʔλ͕εϞʔϧɾߴ͍औಘίετ
    • Φʔϓϯσʔλͷ੔උ͸·ͩಓ൒͹
    • Materials Projectʢถࠃʣ
    • MatNaviʢ෺࣭ࡐྉݚڀػߏʣͳͲऔΓ૊Έ͕ਐߦத
    • EUͰ͸ɺެతݚڀ݁ՌͷΦʔϓϯσʔλԽͷ࿮૊Έ͕੔උத
    • γϛϡϨʔγϣϯͰσʔλΛ४උɺ࣮ݧͱ૊Έ߹Θ͕ͤݱ࣮త
    • ࣮ࡍͷ࣮ݧͱͷΪϟοϓ͕՝୊ͱͳΔ
    • ࿹ͷݟͤॴͰ΋͋Δ
    • NLPΛ༻͍ͨจݙαʔϕΠͰσʔλऩू͢ΔࢼΈ΋͋Δ
    • σʔλͷϞμϦςΟ͕ଟ༷
    • ςʔϒϧɺܥྻɺը૾ͳͲ͕ࠞࡏ
    
    ࡐྉՊֶʹ͓͚Δػցֶशͷࠔ೉
    ʢࡐྉܭଌͷཱ৔͔Βʣ
    

    View Slide

11. 1. σʔλ͕εϞʔϧɾߴ͍औಘίετ
    • ίετײͷࢀߟʢ͋͘·ͰҰྫͰ͕͢…ʣ
    • γϛϡϨʔγϣϯɿ1݅͋ͨΓ਺ඵʙ਺࣌ؒ
    • ࣮ݧɿ1݅͋ͨΓ1೔ʢ߹੒+ଌఆʣ
    • ࢼྉ1ݸ͋ͨΓ1000ԁʙ10ສԁͷՁ֨ײʢݪྉ͕ߴ͍ʣ
    • ࣗಈԽɾޮ཰Խ΋ਐΈͭͭ͋Δ
    
    ࡐྉՊֶʹ͓͚Δػցֶशͷࠔ೉
    ʢࡐྉܭଌͷཱ৔͔Βʣ
    ݸͰ૊੒ ݸͰ਺ඦ૊੒
    w ૊੒܏ࣼബບ
    w ຕͰ૊੒Λ໢ཏͰ͖Δ
    

    View Slide

12. 1. σʔλ͕εϞʔϧɾߴ͍औಘίετ
    2. WhyʁΛ໰ΘΕΔέʔε͕ଟ͍
    3. ໰୊ઃఆͷ೉͠͞
    
    ࡐྉՊֶʹ͓͚Δػցֶशͷࠔ೉
    ʢࡐྉܭଌͷཱ৔͔Βʣ
    

    View Slide

13. 2. WhyʁΛ໰ΘΕΔέʔε͕ଟ͍
    • ࡐྉՊֶऀ͸ɺ୯ʹԿ͔Λ༧ଌ͢Δ͚ͩͰ͸ͳ͘ɺ

    ͦͷ༧ଌͷഎޙʹ͋Δ࢓૊Έʢ=෺ཧʣΛ஌Γ͍ͨ
    • ʮۚͷཛΛੜΉܲͦͷ΋ͷΑΓɺͦͷܲͷ࢓૊Έʹڵຯ͕͋Δʯ
    • ࠷ۙͷ࿦จͰ͸ɺ

    MLʹΑΔ༧ଌ + ϞσϧͷղੳʹΑΔ஌ࣝൃݟ ͷߏ੒͕ଟ͍

    • ༧ଌ݁ՌΛແ৚݅ʹड͚ೖΕΒΕΔ΄ͲMLʹ਌͠ΜͰ͍ͳ͍
    • ೲಘײ͕ͳ͍ͱ࢖͍ͮΒ͍
    
    ࡐྉՊֶʹ͓͚Δػցֶशͷࠔ೉
    ʢࡐྉܭଌͷཱ৔͔Βʣ
    

    View Slide

14. 1. σʔλ͕εϞʔϧɾߴ͍औಘίετ
    2. WhyʁΛ໰ΘΕΔέʔε͕ଟ͍
    3. ໰୊ઃఆͷ೉͠͞
    
    ࡐྉՊֶʹ͓͚Δػցֶशͷࠔ೉
    ʢࡐྉܭଌͷཱ৔͔Βʣ
    

    View Slide

15. 3. ໰୊ઃఆͷ೉͠͞
    • ͦ΋ͦ΋ɺͲΜͳ໰୊͕ػցֶशͰղ͚͏Δ͔ݕ౼͍ͯ͠Δஈ֊
    • ղ͘΂͖໰୊Λݟ͚ͭग़͢εΩϧʢʁʣ͕͔ͳΓॏཁ
    • ΞΠσΟΞॏࢹͷ࿦จ͕ଟ͍

    • ՝୊͕֤࿦తͰɺීวతͳλεΫ͕গͳ͍
    • ςΫχοΫͷڞ༗΍ɺڝ૪͕೉͍͠
    • ը૾ೝࣝίϯϖͳͲͱ͸ঢ়گ͕ҟͳ͍ͬͯΔ
    • ੒ޭͷࢦඪʢείΞ΍KPIʣ͕͸͖ͬΓ͠ͳ͍
    • ఏҊख๏͸ɺ࣮ੈքͰ࣮ݧͯ͠͏·͍͔͘͘ௐ΂Δ͔͠ͳ͍
    • ࣮ݧ͕େมͳܥͩͱ͔ͳΓͭΒ͍
    • δϨϯϚɿେมͳ࣮ݧ΄Ͳޮ཰Խͷҙຯ͕͋Δ
    
    ࡐྉՊֶʹ͓͚Δػցֶशͷࠔ೉
    ʢࡐྉܭଌͷཱ৔͔Βʣ
    

    View Slide

16. ࡐྉଌఆσʔλͷྫ
    
    9ઢճંύλʔϯ

    ݁থߏ଄ɺ૊੒
    9ઢٵऩεϖΫτϧ
    ిࢠঢ়ଶɺہॴߏ଄
    9ઢܬޫεϖΫτϧ
    ૊੒
    
    

    View Slide

17. ࡐྉଌఆσʔλͷྫ
    w ͍ͣΕ΋ɺੜσʔλΛղੳͯ͠ಘΒΕΔ෺ཧྔ͕ॏཁ
    
    9ઢճંύλʔϯ

    ݁থߏ଄ɺ૊੒
    9ઢٵऩεϖΫτϧ
    ిࢠঢ়ଶɺہॴߏ଄
    9ઢܬޫεϖΫτϧ
    ૊੒
    • ੜσʔλ͸ෳࡶɺݟͯ΋Α͘Θ͔Βͳ͍
    • σʔλղੳʹ͸ϞσϧϑΟοςΟϯάΛߦ͏͜ͱ͕ଟ͍
    • ࢼߦࡨޡ͕ඞཁɺ͕͔͔࣌ؒΔʢʙ1೔/dataʣ
    

    View Slide

18. 
    X-ray
    sample
    Detectors
    ͍͋ͪγϯΫϩτϩϯ BL5A (powder XRD beamline)
    σʔλͷ௡೾
    ݱࡏͷݚڀࢪઃͰ͸ຖ೔100GBҎ্ͷσʔλ͕ੜΈग़͞ΕΔ
    • ྫɿ 5000ຊͷXRD͕24࣌ؒͰऔಘ͞ΕΔ (AichiSR)
    ͜ΕΒΛखಈͰղੳͰ͖ΔͩΖ͏͔ʁ
    

    View Slide

19. • Research purpose
    • ػցֶशΛ༻͍ͯɺߴ଎͔ͭࣗಈͷσʔλղੳٕज़Λ։ൃ͢Δ
    • έʔεελσΟ:
    1. Xઢٵऩ෼ޫ๏ (XAS: X-ray Absorption Spectroscopy)
    • σʔλͷੑ࣭Λௐ΂ͨݚڀ

    ྨࣅ౓Λௐ΂Δ͚ͩͰ෺ཧྔ͕Θ͔ͬͨ
    2. Xઢճં๏ (XRD: X-ray Diffraction)
    • ڭࢣ͋ΓֶशͷԠ༻ɺܾఆ໦Λؤுͬͯௐ΂ͯ஌ࣝൃݟ
    Outline
    

    View Slide

20. • Research purpose
    • ػցֶशΛ༻͍ͯɺߴ଎͔ͭࣗಈͷσʔλղੳٕज़Λ։ൃ͢Δ
    • έʔεελσΟ:
    1. Xઢٵऩ෼ޫ๏ (XAS: X-ray Absorption Spectroscopy)
    • σʔλͷੑ࣭Λௐ΂ͨݚڀ

    ྨࣅ౓Λௐ΂Δ͚ͩͰ෺ཧྔ͕Θ͔ͬͨ
    2. Xઢճં๏ (XRD: X-ray Diffraction)
    • ڭࢣ͋ΓֶशͷԠ༻ɺܾఆ໦Λؤுͬͯௐ΂ͯ஌ࣝൃݟ
    Outline
    

    View Slide

21. σʔλͷྨࣅ౓ʹண໨ͨ͠

    9ઢٵऩεϖΫτϧ͔Βͷ෺ཧྔਪఆ
    έʔεελσΟ
    : 4V[VLJ et alMicroscopy and Microanalysis
    
    : 4V[VLJ et alnpj Computational Materials,
    
    

    View Slide

22. Photon Energy (eV)
    w ෺࣭ʹ9ઢΛೖࣹ͠ɺͦͷٵऩεϖΫτϧΛௐ΂Δجຊతͳख๏
    w ࡐྉͷ੒෼΍Խֶ݁߹ɺ࣓ؾঢ়ଶͳͲ͕Θ͔Δ
    9ઢٵऩ෼ޫ๏
    ,&,1IPUPO'BDUPSZ#-
    
    

    View Slide

23. 2ͲΕʹࣅ͍ͯ·͔͢ʁ
    9ઢٵऩ෼ޫ๏
    ࢎԽϚϯΨϯʢ࣮ଌʣ
    γϛϡϨʔγϣϯ
    γϛϡϨʔγϣϯ γϛϡϨʔγϣϯ
    Mn 2+,

    10Dq = 0.9 eV
    Mn 3+,

    10Dq = 0.9 eV
    Mn 2+,

    10Dq = 1.3 eV
    
    

    View Slide

24. w ෺࣭ʹ9ઢΛೖࣹ͠ɺͦͷٵऩεϖΫτϧΛௐ΂Δجຊతͳख๏
    w ࡐྉͷ੒෼΍Խֶ݁߹ɺ࣓ؾঢ়ଶͳͲ͕Θ͔Δ
    w యܕతʹ͸ɺଌఆͨ͠εϖΫτϧΛจݙ΍ܭࢉ౳ͱݟൺ΂ɺ

    ࣅͨ΋ͷΛ୳͢͜ͱͰ෺ཧྔΛਪఆʢσʔλղੳʣ͍ͯ͠Δ
    w ਓྗͷ࠷ۙ๣๏
    w ໰୊఺
    w ଌఆ͸ສ݅Iɺ

    ͔݅͠͠ͷσʔλղੳʹ਺࣌ؒʙ਺೔Ҏ্
    w ղੳऀͷओ؍͕আ͚ͳ͍
    9ઢٵऩ෼ޫ๏
    EF(SPPU 'FUBM 1IZT3FW# r
    
    EF(SPPU ',PUBOJ "$PSF-FWFM4QFDUSPTDPQZPG4PMJET $3$1SFTT
    
    
    

    View Slide

25. • ຊ࣭తʹ͸ɺεϖΫτϧͷྨࣅ౓ΛଌΔई౓͕ͳ͍͜ͱ͕໰୊
    • զʑͷΞΠσΟΞ:

    దͨ͠ڑ཭ई౓Λ࢖͑͹εϖΫτϧͷྨࣅ౓ΛධՁͰ͖ͦ͏
    • σʔλͷڑ཭ΛଌΔͱ͍͏ൃ૝͸ɺ

    ࡐྉՊֶऀʹͱͯ͠͸ͳ͔ͳ͔ग़ͯ͜ͳ͔ͬͨ
    • ଟ༷ମԾઆ͔Βண૝
    • ԾఆΛಋೖ
    • ʮࣅͨੑ࣭ͷ෺࣭ͷεϖΫτϧ͸ࣅ͍ͯΔ͸ͣʯ
    • దͨ͠ڑ཭ई౓ͱ͸ʁ
    • Euclidian distance, cosine, JS divergence…
    • ޙͰٞ࿦͠·͢
    Key concepts
    

    View Slide

26. 4XJTT3PMM
    ࣍ݩʹຒΊࠐ·Εͨ

    ࣍ݩଟ༷ମ
    w ଟ༷ମԾઆ
    • “many data sets have the property that the data points all lie close to a manifold of
    much lower dimensionality than that of the original data space” (Bishop 2006)
    w ʰߴ࣍ݩͳσʔλͰ͋ͬͯ΋ɺ

    ɹ࣮ࡍʹ͸ΑΓ௿࣍ݩͳ෦෼ू߹ʹ෼෍͍ͯ͠Δʱ
    w ʮσʔλͷຊ࣭తͳ࣍ݩʢ࣮ޮ࣍ݩʣ͸௿͍

    ʢ͜ͱ͕ଟ͍ʣʯ

    w ༷ʑͳ࣍ݩ࡟ݮΞϧΰϦζϜ͕ߟҊ
    σʔλͷ࣮ޮ࣍ݩ
    *40."1 U4/&
    .%4
    
    

    View Slide

27. 4XJTT3PMM
    ࣍ݩʹຒΊࠐ·Εͨ

    ࣍ݩଟ༷ମ
    σʔλͷ࣮ޮ࣍ݩ
    w զʑͷΞΠσΟΞ
    w ܭଌσʔλͷ࣮ޮ࣍ݩ΋ɺ

    ෺ཧతύϥϝʔλఔ౓ʢ਺ݸʣͷ࣍ݩ͔͠ͳ͍ͩΖ͏
    w ܭଌͱσʔλղੳͱ͸͢ͳΘͪɺ

    ߴ࣍ݩσʔλ͔Β௿࣍ݩͷಛ௃ʢ෺ཧྔʣΛऔΓग़͢͜ͱɻ
    w ଟ༷ମֶशΛ༻͍ͨσʔλͷ࣍ݩ࡟ݮΛண૝
    w ଟ༷ମԾઆ
    • “many data sets have the property that the data points all lie close to a manifold of
    much lower dimensionality than that of the original data space” (Bishop 2006)
    w ʰߴ࣍ݩͳσʔλͰ͋ͬͯ΋ɺ

    ɹ࣮ࡍʹ͸ΑΓ௿࣍ݩͳ෦෼ू߹ʹ෼෍͍ͯ͠Δʱ
    w ʮσʔλͷຊ࣭తͳ࣍ݩʢ࣮ޮ࣍ݩʣ͸௿͍

    ʢ͜ͱ͕ଟ͍ʣʯ
    
    

    View Slide

28. w ଟॏ߲͓Αͼ഑Ґࢠ৔ཧ࿦ʹج͖ͮ

    .OQ9ઢٵऩεϖΫτϧΛγϛϡϨʔγϣϯ
    w Ձ਺ɿ
    w ݁থ৔ύϥϝʔλ%RɿʙF7
    w ݁থ৔ͷରশੑ͸໘ମ഑Ґʹݻఆ
    w ߹ܭຊͷ9"4εϖΫτϧΛ४උ
    w εϖΫτϧຊ͸఺ͷσʔλϙΠϯτ
    w ࣍ݩ࡟ݮʹ͸ଟ࣍ݩई౓ߏ੒๏ʢ.%4ʣΛ༻͍ͨ
    w σʔλಉ࢜ͷڑ཭ΛͳΔ΂͘อͬͨ··ɺ

    ΑΓ௿࣍ݩͳۭؒʹ͓͍ͯσʔλΛදݱ͢Δ
    w ࣍ݩˠ࣍ݩʹ࣍ݩ࡟ݮ
    σʔλͷ४උ
    Photon Energy (eV)
    10
    Dq
    (eV)
    Intensity
    ܭࢉͨ͠.Oͷ9"4
    EF(SPPU ',PUBOJ "$PSF-FWFM4QFDUSPTDPQZPG4PMJET $3$1SFTT
    
    #PSH *(SPFOFO 1.PEFSONVMUJEJNFOTJPOBMTDBMJOHUIFPSZBOEBQQMJDBUJPOT 4QSJOHFS7FSMBH/FX:PSL
    
    
    

    View Slide

29. ࣮ଌͨ͠.O0ͷεϖΫτϧ

    ʢEF(SPPUFUBMʣ
    ਪఆ͞Εͨ෺ཧྔʹରԠ͢ΔεϖΫτϧ

    ʢܭࢉɺ.O %RF7ʣ
    ैདྷ๏ͱ΋Α͘Ұக
    ݁Ռɿ࣍ݩ࡟ݮͱՄࢹԽ
    w εϖΫτϧಉ࢜ͷྨҎ౓͕

    ෺ཧྔʹରԠͯ͠มԽͨ͠

    ʢಋೖͨ͠Ծఆ͸ଥ౰ͩͬͨʂʣ
    w ྨࣅ౓͔Βਪఆ͞Εͨ෺ཧྔ͸

    ख़࿅ऀʹΑΔಉఆʢैདྷ๏ʣͱҰக
    w σʔλͷྨࣅ౓ʹ΋ͱ͍ͮͯ

    ෺ཧྔΛਪఆͰ͖Δͱࣔͤͨ
    ଟ࣍ݩई౓ߏ੒๏ʢ.%4ʣʹΑΔ࣍ݩ࡟ݮ݁Ռ
    ֤఺͕ຊͷ9"4εϖΫτϧɺ఺ಉ࢜ͷڑ཭͸εϖΫτϧͷ

    ྨҎ౓ʹରԠ͢Δɻ৭͸Ձ਺ɺ਺ࣈ͸%RʢF7ʣYΛࣔ͢
    ʢ͜͜Ͱ͸&VDMJEFBO%JTUBODFΛ༻͍͍ͯΔʣ
    ྨࣅ౓ʹج͍ͮͯɺ

    ࣮ݧσʔλ͕ଥ౰ͳҐஔʹ

    ϓϩοτ͞Ε͍ͯΔʢ੺ؙʣ
    
    

    View Slide

30. • ຊ࣭తʹ͸ɺεϖΫτϧͷྨࣅ౓ΛଌΔई౓͕ͳ͍͜ͱ͕໰୊
    • զʑͷΞΠσΟΞ:

    దͨ͠ڑ཭ई౓Λ࢖͑͹εϖΫτϧͷྨࣅ౓ΛධՁͰ͖ͦ͏
    • ԾఆΛಋೖ
    • ʮࣅͨੑ࣭ͷ෺࣭ͷεϖΫτϧ͸ࣅ͍ͯΔ͸ͣʯ
    • ଟ༷ମԾઆ͔Βண૝
    • దͨ͠ڑ཭ई౓ͱ͸ʁ
    • ඞཁͳੑ࣭:
    • εϖΫτϧͷಛ௃Λଊ͑ΒΕΔ͜ͱ
    • ϐʔΫͷҐஔɺܗ…
    • ϊΠζͷ෇Ճʹ͍ͭͯϩόετͰ͋Δ͜ͱʢ࣮༻্ॏཁʣ
    • ϊΠζͱଌఆ࣌ؒ͸τϨʔυΦϑ
    Key conceptsʢ࠶ܝʣ
    

    View Slide

31. ϊΠζͷ෇Ճʹର͢Δݎ࿚ੑ
    w ϊΠζΛؚΜͩεϖΫτϧΛ࠶ݱ͢ΔͨΊɺ

    ϊΠζΛΨ΢ε෼෍Ͱۙࣅ͠ɺεϖΫτϧʹՃࢉ
    w 4/ൺΛม͑ͯσʔληοτΛ࡞੒
    w 4/ൺϐʔΫ࠷େ஋ϊΠζͷඪ४ภࠩͱఆٛ
    
    9ઢٵऩεϖΫτϧ

    ʢγϛϡϨʔγϣϯʣ
    Ψ΢εϊΠζΛՃࢉ
    
    

    View Slide

32. ϊΠζͷ෇Ճʹର͢Δݎ࿚ੑ
    ϊΠζ෇Ճʹର͢Δڑ཭ؔ਺ͷݎ࿚ੑΛൺֱ
    w ϐΞιϯ૬ؔ܎਺͕ϕετɺDPTJOF΋ݎ࿚
    w ϐʔΫͷڧ౓มԽʹରͯ͠ڑ཭͕ෆมͳͨΊͱਪ࡯͞ΕΔ
    w ϊΠδʔͳσʔλ͔ΒͰ΋෺ཧྔ͕ਪఆͰ͖Ε͹ɺ

    ଌఆ࣌ؒΛେ෯ʹ୹ॖͰ͖Δ
    w 4/ൺ͕ͰΑ͚Ε͹ଌఆ࣌ؒ͸ͰࡁΉ
    ϊΠζେ
    ϦϑΝϨϯεʹର͢Δྨࣅ౓
    ྑ͍
    
    

    View Slide

33. w εϖΫτϧͷྨҎ౓ΛଌΔ͚ͩͰɺͦͷࡐྉͷ෺ཧྔΛਪఆͰ͖ͨ
    w σʔλղੳΛେ෯ʹߴ଎ԽɺࣗಈԽͰ͖Δͱݟࠐ·ΕΔ

    w ద੾ͳྨࣅੑई౓͕બఆ͞ΕͨʢϐΞιϯ૬ؔ܎਺͕ϕετʣ
    w ࡐྉσʔλ΁ͷػցֶशͷԠ༻ͷجૅͱͳΔ
    w ϊΠζ΍෼ղೳͷѱԽʹݎ࿚ͳ෺ཧྔਪఆ͕Մೳ
    w ෆ҆ఆ෺࣭΍ߴ଎ݱ৅ͷଌఆͳͲɺ৽͍͠෺ཧ͕୓͚ΔՄೳੑ
    લ൒ͷ·ͱΊ
    

    View Slide

34. • Research purpose
    • ػցֶशΛ༻͍ͯɺߴ଎͔ͭࣗಈͷσʔλղੳٕज़Λ։ൃ͢Δ
    • έʔεελσΟ:
    1. Xઢٵऩ෼ޫ๏ (XAS: X-ray Absorption Spectroscopy)
    • σʔλͷੑ࣭Λௐ΂ͨݚڀ

    ྨࣅ౓Λௐ΂Δ͚ͩͰ෺ཧྔ͕Θ͔ͬͨ
    2. Xઢճં๏ (XRD: X-ray Diffraction)
    • ڭࢣ͋ΓֶशͷԠ༻ɺܾఆ໦Λؤுͬͯௐ΂ͯ஌ࣝൃݟ
    Outline
    

    View Slide

35. ػցֶशΛ༻͍ͨ

    9ઢճંύλʔϯ͔Βͷ݁থߏ଄ਪఆ
    έʔεελσΟ
    : 4V[VLJet alMicroscopy and Microanalysis
    
    

    View Slide

36. ࡐྉ։ൃͷ࢝఺ɿ݁থߏ଄ղੳ
    • ෺࣭ͷػೳ͸݁থߏ଄ʹΑΓࢧ഑͞ΕΔ
    • ిؾ఻ಋੑɺ࣓ੑɺ৮ഔػೳɺFUDʜ
    • ݱ୅ͷࡐྉ։ൃͰ͸ݪࢠεέʔϧͰࡐྉΛ࡞ΓࠐΉ
    ݁থߏ଄ʢʹݪࢠͷฒͼํʣͷಉఆ͕ۃΊͯॏཁ
    • Ϧν΢ϜΠΦϯి஑ͷྫ
    • ిۃࡐྉ-J$P0
    ʹଞݩૉΛఴՃͯ݁͠থߏ଄Λ҆ఆԽ͠ɺ

    ༰ྔྼԽΛ཈͑ͨిۃࡐྉ%-$0͕։ൃ
    Liu, Q. et al. Nature Energy (2018).
    %-$0ͷ9ઢճંύλʔϯ
    9ઢճં๏ʹΑΔ݁থߏ଄ղੳ͸ࡐྉ։ൃͷ࠷ॏཁٕज़ͷҰͭ
    %-$0ͷ݁থߏ଄
    "M
    -B
    -J
    
    

    View Slide

37. ࡐྉ։ൃͷ࢝఺ɿ݁থߏ଄ղੳ
    • ෺࣭ͷػೳ͸݁থߏ଄ʹΑΓࢧ഑͞ΕΔ
    • ిؾ఻ಋੑɺ࣓ੑɺ৮ഔػೳɺFUDʜ
    • ݱ୅ͷࡐྉ։ൃͰ͸ݪࢠεέʔϧͰࡐྉΛ࡞ΓࠐΉ
    ݁থߏ଄ʢʹݪࢠͷฒͼํʣͷಉఆ͕ۃΊͯॏཁ
    • Ϧν΢ϜΠΦϯి஑ͷྫ
    • ిۃࡐྉ-J$P0
    ʹଞݩૉΛఴՃͯ݁͠থߏ଄Λ҆ఆԽ͠ɺ

    ༰ྔྼԽΛ཈͑ͨిۃࡐྉ%-$0͕։ൃ
    Liu, Q. et al. Nature Energy (2018).
    %-$0ͷ9ઢճંύλʔϯ
    9ઢճં๏ʹΑΔ݁থߏ଄ղੳ͸ࡐྉ։ൃͷ࠷ॏཁٕज़ͷҰͭ
    %-$0ͷ݁থߏ଄
    "M
    -B
    -J
    
    ੈքॳͷϦν΢ϜΠΦϯೋ࣍ి஑ͷిۃ

    ʢࠓ೥ͷϊʔϕϧԽֶ৆ʹʂʣ
    

    View Slide

38. 9ઢճંʢ93%ʣͷݪཧ
    • 9ઢճંʢ93%ʣ
    • ೾ͷڧΊ͍͋ɺऑΊ͍͋Λར༻ͯ݁͠থߏ଄Λௐ΂Δ
    • ϐʔΫͷҐஔ΍ߴ͞ͰɺݪࢠͷҐஔ΍छྨ͕Θ͔Δ
    • ϒϥοά৚݅ETJOВЕ
    
    Xઢ ݕग़ث
    ֯౓Λม͑ͯଌఆ
    ճં֯
    ճંڧ౓
    Rigaku Corp.

    https://www.rigaku.com/en/techniques/xrd
    

    View Slide

39. ݁থߏ଄ղੳͷྲྀΕ
    • 9ઢճંʢ93%ʣ͸݁থߏ଄ղੳͷجຊख๏ͱͯ͠ීٴ
    • 93%ύλʔϯ͔Β݁থߏ଄͕Θ͔Δʢ೥ϊʔϕϧ৆ʣ
    • ෺࣭ݚڀɾ࢈ۀΛࢧ͑ΔෆՄܽͳπʔϧͷҰͭ
    • %/"ͷॏཐટߏ଄ͷൃݟʹ΋ߩݙʢ೥ϊʔϕϧ৆ʣ
    93%ଌఆ૷ஔ /E࣓ੴͷ93%ύλʔϯ /E࣓ੴͷ݁থߏ଄
    ωΦδϜʢ/Eʣ࣓ੴ

    ʢ/E
    'F
    #ʣ
    93%ύλʔϯͷ௚ײతཧղ͸೉͘͠ɺ

    ݁থߏ଄ͷਪఆ͸ઐ໳Ոʹͱͬͯ΋༰қͰ͸ͳ͍
    ˠޮ཰తͳσʔλղੳ͕՝୊ʹ
    
    ϐ
    ʔΫϑΟοςΟϯά
    * Chen, P. A. et al. J. Magn. Mag. Mater. 370, 45–53 (2014).
    

    View Slide

40. 93%σʔλղੳͷ՝୊
    • ଌఆσʔλͷղੳʹཁ͢Δ࣌ؒͱ࿑ྗ͕

    ࡐྉݚڀͷϘτϧωοΫͱͳ͍ͬͯΔ
    • 93%ଌఆɿ݅EBZ
    • 93%ղੳɿ ݅EBZɹˠ͔͠΋σʔλͷେ൒͕ະ׆༻
    • ࣗಈंେख๭ࣾ͸ਓւઓज़ͰରԠ
    • 93%ղੳͷ೉͠͞
    • શݪࢠͷ࣍ݩߏ଄͕ͭͷύλʔϯʹॏͳΓ߹͍ͬͯΔ
    • ղੳతʹ݁থߏ଄ΛٻΊΔ͜ͱ͸ෆՄೳʢෳૉ਺ͷҐ૬ΛଌఆͰ͖ͳ͍ͨΊʣ
    
    • طଘख๏ʢύϥ
    ϝʔλϑ
    Οος
    Οϯάʣ
    ͸ॳظ஋ܾఆ͕ඇৗʹ೉͍͠
    
    ݁থߏ଄ 93%ύλʔϯ
    

    View Slide

41. Overview
    1SFQBSJOH93%%BUBCBTF
    #VJMEJOH1SFEJDUJPO.PEFM ,OPXMFEHF%JTDPWFSZ
    &YQFSJNFOUBM7BMJEBUJPO
    
    

    View Slide

42. Overview
    1SFQBSJOH93%%BUBCBTF
    #VJMEJOH1SFEJDUJPO.PEFM ,OPXMFEHF%JTDPWFSZ
    &YQFSJNFOUBM7BMJEBUJPO
    
    ͜ͷ෦෼ͷ࿩Λ͠·͢
    

    View Slide

43. 0WFSWJFX
    1SFQBSJOH93%%BUBCBTF
    #VJMEJOH1SFEJDUJPO.PEFM ,OPXMFEHF%JTDPWFSZ
    &YQFSJNFOUBM7BMJEBUJPO
    
    

    View Slide

44. ݁থߏ଄༧ଌϞσϧߏங
    • ڭࢣ͋ΓֶशͷλεΫͱͯ͠ఆٛʢܗͱେ͖͞Λ༧ଌʣ
    
    ܗΛ༧ଌ େ͖͞Λ༧ଌ
    X-ray Diffraction Machine Learning Pipeline
    Crystal System Prediction
    (Classification)
    Lattice Parameter Prediction
    (Regression)
    1 ms
    Error < 0.8%
    Cubic,
    ɾ݁থܥʢΫϥεʣ

    ɾۭؒ܈ʢΫϥεʣ
    

    View Slide

45. ݁থߏ଄༧ଌϞσϧߏங
    • ڭࢣ͋ΓֶशͷλεΫͱͯ͠ఆٛʢܗͱେ͖͞Λ༧ଌʣ
    
    w࣮ݧʹద༻͢ΔͨΊɺ੍໿͋Γ
    w࣮ݧ৚͕݅ຖճมΘΔ
    w σʔλ௕ɺεΩϟϯ֯౓ൣғɺ೾௕ͳͲ
    w࣮ݧ৚݅ʹԠͯ͡ߴ଎ʹ࠶܇࿅͍ͨ͠
    wࡶͳϋΠύϥͰ΋ͦͦ͜͜ಈ͍ͯ΄͍͠
    wΞϧΰϦζϜʹ͸&YUSB5SFFTΛ༻͍ͨʢ3BOEPN'PSFTUͷѥछʣ
    w 3BOEPN'PSFTUΑΓ܇࿅͕਺ഒ଎͍ɺνϡʔχϯά΋؆୯
    w ಛ௃ྔॏཁ౓ɺOEҎ߱ͷ༧ଌ݁Ռ΋ಘΒΕΔ
    wղੳ͠΍͍͢γϯϓϧͳಛ௃ྔΛબ୒ʢ࣍ݩʣ
    w ϐʔΫҐஔʢ௿֯ଆ͔Βຊʣɺ߹ܭϐʔΫຊ਺
    X-ray Diffraction Machine Learning Pipeline
    Crystal System Prediction
    (Classification)
    Lattice Parameter Prediction
    (Regression)
    1 ms
    Error < 0.8%
    Cubic,
    

    View Slide

46. ݁Ռɿ݁থߏ଄ͷ༧ଌ݁Ռ
    
    ݁থܥͷࠞಉߦྻʢ7Ϋϥεʣ
    HgS
    a = 5.8717 Å
    CsCl
    a = 4.1150 Å
    LiMn
    2
    O
    4
    a = 8.245 Å
    LiNi
    0.5
    Mn
    2
    Ti
    1.2
    O
    4
    a = 8.324 Å
    Hydrokenomicrolite
    a = 10.454 Å
    ֨ࢠఆ਺ͷ༧ଌ݁Ռ
    R2=0.9892
    • ඇৗʹߴ͍ੑೳΛୡ੒ͨ͠ʢ͜Ε͸γϛϡϨʔγϣϯσʔλͰͷ݁Ռʣ
    • ෼ྨਫ਼౓ "DD
    ͸݁থܥɺۭؒ܈ɺ֨ࢠఆ਺ͷ3.4&
    • εΫϦʔχϯάʹ͸े෼ͳਫ਼౓ͱߟ͑Δ
    

    View Slide

47. 0WFSWJFX
    1SFQBSJOH93%%BUBCBTF
    #VJMEJOH1SFEJDUJPO.PEFM ,OPXMFEHF%JTDPWFSZ
    &YQFSJNFOUBM7BMJEBUJPO
    
    

    View Slide

48. ػցֶशϞσϧ͔Βͷ஌ࣝൃݟ
    • جຊతํ਑
    • ݹయతΞϧΰϦζϜͰటष͘ؤுΔ
    • ख๏తʹ৽͍͠࿩͸ग़͖ͯ·ͤΜʢ͢Έ·ͤΜʣ
    • ࣮໰୊ͷྫͱͯ͠ࢀߟʹͳΕ͹޾͍Ͱ͢
    

    View Slide

49. ݁ՌɿػցֶशϞσϧ͔Βͷ஌ࣝൃݟ
    • ண໨͢ΔϐʔΫຊ਺ʢಛ௃ྔͷ਺ʣͱɺ༧ଌਫ਼౓Λݕূ
    • ඞཁे෼ͳܭଌ৚݅ʢεΩϟϯ֯౓ʣͷܾఆ͕Ͱ͖Δ
    • ಘΒΕͨ஌ݟ
    • ʮ݁থܥ෼ྨͷͨΊʹண໨͢΂͖ϐʔΫ͸ࠨଆ͔Βຊఔ౓ͰΑ͍ʯ
    • ೥Ҏ্࢖ΘΕ͍ͯΔܦݧଇʢϋφϫϧτ๏˞ʣͱ΋ྨࣅ

    ˞ʮڧ౓ॱʹຊͷϐʔΫʹண໨ͤΑʯ
    ண໨͢ΔϐʔΫຊ਺
    ݁থܥͷ༧ଌਫ਼౓
    Q Q Qʜ
    XRDύλʔϯͷྫ
    

    View Slide

50. ݁ՌɿػցֶशϞσϧ͔Βͷ஌ࣝൃݟ
    • ܇࿅ࡁΈͷ&YUSB5SFF͔Βಛ௃ྔͷॏཁ౓Λௐ΂ͨ
    • ܦݧଇͷ۩ମԽΛૂͬͨ
    • ݁থͷܗͷࣝผʹ͸ϐʔΫຊ਺ͷ৘ใ͕ॏཁɿׂͱࣗ໌
    • ղऍɿʮϐʔΫͷຊ਺͸݁থͷରশੑʹରԠ͢Δʢফ໓ଇʣʯ
    • ௿֯ଆͷϐʔΫҐஔʢQʙQ͋ͨΓʣ΋ॏཁͦ͏
    • ෺ཧతͳղऍ͸ඇࣗ໌
    ֤ಛ௃ͷॏཁ౓
    ಛ௃ྔ
    ߹ܭϐʔΫຊ਺
    QʢϐʔΫҐஔʣ
    Q
    Q
    Q
    Q
    Q
    Q
    Q
    Q
    Q
    Q Q Qʜ
    XRDύλʔϯͷྫ
    

    View Slide

51. ݁ՌɿػցֶशϞσϧ͔Βͷ஌ࣝൃݟ
    • γϯϓϧͳܾఆ໦Λ܇࿅ɾՄࢹԽͯࣝ͠ผنଇΛݕূ
    ޙ೔ެ։༧ఆ
    

    View Slide

52. 0WFSWJFX
    1SFQBSJOH93%%BUBCBTF
    #VJMEJOH1SFEJDUJPO.PEFM ,OPXMFEHF%JTDPWFSZ
    &YQFSJNFOUBM7BMJEBUJPO
    
    

    View Slide

53. ఏҊख๏Λ࣮ݧతʹݕূ
    w ࢼྉɿ70ʢೋࢎԽόφδ΢Ϝʣ
    w ஝೤ࡐྉͱͯ͠஫໨͞ΕΔ࣍ੈ୅ࡐྉʮ෺Λˆʹอͭණʯ
    w ߏ଄૬సҠʹ൐͏જ೤Λར༻
    w ಉ࣌ʹۚଐઈԑମసҠ͢ΔͨΊɺεΠονϯάૉࢠʹ΋ظ଴
    w ిࢠͷεϐϯͱيಓͷடংԽʹΑΓ෺ੑ͕ൃݱ
    w ٯʹݴ͑͹ɺ݁থߏ଄Λௐ΂Ε͹ిࢠঢ়ଶ͕Θ͔Δ
    67℃
    ௿Թ૬ɿMonoclinicߏ଄

    ʢઈԑମʙ൒ಋମʣ
    ߴԹ૬ɿTetragonalߏ଄

    

    View Slide

54. 
    • ͍͋ͪγϯΫϩτϩϯʹͯ70ͷ93%ଌఆΛ࣮ࢪʢ#-4ʣ
    • ΄΅ޫ଎Ͱӡಈ͢Δిࢠ͔Β์ࣹ͞ΕΔେڧ౓9ઢΛར༻
    XઢϏʔϜ ࢼྉ
    XઢΧϝϥ
    ์ࣹޫ࣮ݧͰݕূ༻σʔλΛऔಘ
    

    View Slide

55. 
    • Թ౓มԽʹ൐͏70ͷߏ଄૬సҠ͕໌ྎʹ؍࡯͞Εͨ
    • 93%ຊ͸ඵͰࣗಈଌఆՄೳʢ࣌ؒͰຊʣ
    खಈͰͷղੳ݁ՌʢHSPVOEUSVUIʣ
    70ͷ93%ύλʔϯʢ࣮ଌʣ
    ௿Թ૬ʢ31℃ʣ
    ௿Թ૬ʢ62℃ʣ
    ߴԹ૬ʢ64℃ʣ
    ૬సҠ
    

    View Slide

56. 
    ݁থܥ༧ଌ݁Ռ
    ۭؒ܈༧ଌ݁Ռʢ5PQʣ
    139 [I4/mmm]
    194 [P63/mmc]
    198 [P213]
    140 [I4/mcm]
    136 [P42/mnm]
    Tetragonal : P42/mnm
    • ࣌ؒʢਓྗʣˠNTʢఏҊख๏ʣ
    • ݁থܥ͸ਖ਼ղʢΫϥεʣ
    • ۭؒ܈΋501Ͱਖ਼ղʢΫϥεʣ
    70ʢߴԹ૬ʣͷ93%ύλʔϯʢ࣮ଌʣ
    ࣮σʔλΛ࢖ͬͯఏҊख๏Λݕূ
    • ࣮σʔλͰ΋࢖͑Δ͜ͱΛ֬ೝ
    

    View Slide

57. ޙ൒ͷ·ͱΊ
    • ػցֶशΛ༻͍9ઢճંύλʔϯ͔Β݁থߏ଄Λ༧ଌͨ͠
    • εΫϦʔχϯάʹ͸े෼ͳਫ਼౓Λಘͨ
    • ࣮σʔλͰ΋͏·͘ಈ͍ͨ
    • ஌ࣝൃݟ
    • ΍ͬͨ͜ͱ
    • &YUSB5SFFTͷಛ௃ྔॏཁ౓ͷղऍ
    • ܾఆ໦ʹΑΔۙࣅͱՄࢹԽɺղऍ
    • ख़࿅ऀͷצͷʢͷҰ෦ʣ۩ମԽ͕Ͱ͖ͨ
    • ܭଌޮ཰ԽͷͨΊͷࢦ਑Λಘͨ

    • ݸਓతࡶײ
    • ܾఆ໦ϕʔεͷख๏͕ࢥͬͨΑΓڧྗͰڻ͍ͨ
    

    View Slide

58. ऴΘΓʹɿ.BUFSJBMT*OGPSNBUJDTݚڀಈ޲
    
    • ࠓޙͷ.*ͷํ޲ੑʹ͍ͭͯ

    ͭͷϙΠϯτΛࢦఠ
    1. Share meaningful data
    • σʔλ͸ܗࣜΛඪ४Խ͠ެ։͢΂͖
    
    2. Spur collaboration with competitions
    • େձͰڝ͑͹ٕज़ൃల͕Ճ଎͢Δ

    ʢKaggle౳ʣ
    
    3. Develop a shared language
    • ࡐྉͱػցֶशͷݚڀऀͷަྲྀΛ

    ଅਐ͢΂͖
    
    4. Accelerate and automate
    Luna et al., Nature 552, 23-27 (2017)
    

    View Slide

59. ऴΘΓʹɿ.BUFSJBMT*OGPSNBUJDTݚڀಈ޲
    
    • ࠓޙͷ.*ͷํ޲ੑʹ͍ͭͯ

    ͭͷϙΠϯτΛࢦఠ
    1. Share meaningful data
    • σʔλ͸ܗࣜΛඪ४Խ͠ެ։͢΂͖
    
    2. Spur collaboration with competitions
    • େձͰڝ͑͹ٕज़ൃల͕Ճ଎͢Δ

    ʢKaggle౳ʣ
    
    3. Develop a shared language
    • ࡐྉͱػցֶशͷݚڀऀͷަྲྀΛ

    ଅਐ͢΂͖
    
    4. Accelerate and automate
    Luna et al., Nature 552, 23-27 (2017)
    ݸਓతͳࡶײɿ
    • ๖ժظͳ͕ΒɺϓϨΠϠʔͷ૿Ճ͕ஶ͍͠
    
    • ౎߹ͷ͍͍໰୊Λղ͚ͩ͘ʹݶΒͣɺ

    ࣮໰୊΁ͷద༻ΛਐΊΔͷ͕࣍ͷεςοϓ
    
    • ͥͻίϥϘ͍ͤͯͩ͘͞͞ʂ
    

    View Slide