Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Network Neuroscience入門

Kohei Ichikawa
February 22, 2019

Network Neuroscience入門

2/22の計算論的神経科学勉強会のスライドです。

Kohei Ichikawa

February 22, 2019
Tweet

More Decks by Kohei Ichikawa

Other Decks in Science

Transcript

  1. ωοτϫʔΫՊֶೖ໳ جૅ֓೦ w ωοτϫʔΫʢάϥϑʣͱ͸ɿ
 ϊʔυʢ௖఺ʣͱͦͷؒͷ࿈݁ؔ܎Λද͢ΤοδʢลʣͰߏ੒͞ΕΔσʔλ ܕ͕άϥϑͰɺϊʔυ΍Τοδʹ෺ཧతҙຯΛ࣋ͨͤͨ΋ͷ͕ωοτϫʔΫ w άϥϑ͸େ͖͘෼ྨ͢Δͱ༗޲ແ޲άϥϑɺόΠφϦॏΈ෇͖άϥϑͷ ͭʹ෼͚ΒΕΔɻ w

    ೴ͷωοτϫʔΫΛߟ͑Δͱ͖͸෺ཧత઀ଓʹΑͬͯߏ੒͞ΕΔ BOBUPNJDBMOFUXPSLͱμΠφϛΫεͷ૬͔ؔΒܾ·ΔGVODUJPOBMOFUXPSL ͕͋ΓɺOFUXPSLOFVSPTDJFODFͰ͸ͲͪΒ΋ѻ͏ɻ w άϥϑͷ౳Ձͳදݱͱͯ͠ྡ઀ߦྻ͕͋Δɻ ༗޲όΠφϦάϥϑͷྫ
  2. ωοτϫʔΫՊֶೖ໳ w DMVTUFSJOHDPF⒏DJFOU w ͋ΔϊʔυJΛߟ͑ͨ࣌ɺͦΕʹ઀ଓ͍ͯ͠Δϊʔυؒͷ઀ଓ͕औΓಘ Δ઀ଓͷ͏ͪͲͷఔ౓઎Ί͍ͯΔ͔Λද͢ྔ w ωοτϫʔΫʹ͍ͭͯߟ͑Δͱ͖͸શϊʔυͷฏۉΛͱΔ C =

    1 n ∑ i∈N Ci = 1 n ∑ i∈N 2ti ki (ki − 1) ti = 1 2 ∑ j,h∈N aij aih ajh ࠨਤͷྫͩͱɺCʹ઀ଓ͍ͯ͠ΔϊʔυؒͷՄೳͳ઀ଓ͸ ͋Γɺͦͷ͏࣮ͪࡍʹଘࡏ͢ΔΤοδ͸ຊͷͨΊɺ $ʢ༗޲άϥϑͷͨΊ্ͷఆٛͱ͸ҟͳΔ͜ͱʹ஫ҙʣ
  3. ωοτϫʔΫՊֶೖ໳ w QBUIMFOHUI w TIPSUFTUQBUIMFOHUI
 ϊʔυJͱϊʔυKΛ݁Ϳ࠷୹ܦ࿏ɻJ KؒʹΤοδ͕͋Ε͹ͰɺJͱK͕ ܨ͕͍ͬͯͳ͍৔߹͸㱣ʹͳΔɻ dij =

    ∑ auv ∈gi↔j auv ͸JͱKΛ݁Ϳ࠷୹ܦ࿏ʣ w DIBSBDUFSJTUJDQBUIMFOHUI
 TIPSUFTUQBUIMFOHUIΛωοτϫʔΫશମʹ౉ͬͯฏۉͨ͠΋ͷɻ L = 1 n ∑ i∈N Li = 1 n ∑ i∈N ∑ j∈N,j≠i dij n − 1 gi↔j
  4. ωοτϫʔΫՊֶೖ໳ w NPEVMBSJUZ w Ϟδϡʔϧͱશࣸ૾
 ɹɹɹɹɹɹʹΑͬͯϞδϡʔϧ෼ղ͸Ұҙʹఆ·Δɻ w ͜ͷ෼ղʹରͯ͠NPEVMBSJUZ2͸ M ∈

    {1,2,...,m}(m < n) f : N → M Q = ∑ u∈M [ euu − (∑ v∈M euv) 2 ] w ͜͜Ͱɺ͸ϞδϡʔϧVʹؚ·ΕΔϊʔυͷΤοδͷ͏ͪϞδϡʔϧ Wʹؚ·ΕΔϊʔυʹ઀ଓ͍ͯ͠ΔΤοδͷׂ߹Λද͢ɻ euv w BSHNBY2ͷܭࢉ͸/1IBSEͷͨΊɺ༷ʑͳώϡʔϦεςΟΫε͕ఏҊ ͞Ε͍ͯΔɻ
  5. ωοτϫʔΫՊֶೖ໳ w DFOUSBMJUZ w ೔ຊޠͰ͸த৺ੑई౓ɻωοτϫʔΫ಺Ͱݸʑͷϊʔυ͕࣋ͭॏཁੑ Λද͢ྔɻ w DMPTFOFTTDFOUSBMJUZଞͷ఺ͱڑ཭͕͍ۙ΄Ͳத৺ੑ͕ߴ͍ͱ͢Δɻ QBUIMFOHUIͷٯ਺ɻ L−1

    i = n − 1 ∑ j∈N,j≠i dij w CFUXFFOOFTTDFOUSBMJUZͦͷ఺Λ௨Δܦ࿏͕ଟ͍΄Ͳɺத৺ੑ͕ߴ ͍ͱ͢Δɻ bi = 1 (n − 1)(n − 2) ∑ h,j∈N h≠j,h≠i,j≠i ρhj (i) ρhj ʢ͜͜Ͱɺ͸I͔ΒK΁ͷ࠷୹ܦ࿏ͷݸ਺ɺ͸ͦͷ͏ͪϊʔυJΛ ௨Δ΋ͷͷݸ਺ʣ ρhj ρhj (i)
  6. ωοτϫʔΫՊֶೖ໳ w 8BUUTBOE4USPHBU[ TNBMMXPSMEOFUXPSL w ͍ΘΏΔεϞʔϧϫʔϧυωοτϫʔΫͷੜ੒Ϟσϧɻ
 ϊʔυ਺/ ฏۉ౓਺L Τοδͷܨ͔͗͑֬཰QΛ༩͑ͯɺ࣍ݩ֨ࢠά ϥϑΛ࡞੒ͨ͠ޙ֬཰QͰΤοδͷܨ͔͗͑Λߦ͏ɻQͷ࣌͸ϥϯμ

    ϜωοτϫʔΫʹͳΔɻ w DIBSBDUFSJTUJDQBUIMFOHUIখ w DMVTUFSJOHDPF⒏DJFOUେ
 ˠεϞʔϧϫʔϧυੑ w εϞʔϧϫʔϧυੑ͸༷ʑͳωοτϫʔΫʹ
 ݟΒΕΔීวతͳੑ࣭ɻ
  7. ωοτϫʔΫՊֶೖ໳ w #BSBCBTJ TDBMFGSFFOFUXPSL w εέʔϧϑϦʔωοτϫʔΫͷੜ੒Ϟσϧɻ খ͞ͳωοτϫʔΫ͔Βελʔτͯ͠ɺͦ ͜ʹ࣍ʑͱϊʔυΛՃ͑Δ͜ͱͰ࡞͍ͬͯ ͘ɻͦͷࡍʢطଘͷάϥϑͷʣͲͷϊʔυ ʹ઀ଓ͢Δ͔͸ϊʔυͷ౓਺ʹൺྫ͢Δ֬

    ཰ʹΑܾͬͯΊΒΕΔɻ w ౓਺෼෍͕΂͖৐෼෍ʹै͍ɺಛ௃తͳε έʔϧ͕ݟΒΕͳ͍ͨΊεέʔϧϑϦʔͱ ݺ͹ΕΔɻ w εέʔϧϑϦʔੑ΋888ͳͲ༷ʑͳωο τϫʔΫʹݟΒΕΔීวతͳੑ࣭͕ͩɺ෺ ཧత੍໿ͷͨΊ೴Ͱ͸΄ͱΜͲݟΒΕͳ͍ɻ
  8. ωοτϫʔΫ͔ΒԿ͕෼͔Δʁ ࿦จ঺հᶃ
 %ZOBNJDSFDPOpHVSBUJPOPGIVNBOCSBJOOFUXPSLT EVSJOHMFBSOJOH #BTTFUU %4 8ZNCT /' 1PSUFS ."

    .VDIB 1+ $BSMTPO +. (SBGUPO 45   IUUQTEPJPSHQOBT w .PEVMBSJUZʹண໨ͯ͠ɺλεΫͷशख़౓ͱG.3*͔ΒଌఆՄೳͳྔͱͷ ؒͷ૬ؔΛࣔͨ͠ɻ w ಈతͳωοτϫʔΫͷղੳʹΑͬͯɺ೴಺ͷϞδϡϥʔߏ଄͸ֶश࣌ʹ มԽ͍ͯ͠Δ͜ͱ͕෼͔ͬͨɻ
  9. ωοτϫʔΫ͔ΒԿ͕෼͔Δʁ ࿦จ঺հᶄ #SBJODPOOFDUJWJUZEZOBNJDTEVSJOHTPDJBMJOUFSBDUJPO SFqFDUTPDJBMOFUXPSLTUSVDUVSF 4DINÅM[MF 3 #SPPL0`%POOFMM . (BSDJB +0

    $BTDJP $/ #BZFS + #BTTFUU %4 ʜ'BML &#   IUUQTEPJPSHQOBT w ࣾձతഉআ͞Ε͍ͯΔঢ়گʹ߹͏ͱɺ.FOUBMJ[JOHOFUXPSLͱݺ͹ΕΔ ෦Ґͷ݁߹ڧ౓͕ڧ͘ͳΔɻ
  10. ߏ଄͔ΒμΠφϛΫε΁ w ཧ࿦తΞϓϩʔν͔Β͸Կ͕ݴ͑Δ͔ʁ w ϚΧΫβϧͰܭࢉͨ͠Α͏ʹɺ೴಺ͷωοτϫʔΫʹ͓͍ͯ΋εϞʔϧ ϫʔϧυੑ͕޿͘ݟΒΕΔɻ a small-world topology can

    support both segregated/specialized and distributed/integrated information processing. #BTTFUU %4 #VMMNPSF &  4NBMMXPSMECSBJOOFUXPSLT/FVSPTDJFOUJTU   r IUUQTEPJPSH ͜ΕΛܭࢉϞσϧ͔Β໌Β͔ʹ͍ͨ͠ɻ ˠ$PNQMFYJUZͷಋೖʢڭՊॻୈষʹରԠʣ
  11. ߏ଄͔ΒμΠφϛΫε΁ $PNQMFYJUZ 4QPSOT 5POPOJ &EFMNBO In brains of higher vertebrates,

    the functional segregation of local areas that differ in their anatomy and physiology contrasts sharply with their global integration during perception and behavior. 5POPOJ ( 4QPSOT 0 &EFMNBO (.  "NFBTVSFGPSCSBJODPNQMFYJUZSFMBUJOHGVODUJPOBMTFHSFHBUJPOBOEJOUFHSBUJPOJOUIFOFSWPVTTZTUFN1SPDFFEJOHTPG UIF/BUJPOBM"DBEFNZPG4DJFODFT   rIUUQTEPJPSHQOBT ෦෼ಉ࢜Ͱ͸ಠཱʢ෼཭ʣ͍ͯ͠Δ͕ɺશମͱͯ͠ΈͨΒ౷߹͍ͯ͠ΔγεςϜ ͕ཧ૝తɻͦͷࢦඪͱͯ͠ͷ$PNQFYJUZɻ
  12. ߏ଄͔ΒμΠφϛΫε΁ NVUVBMJOGPSNBUJPOɾJOUFHSBUJPO ܥΛͭͷ෦෼ܥʹ෼͚ɺͦΕΒΛͱද͢ʢL͸੒෼ͷ਺ɺK͸Մೳͳݸ ͷ෼ׂʹ͚ͭΒΕͨJOEFYʣɻ ෦෼ܥಉ࢜ͷಠཱੑ͸NVUVBMJOGPSNBUJPO͔Β෼͔Δɻ Xk j , X −

    Xk j n Ck MI(Xk j ; X − Xk j ) = H(Xk j ) + H(X − Xk j ) − H(X) ʢͨͩ͠ɺ)͸Τϯτϩϐʔʣ ͞ΒʹɺJOUFHSBUJPOͱݺ͹ΕΔ࣍ͷྔΛಋೖ͢Δɻ I(X) = n ∑ i=1 H(xi ) − H(X) JOUFHSBUJPOͱNVUVBMJOGPSNBUJPO͸࣍ͷؔ܎Ͱ݁͹Ε͍ͯΔɻ I(X) = n−1 ∑ i=1 MI({xi }; {xi+1,...,n })
  13. ߏ଄͔ΒμΠφϛΫε΁ ΑΓܭࢉྔ͕খ͍͞ࢦඪͱͯ͠ɺ࣍ͷͭͷ౳ՁͳදݱΛ࣋ͭDPNQMFYJUZ΋ఏҊ͞Εͯ ͍Δʢಋग़͸লུɺOFVSBMDPNQFYJUZͱ౳ՁͰ͸ͳ͍͕ඇৗʹ͍ۙʣɻ C(X) = H(X) − ∑ i H(xi

    ∣ X − xi ) = ∑ i MI(xi ; X − xi ) − I(X) = (n − 1)I(X) − n⟨I(X − xi )⟩ ઢܗఆৗ֬཰աఔΛߟ͑ͨͨΊɺ) 9 ͕ղੳతʹղ͚Δʢಋग़͸লུʣ H(X) = 1 2 ln((2πe)n ⋅ ∣ COV(X) ∣ )
  14. ߏ଄͔ΒμΠφϛΫε΁ w·ͨɺٯʹϥϯμϜͳωοτϫʔΫ͔Βग़ൃͯ͠ɺ$PNQMFYJUZ͕ େ͖͘ͳΔΑ͏ʹ.$.$ͰωοτϫʔΫΛൃలͤ͞ΔͱεϞʔϧ ϫʔϧυੑ͕ݱΕͨɻ ʢਤ͸લϖʔδͱ΋4QPSOT 0 5POPOJ ( &EFMNBO (.

     5IFPSFUJDBMOFVSPBOBUPNZBOEUIFDPOOFDUJWJUZPGUIF DFSFCSBMDPSUFY#FIBWJPVSBM#SBJO3FTFBSDI  r rIUUQTEPJPSH4  ΑΓʣ
  15. ·ͱΊ w ࠓ೔΍ͬͨ͜ͱ w ωοτϫʔΫղੳͷجૅʢωοτϫʔΫΛಛ௃෇͚Δछʑͷྔͷಋೖͱͦͷ ܭࢉํ๏ʣΛಋೖͨ͠ɻ w ؆୯ͳωοτϫʔΫΛ࢖࣮ͬͯࡍʹܭࢉͯ͠Έͨɻ w ʢ࠷ۙͷʣωοτϫʔΫՊֶͷχϡʔϩαΠΤϯε΁ͷԠ༻ྫΛ঺հͨ͠ɻ

    w $PNQMFYJUZͱݺ͹ΕΔྔΛಋೖͯ͠ɺOFUXPSLUPQPMPHZͱμΠφϛΫεʢ৘ ใॲཧʣͱͷؔ܎ੑΛٞ࿦ͨ͠ɻ w ࠓ೔ѻ͑ͳ͔ͬͨ͜ͱ w ೴ͷ؍ଌσʔλ͔ΒωοτϫʔΫΛߏங͢Δํ๏ w ωοτϫʔΫͷʢ෺ཧతʣޮ཰ੑɺؤ݈ੑ w FUDʜ