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

Synapse: 文脈と時間経過に応じて推薦手法の選択を最適化するメタ推薦システム/smash21-synapse

Cd3d2cb2dadf5488935fe0ddaea7938a?s=47 monochromegane
September 16, 2021
Research
0
37

Synapse: 文脈と時間経過に応じて推薦手法の選択を最適化するメタ推薦システム/smash21-synapse

2021.09.16 SMASH21 Summer Symposium
https://sites.google.com/view/sig-macc/smash/smash21-summer-symposium

Cd3d2cb2dadf5488935fe0ddaea7938a?s=128

monochromegane

September 16, 2021
Tweet

More Decks by monochromegane

See All by monochromegane
Cd3d2cb2dadf5488935fe0ddaea7938a?s=48 monochromegane
0
56
Cd3d2cb2dadf5488935fe0ddaea7938a?s=48 monochromegane
1
2.5k
Cd3d2cb2dadf5488935fe0ddaea7938a?s=48 monochromegane
2
99
Cd3d2cb2dadf5488935fe0ddaea7938a?s=48 monochromegane
0
1.1k
Cd3d2cb2dadf5488935fe0ddaea7938a?s=48 monochromegane
0
390
Cd3d2cb2dadf5488935fe0ddaea7938a?s=48 monochromegane
0
1.8k
Cd3d2cb2dadf5488935fe0ddaea7938a?s=48 monochromegane
0
230
Cd3d2cb2dadf5488935fe0ddaea7938a?s=48 monochromegane
0
2.3k
Cd3d2cb2dadf5488935fe0ddaea7938a?s=48 monochromegane
0
4.6k

Other Decks in Research

See All in Research
14da6ebc2e909305afdb348e7970de81?s=48 wingnus
0
870
02335420aa18081affbcb8974de06429?s=48 connect24h
1
2.1k
638dd29baf6aa478d8eb0aeb0075c404?s=48 sshimizu2006
2
590
Cb3c74960f845f8309702eb95f7e2f2a?s=48 kuri8ive
0
130
18923a96dc13fe272f16c77167102f42?s=48 koheishinden
0
150
6f881fac818f465f2b375ed7e335cf2a?s=48 tathi
0
180
9782dd351a8f1737d4ef02d839e821f9?s=48 kitachan_black
0
11k
F4ac14214e26fafd85efecbc95e677c3?s=48 hayataka88
5
4.3k
9782dd351a8f1737d4ef02d839e821f9?s=48 kitachan_black
0
3.1k
92c88c6409efa2efefadcdeedfb94cd3?s=48 100_tools
0
160
9e650916f36300d64c9c61eeb4ab697e?s=48 a1da4
0
130
0d4ef9af8e4f0cf5c162b48ba24faea6?s=48 ept
0
110

Featured

See All Featured
3d6ace9554821d552146413bcdf874f6?s=48 trishagee
38
5.9k
9cde37f47e4a800ea081ea42de8d749a?s=48 dougneiner
56
5.8k
027d1ebf66cc039a0bd3b55eeadbe75d?s=48 sachag
272
17k
1519587b1a0febb658c36c94f6272b0d?s=48 roundedbygravity
249
22k
Dafc4723e9a1c067796c0fec6f509774?s=48 lemiorhan
634
51k
2bb923c57a1fdc3a2eb484bb8d565fd2?s=48 ufuk
62
7.4k
32de0bd2ba869609d26fd052a4622778?s=48 cromwellryan
109
6.7k
766b9746b59e6f09e280cd33cf4ed419?s=48 skipperchong
14
1.2k
0c6fd6a08d0f898159cda7cafcacf07f?s=48 afnizarnur
181
14k
7edec06b5435e0dac07a353b2cc26253?s=48 geeforr
337
30k
E14f55d3f939977cecbf51b64ff6f861?s=48 keithpitt
405
20k
E195ae45320d9202eaa01c9f1d31a416?s=48 marktimemedia
15
800

Transcript

  1. ࡾ୐ ༔հ1,2ɺ็ ߃ݑ3 1. Pepabo R&D Institute, GMO Pepabo, Inc.,

    2. ۝भେֶ େֶӃγεςϜ৘ใՊֶ෎ ৘ใ஌ೳ޻ֶઐ߈ 3. ۝भେֶ େֶӃγεςϜ৘ใՊֶݚڀӃ ৘ใ஌ೳ޻ֶ෦໳ 2021.09.16 SMASH21 Summer Symposium Synapse: จ຺ͱ࣌ؒܦաʹԠͯ͡ ਪનख๏ͷબ୒Λ࠷దԽ͢Δ ϝλਪનγεςϜ

  2. 1. ͸͡Ίʹ 2. ؔ࿈ݚڀ 3. ఏҊख๏ 4. ධՁ 5. ͓ΘΓʹ

    2 ໨࣍

  3. 1. ͸͡Ίʹ

  4. • ৘ใγεςϜʹ͓͚Δ৘ใաଟ໰୊Λղܾ͢ΔɺਪનγεςϜͷಋೖ • ͳΜΒ͔ͷํࡦʢ=ਪનख๏ʣʹج͖ͮଟ਺ͷબ୒ࢶ͔Βར༻ऀ͕ڵຯΛ࣋ ͭ΋ͷΛఏҊ͢ΔγεςϜ 4 എܠ 6TFST 3FDPNNFOEBUJPO .FUIPE

    1SPEVDUT IUUQTJDPOTDPN

  5. • ৘ใγεςϜʹ͓͚Δ৘ใաଟ໰୊Λղܾ͢ΔɺਪનγεςϜͷಋೖ • ͳΜΒ͔ͷํࡦʢ=ਪનख๏ʣʹج͖ͮଟ਺ͷબ୒ࢶ͔Βར༻ऀ͕ڵຯΛ࣋ ͭ΋ͷΛఏҊ͢ΔγεςϜ • ਺ଟ͘ͷਪનख๏͕ఏҊ͞Ε͍ͯΔ 5 എܠ 6TFST

    3FDPNNFOEBUJPO .FUIPET 1SPEVDUT IUUQTJDPOTDPN

  6. • ৘ใγεςϜʹ͓͚Δ৘ใաଟ໰୊Λղܾ͢ΔɺਪનγεςϜͷಋೖ • ͳΜΒ͔ͷํࡦʢ=ਪનख๏ʣʹج͖ͮଟ਺ͷબ୒ࢶ͔Βར༻ऀ͕ڵຯΛ࣋ ͭ΋ͷΛఏҊ͢ΔγεςϜ • ਺ଟ͘ͷਪનख๏͕ఏҊ͞Ε͍ͯΔ → ޮՌతͳʮਪનख๏ͷબఆʯ͕ॏཁ 6

    എܠ 6TFST #FTU3FDPNNFOEBUJPO .FUIPE 1SPEVDUT IUUQTJDPOTDPN

  7. 7 എܠ • ޮՌతͳਪનख๏͸ঢ়گ΍࣌ؒͷܦաʹΑͬͯҟͳΔ • ͔͠͠ͳ͕Βɺ࣮؀ڥͰͷܧଓతͳਪનख๏ͷධՁʹ͸ػձଛࣦ͕൐͏ ӡ༻্ͷ՝୊ • ৘ใγεςϜʹ͓͚Δ৘ใաଟ໰୊Λղܾ͢ΔɺਪનγεςϜͷಋೖ •

    ͳΜΒ͔ͷํࡦʢ=ਪનख๏ʣʹج͖ͮଟ਺ͷબ୒ࢶ͔Βར༻ऀ͕ڵຯΛ࣋ ͭ΋ͷΛఏҊ͢ΔγεςϜ • ਺ଟ͘ͷਪનख๏͕ఏҊ͞Ε͍ͯΔ → ޮՌతͳʮਪનख๏ͷબఆʯ͕ॏཁ

  8. • ਪનख๏ͷ༏ྼ͸ঢ়گʢ=จ຺ʣͱ࣌ؒͷܦաʹΑͬͯࠨӈ͞ΕΔ • ޮՌతͳਪનख๏Λػձଛࣦ͕ͳ͍Α͏ʹจ຺ͱ࣌ؒͷܦաʹԠͯ͡࢖͍෼͚ ͍ͨ • จ຺ͱ࣌ؒͷܦաʹԠͯ͡ਪનख๏ͷબ୒Λࣗಈ͔ͭܧଓతʹ࠷దԽ͢Δ ϝλਪનγεςϜ • →

    ࠷ળͳਪનख๏ͷબ୒Λଟ࿹όϯσΟοτ໰୊ͱΈͳͯ͠ղ͘ 8 ݚڀͷ໨తͱఏҊͷࠎࢠ

  9. 2. ؔ࿈ݚڀ

  10. ଟ࿹όϯσΟοτ໰୊

  11. • ʮ࿹ʯͱݺ͹ΕΔෳ਺ͷީิ͔ΒಘΒΕΔใुΛ࠷େԽ͢Δ໰୊ • ϓϨΠϠʔ͸Ұ౓ͷࢼߦͰ1ͭͷ࿹Λબ୒͠ɺใुΛಘΔ • ͦΕͧΕͷ࿹͸͋Δใु෼෍ʹै͍ใुΛੜ੒ • ͨͩ͠ɺϓϨΠϠʔ͸͜ͷใु෼෍Λࢼߦͷ݁Ռ͔Βਪଌ͢Δඞཁ͕͋Δ 11 ଟ࿹όϯσΟοτ໰୊

    • ϓϨΠϠʔ͸͋Δ࣌఺ͷ࿹ͷධՁʹج͖ͮʮ׆༻ʯͱʮ୳ࡧʯΛฒߦͯ͠ߦ͏ • ͜ͷτϨʔυΦϑΛղফ͢ΔͨΊʹ༷ʑͳղ๏͕ఏҊ͞Ε͍ͯΔ

  12. • ࿹͝ͱͷใु෼෍͸ৗʹಉ͡Ͱ͋Δͱ͍͏Ծఆ • → ঢ়گ΍ଐੑ͝ͱʹใु෼෍͕ҟͳΔͷͰ͸ͳ͍͔ʁ • ྫʣ೥୅͝ͱʹਓؾͷ঎඼͕ҧ͏ɺ࠷ۙɺಉ͡ΧςΰϦͷ঎඼Λങͬͨ • → ࣌ؒͷܦաʹΑͬͯ΋ใु෼෍͕มԽ͢ΔͷͰ͸ͳ͍͔ʁ

    • ྫʣ৽͍͠χϡʔεهࣄɺҰ࣌తͳྲྀߦ 12 ଟ࿹όϯσΟοτ໰୊ͷ֦ு • ʮจ຺෇͖ʯ·ͨ͸ʮඇఆৗʯͳଟ࿹όϯσΟοτ໰୊ͱ֦ͯ͠ு͞Ε͍ͯΔ

  13. ਪનख๏ͷબఆʹ͓͚Δ ؍఺ͷ੔ཧ

  14. • ਪનख๏ͷ૬ରతͳੑೳͷ༏ྼΛॿ௕͢ΔཁҼͷଘࡏ 14 ਪનख๏ͷબఆʹ͓͚Δจ຺ͷߟྀͷॏཁੑ <>&LTUSBOE .BOE3JFEM +8IFOSFDPNNFOEFSTGBJMQSFEJDUJOHSFDPNNFOEFSGBJMVSFGPSBMHPSJUINTFMFDUJPOBOEDPNCJOBUJPO  1SPDFFEJOHTPGUIFTJYUI"$.DPOGFSFODFPO3FDPNNFOEFSTZTUFNT QQr

      <>#SBVOIPGFS . $PEJOB 7BOE3JDDJ '4XJUDIJOHIZCSJEGPSDPMETUBSUJOHDPOUFYUBXBSFSFDPNNFOEFSTZTUFNT 1SPDFFEJOHTPGUIF UI"$.$POGFSFODFPO3FDPNNFOEFSTZTUFNT QQr   <>"OEFSTPO " .BZTUSF - "OEFSTPO * .FISPUSB 3BOE-BMNBT ."MHPSJUINJDF⒎FDUTPOUIFEJWFSTJUZPGDPOTVNQUJPOPOTQPUJGZ  1SPDFFEJOHTPG5IF8FC$POGFSFODF QQr   ઌߦݚڀ ֓ཁ ਪનख๏ͷબఆʹؔ͢ΔཁҼͷ෼ੳɾར༻ <> ϋΠϒϦουਪનʹ͓͚Δਪનख๏ ͷબఆࣦഊͷݪҼΛ୳ͬͨݚڀ ར༻ऀ͝ͱʹਪનख๏Λ࢖͍෼͚Δ͜ͱͰੑೳ͕վળ͞ΕΔͱใࠂ͠ɼ ͦͷཁҼͷ෼ੳ͕ඞཁ <> ίʔϧυελʔτͷঢ়گʹݶఆͨ͠ ϋΠϒϦουਪનͷݚڀ ਪનख๏ͷબ୒ʹӨڹΛٴ΅͢ཁҼͱͯ͠ɼར༻ऀ΍঎඼ɼίϯςΩε τʹର͢ΔධՁͷ஝ੵ۩߹Λར༻ <> ԻָετϦʔϛϯάαʔϏεʹ͓͚ Δࢹௌ܏޲ͱਪનख๏ͷޮՌͷݚڀ ਪનख๏ͷબ୒ʹӨڹΛٴ΅͢ཁҼͱͯ͠ɼࢹௌ܏޲ͷଟ༷ੑʹண໨Ͱ ͖Δ͜ͱΛࣔͨ͠

  15. • ར༻ऀͷᅂ޷৘ใͷ஝ੵʹ൐͍ɺਪનਫ਼౓ͷ޲্͕ظ଴Ͱ͖Δਪનख๏ͷଘࡏ 15 ਪનख๏ͷબఆʹ͓͚Δ࣌ؒͷܦաͷߟྀͷॏཁੑ ಺༰ϕʔεͷਪનख๏͸Ұఆͷਫ਼౓ ᅂ޷৘ใΛ༻͍Δਪનख๏͸ਫ਼౓͕ঃʑʹ޲্ ਪનख๏ͷ༗ޮੑ͕ඇఆৗͰ͋Δ͜ͱ Λલఏͱͨ͠ղ๏͕ඞཁͱͳΔ

  16. ଟ࿹όϯσΟοτ໰୊ͷղ๏Λ༻͍ͨ ਪનख๏ͷબఆ

  17. • ઌߦݚڀʹ͓͚Δจ຺ͱ࣌ؒͷܦաͷߟྀ 17 ଟ࿹όϯσΟοτ໰୊ͷղ๏Λ༻͍ͨਪનख๏ͷબఆ ઌߦݚڀ จ຺ ࣌ؒͷܦա <> º º

    <> <> ˚ º ࿹ͱͳΔਪનख๏͕จ຺Λѻ͏͕ਪનख๏ͷબఆʹ͸จ຺Λߟྀ͠ͳ͍ <> <> ˓ º <>͸ਪન࣌ͷӾཡதͷ঎඼ಛੑΛจ຺ͱͯ͠ར༻ʢ<>͸ఏࣔͳ͠ʣ ˎ͍ͣΕͷख๏΋࣌ؒͷܦաͷߟྀ͕े෼Ͱ͸ͳ͍ <>'FM ’DJP $; 1BJYB _P ,7 #BSDFMPT $"BOE1SFVY 1"NVMUJBSNFECBOEJUNPEFMTFMFDUJPOGPSDPMETUBSUVTFSSFDPNNFOEBUJPO 1SPDFFEJOHTPGUIFUI $POGFSFODFPO6TFS.PEFMJOH "EBQUBUJPOBOE1FSTPOBMJ[BUJPO QQr   <>#SPE FO # )BNNBS . /JMTTPO #+BOE1BSBTDIBLJT %&OTFNCMFSFDPNNFOEBUJPOTWJBUIPNQTPOTBNQMJOHBOFYQFSJNFOUBMTUVEZXJUIJOF DPNNFSDF SEJOUFSOBUJPOBMDPOGFSFODFPOJOUFMMJHFOUVTFSJOUFSGBDFT QQr   <>$BO _BNBSFT 3 3FEPOEP .BOE$BTUFMMT 1.VMUJBSNFESFDPNNFOEFSTZTUFNCBOEJUFOTFNCMFT 1SPDFFEJOHTPGUIFUI"$.$POGFSFODFPO 3FDPNNFOEFS4ZTUFNT QQr   <>4BOUBOB .3 .FMP -$ $BNBSHP ') #SBOE _BP # 4PBSFT " 0MJWFJSB 3.BOE$BFUBOP 4$POUFYUVBM.FUB#BOEJUGPS3FDPNNFOEFS4ZTUFNT 4FMFDUJPO 'PVSUFFOUI"$.$POGFSFODFPO3FDPNNFOEFS4ZTUFNT QQr   <>ࡾ୐༔հɼ็߃ݑ4ZOBQTFจ຺ʹԠͯ͡ܧଓతʹਪનख๏ͷબ୒Λ࠷దԽ͢ΔਪનγεςϜ ిࢠ৘ใ௨৴ֶձ࿦จࢽ% 7PM /P QQr  

  18. 3. ఏҊख๏

  19. • จ຺ͱ࣌ؒͷܦաʹԠͯ͡ਪનख๏ͷબ୒Λࣗಈ͔ͭܧଓతʹ࠷దԽ͢Δ ϝλਪનγεςϜ • ػձଛࣦΛ཈͑ͨจ຺ͱ࣌ؒͷܦաʹԠͨ͡࠷ળͳબ୒Λୡ੒͢ΔͨΊɺ จ຺෇͖ɺ͔ͭɺඇఆৗͳଟ࿹όϯσΟοτ໰୊ͷղ๏Λ༻͍Δ • ਪનख๏ͷ༗ޮੑ͕ٯస͢Δঢ়گʢޙड़ʣʹରॲ͢ΔͨΊɺैདྷͷղ๏Λ ֦ு͠ɺੵۃతͳ୳ࡧΛଅ͢ •

    બ୒ͨ͠ਪનख๏ͱ͜Εʹର͢Δར༻ऀͷ൓ԠΛه࿥͠ɺબ୒ͷܧଓతͳ վળʹ༻͍Δ 19 ఏҊγεςϜ (Synapse)

  20. 20 ఏҊγεςϜ (Synapse)

  21. จ຺ͱ࣌ؒͷܦաΛߟྀͨ͠ ଟ࿹όϯσΟοτ໰୊ͷղ๏

  22. 22 Time-varying Thompson Sampling (TVTP) [15] Dynamic Context drift Modeling

    จ຺෇͖ɺ͔ͭɺඇఆৗͳใुͷมԽΛ ѻ͏ͨΊใुͷมಈΛ૊ΈࠐΜͩϞσϧ ͕༻͍ΒΕΔ <>;FOH $ 8BOH 2 .PLIUBSJ 4BOE-J 50OMJOFDPOUFYUBXBSFSFDPNNFOEBUJPOXJUIUJNFWBSZJOHNVMUJBSNFECBOEJU 1SPDFFEJOHTPGUIF OE"$.4*(,%%JOUFSOBUJPOBMDPOGFSFODFPO,OPXMFEHFEJTDPWFSZBOEEBUBNJOJOH QQr   <>'JH(SBQIJDBMNPEFMSFQSFTFOUBUJPOGPSCBOEJUQSPCMFN yk,t ⇠ N(xT t (cwk + ✓k ⌘k,t), 2 k ) <latexit sha1_base64="MOOU/xvogEgNwsUMW1ExG3Gil7E=">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</latexit> ίϯςΩετ ఆৗ߲ ඇఆৗ߲ ʢεέʔϧ߲ɺυϦϑτ߲ʣ ؍ଌޡࠩ

  23. 23 Time-varying Thompson Sampling (TVTP) [15] ঢ়ଶۭؒϞσϧ ใुϞσϧͷύϥϝʔλͷࣄޙ෼෍ͱυ Ϧϑτ߲ͷજࡏঢ়ଶͷஞ࣍ਪఆʹཻࢠ ϑΟϧλͱΧϧϚϯϑΟϧλ͕༻͍ΒΕΔ

    <>;FOH $ 8BOH 2 .PLIUBSJ 4BOE-J 50OMJOFDPOUFYUBXBSFSFDPNNFOEBUJPOXJUIUJNFWBSZJOHNVMUJBSNFECBOEJU 1SPDFFEJOHTPGUIF OE"$.4*(,%%JOUFSOBUJPOBMDPOGFSFODFPO,OPXMFEHFEJTDPWFSZBOEEBUBNJOJOH QQr   <>'JH(SBQIJDBMNPEFMSFQSFTFOUBUJPOGPSCBOEJUQSPCMFN yk,t ⇠ N(xT t (cwk + ✓k ⌘k,t), 2 k ) <latexit sha1_base64="MOOU/xvogEgNwsUMW1ExG3Gil7E=">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</latexit> ΧϧϚϯϑΟϧλ ཻࢠϑΟϧλ

  24. 24 Time-varying Thompson Sampling (TVTP) [15] ֬཰Ұக๏ ֤࿹ͰٻΊͨύϥϝʔλͷࣄޙ෼෍ʹै ͍αϯϓϦϯάͨ݁͠ՌΛ࿹ͷબఆʹ༻ ͍Δ͜ͱͰଟ࿹όϯσΟοτղ๏ͱ౷߹

    <>;FOH $ 8BOH 2 .PLIUBSJ 4BOE-J 50OMJOFDPOUFYUBXBSFSFDPNNFOEBUJPOXJUIUJNFWBSZJOHNVMUJBSNFECBOEJU 1SPDFFEJOHTPGUIF OE"$.4*(,%%JOUFSOBUJPOBMDPOGFSFODFPO,OPXMFEHFEJTDPWFSZBOEEBUBNJOJOH QQr   <>'JH(SBQIJDBMNPEFMSFQSFTFOUBUJPOGPSCBOEJUQSPCMFN yk,t ⇠ N(xT t (cwk + ✓k ⌘k,t), 2 k ) <latexit sha1_base64="MOOU/xvogEgNwsUMW1ExG3Gil7E=">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</latexit> a(t) = arg max j=1,K xT t ¯ wk,t 1 ¯ wk,t 1 ⇠ Nm(¯ µwk , ¯ ⌃wk ) <latexit sha1_base64="DJGY+opJHBQ8ZtUO2/R3rcYFv7Q=">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</latexit> ࣄޙ෼෍ ࣄޙ෼෍

  25. 25 TVTPͷ՝୊ a(t) = arg max j=1,K xT t ¯

    wk,t 1 ¯ wk,t 1 ⇠ Nm(¯ µwk , ¯ ⌃wk ) <latexit sha1_base64="DJGY+opJHBQ8ZtUO2/R3rcYFv7Q=">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</latexit> ¯ ⌃wk = 1 p2 p X i=1 2(i) k ⌃(i) wk , where p is number of particles <latexit sha1_base64="xTwhDLbTasGLMML2v9yvADUgJxI=">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</latexit> • ࢼߦճ਺ͷ૿Ճʹ൐͏ਪનख๏ͷબఆͷภΓ • ͋Δ࣌఺ͰධՁͷ௿͍ਪનख๏Λ୳ࡧ͢Δػձ͕ۃ୺ʹ௿Լ • ༗ޮੑ͕ٯస͢ΔΑ͏ͳঢ়گ΁ͷ௥ै͕஗ΕΔ ࢼߦճ਺ͷ૿Ճʹ൐͏ٸܹͳݮগʹΑΓ͋Δ ࣌఺ͷධՁʹج͍ͮͨબ୒ʹݻఆ͞ΕΔ

  26. 26 ఏҊํࣜ: Aggressive Exploration TVTP (AE-TVTP) a(t) = arg max

    j=1,K xT t ¯ wk,t 1 ¯ wk,t 1 ⇠ Nm(¯ µwk , ¯ ⌃wk ) <latexit sha1_base64="DJGY+opJHBQ8ZtUO2/R3rcYFv7Q=">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</latexit> ¯ ⌃wk = 1 p2 p X i=1 2(i) k ⌃(i) wk , where p is number of particles <latexit sha1_base64="xTwhDLbTasGLMML2v9yvADUgJxI=">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</latexit> • ࢼߦճ਺ͷ૿Ճʹ൐͏ਪનख๏ͷબఆͷภΓΛղফ • ͋Δ࣌఺ͰධՁͷ௿͍ਪનख๏Λੵۃతʹ୳ࡧ͢ΔػձΛઃ͚Δ • ࿹ͷ༗༻ੑ͕ٯస͢Δ؀ڥʹ͓͍ͯɺ௥ैੑͷ޲্ΛਤΔ ཻࢠͷฏۉ ཻ֤ࢠͰͷ৐ࢉͷΈ

  27. 4. ධՁ

  28. • ࣮ࡍͷECαΠτ͔Β࠾औͨ͠4ͭͷਪનख๏ͷ঎඼ΧςΰϦ͝ͱͷΫϦοΫ཰ ͷਪҠ࣮੷σʔλΛ༻͍ͯఏҊγεςϜͷ༗ޮੑΛධՁ͢Δ • 2019/6/20ʙ8/4·Ͱͷ໿225ສճͷਪનσʔλ͔Βࢉग़ • 4ͭͷਪનख๏ͱ18ͷ঎඼ΧςΰϦ͝ͱʹɺ1࣌ؒ୯ҐͰूܭ 28 ධՁσʔλͱਪનख๏

  29. • બ୒ͨ͠ਪનख๏͔ΒಘΒΕΔΫϦοΫ਺ͷγϛϡϨʔγϣϯ • ਪનख๏ͷબ୒ʹ͸ɺఏҊํࣜΛؚΉଟ࿹όϯσΟοτͷղ๏Λ༻͍Δ • બ୒ͨ͠ਪનख๏͸࣮੷σʔλͷΫϦοΫ཰Λύϥϝʔλͱ͢ΔϕϧψʔΠ ෼෍ʹै͍ɺਪન݁Ռ͕ΫϦοΫ͞ΕΔ΋ͷͱ͢Δ • ཚ਺Λ༻͍ͨγϛϡϨʔγϣϯΛฏۉԽ͢ΔͨΊ50ճͷγϛϡϨʔγϣϯ ݁ՌͷฏۉΛ݁Ռͱͯ͠༻͍Δ

    • ࣮ࡍͷਪનγεςϜͷڍಈͱ߹ΘͤΔͨΊɺใु͸1࣌ؒ͝ͱʹ·ͱΊͯ ϑΟʔυόοΫ͞ΕΔ΋ͷͱ͢Δ 29 ධՁํ๏(1/2)

  30. • จ຺ͱ࣌ؒͷܦաͷߟྀͷͦΕͧΕͷد༩౓Λ໌Β͔ʹ͢Δ4άϧʔϓͷγ ϛϡϨʔγϣϯΛ࣮ࢪ 30 ධՁํ๏(2/2) ࣌ؒͷܦա º ˓ จ຺ º

    "ىटͷ࠷ળͳਪનख๏ΛશظؒҰ؏ ͯ͠༻͍Δ $࣌఺͝ͱʹධՁͷߴ͍ਪનख๏Λόϯ σΟοτΛ༻͍ͯબఆ ˓ #จ຺͝ͱʹ࠷ળͳਪનख๏Λશظؒ Ұ؏ͯ͠༻͍Δ %จ຺͝ͱ࣌఺͝ͱʹධՁͷߴ͍ਪનख ๏ΛόϯσΟοτΛ༻͍ͯબఆ • จ຺ʹ͸ɺਪન࣌ʹӾཡதͷ঎඼ΧςΰϦΛ༻͍Δ • ଟ࿹όϯσΟοτղ๏͸ɺLTS(จ຺) ɺTVTP(จ຺/࣌ؒͷܦա) ɺAE-TVTP(จ຺/࣌ؒͷܦա)

  31. • Bάϧʔϓʢจ຺ʣ͸ظट࣌఺ʹ͓ ͍ͯAάϧʔϓͱจ຺ʹΑΔࠩҟ͕ ΄΅ͳ͍ͨΊ݁Ռ΋ࠩҟͳ͠ • Cάϧʔϓʢ࣌ؒͷܦաʣ͸ਪનख ๏ͷ༗ޮੑͷมԽʹ௥ैͨ͜͠ͱ Ͱվળ͕ݟΒΕΔ • Dάϧʔϓʢจ຺ͱ࣌ؒͷܦաʣ͸

    ঎඼ΧςΰϦ͝ͱͷมԽʹ௥ै͠ ͨ͜ͱͰߋͳΔվળ͕ݟΒΕΔ 31 ධՁ݁Ռ: AάϧʔϓΛج४ͱͨ͠ྦྷੵใुͷࠩͷൺֱ จ຺ͱ࣌ؒͷܦաͷߟྀͳΒͼʹɺٯస؀ڥͷ௥ ैੑΛߴΊͨఏҊํࣜʹΑͬͯ໿૿Ճ จ຺ͷΈ ࣌ؒͷܦաͷΈ

  32. • Dάϧʔϓͷจ຺͝ͱͷվળ݁ՌΛ ෼ੳ͢Δͱɺਪનख๏ͷ༗ޮੑͷ มಈͷগͳ͍จ຺Ͱ͸ɺશͯͷղ ๏ʹ͓͍ͯ୳ࡧͷίετΛճऩͰ ͖͍ͯͳ͍ • มಈͷେ͖͍จ຺Ͱ͸ੵۃతͳ୳ ࡧʹΑΓAE-TVTP͕େ͖͘վળ͠ ͨɻTVTP͸ਪનख๏͕ݻఆ͞Εվ

    ળʹࢸΒͳ͔ͬͨ 32 ධՁ݁Ռ: ਪનख๏ͷ༗ޮੑͷมಈ౓߹͍ͱվળ཰

  33. 33 ߟ࡯ • ਪનख๏ͷ༗ޮੑ͕ٯస͢Δࠨྻ ʹ͓͍ͯ͸ఏҊख๏͕༗ޮ • ӈྻʹ͓͍ͯ͸ɺఏҊख๏ͷੵۃ తͳ୳ࡧʹىҼͯ͠ɺظؒதܧଓత ʹྦྷੵϦάϨοτ͕૿Ճ͢Δɻ3ׂ ఔ౓Ͱಉ༷ͷࣄ৅Λ֬ೝɻ

    ਪનख๏ͷ༗ޮੑʹٯస͕͋Δ঎඼ΧςΰϦ ࠨ ͱɺ ͳ͍঎඼ΧςΰϦ ӈ ʹ͓͚ΔྦྷੵϦάϨοτͷਪҠ ˎྦྷੵϦάϨοτ͸ਪનख๏ͷ͏ͪ࠷େͷظ଴஋ͱબ୒ͨ͠ਪનख ๏ͷظ଴஋ͷࠩΛظؒ·Ͱʹ߹ܭͨ͠΋ͷ มԽͷͳ͍ظؒʹ͓͍ͯ΋ػձଛࣦΛ ௿ݮ͢ΔదԠతͳ୳ࡧख๏ͷݚڀ΁

  34. 5. ͓ΘΓʹ

  35. • ଟ࿹όϯσΟοτΛ༻͍ͯɺจ຺ͱ࣌ؒͷܦաʹԠͯ͡ਪનख๏ͷબ୒Λࣗಈ త͔ͭܧଓతʹ࠷దԽ͢ΔϝλਪનγεςϜΛఏҊ͠ɺͦͷ༗ޮੑΛࣔͨ͠ • จ຺ͱ࣌ؒͷܦաͷߟྀͳΒͼʹٯస؀ڥͷ௥ैੑΛߴΊͨఏҊํࣜʹΑͬ ͯɺߟྀ͠ͳ͍৔߹ͱൺֱͯ͠ྦྷੵΫϦοΫ਺ͷ޲্ʹܨ͕Δ͜ͱ͕Θ͔ͬͨ • ਪનख๏ͷ༏ྼʹมԽͷͳ͍ظؒͰ͸݁Ռ͕ٯస͢ΔՄೳੑ͕֬ೝ͞Εͨ͜ͱ ͰɺػձଛࣦΛ௿ݮ͢Δղ๏͕ॏཁͰ͋Δ͜ͱ΋ࣔࠦ͞Εͨ •

    ࠓޙͷ՝୊ͱͯ͠ɺਪનख๏ͷબ୒ʹӨڹΛٴ΅͢ޮՌͷߴ͍จ຺ͷൃݟɺͳ Βͼʹಛఆͷจ຺ʹ͓͍ͯධՁͷߴ͍ਪનख๏ͷཱ֬ʹΑΔจ຺͝ͱͷޮՌ޲ ্ͷ࣮ݱ͕ڍ͛ΒΕΔ 35 ͓ΘΓʹ
  36. None