Tweet
Tweet

Slide 1

Slide 1 text

4BOTBOʹ͓͚Δ ໊ࢗը૾σʔλԽϑϩʔͱ ػցֶशͷऔΓ૊Έ 4BOTBOגࣜձࣾ 3%3FTFBSDIFSՏ߹ढ़య

Slide 2

Slide 2 text

2 ࣗݾ঺հ 4BOTBOͱ໊ࢗͷσʔλԽ 4BOTBOגࣜձࣾʹ͍ͭͯ 4BOTBOͷ໊ࢗσʔλԽϑϩʔ ػցֶशͷऔΓ૊Έ ը૾ೝࣝʹ͓͚Δػցֶश ଞػցֶशϓϩδΣΫτ ػցֶशϓϩδΣΫτͷεεϝ ͓ΘΓʹ "HFOEB

Slide 3

Slide 3 text

3 ࣗݾ঺հ 4BOTBOגࣜձࣾ %40$3% ৽ଔೖࣾ೥ ߴઐ େֶӃ ػցֶश͕޷͖ ,BHHMF !WBBBBBORVJTI

Slide 4

Slide 4 text

4BOTBOͱ໊ࢗͷσʔλԽ 4

Slide 5

Slide 5 text

4BOTBOגࣜձࣾͱ͸ 5 ๏ਓ޲໊͚ࢗ؅ཧαʔϏε ݸਓ޲໊͚ࢗΞϓϦ εΩϟφɺΞϓϦܦ༝Ͱ໊ࢗը૾ͷσʔλԽͱ؅ཧ ࣾҎ্ͷاۀಋೖɺ೥ؒ໿ԯຕͷ໊ࢗΛॲཧ ਓखʹΑΔਖ਼֬ͳσʔλԽ

Slide 6

Slide 6 text

̎ͭͷϓϩμΫτ 6 ࣾһͷ໊ࢗަ׵Λձࣾͷࢿ࢈ʹ ࣾ಺ͷ໊ࢗΛҰׅ؅ཧ ʮྡͷσεΫͷ˓˓͞Μ͕࣮͸औҾઌͷ໾һͱ஌Γ߹͍ͩͬͨʯʮૣ͘ݴͬͯΑʙʯ ໊ࢗަ׵Λਓͷܨ͕Γͱଊ͑اۀͷࢿ࢈ʹ

Slide 7

Slide 7 text

̎ͭͷϓϩμΫτ 7 ݸਓ޲໊͚ࢗ4/4ΞϓϦ ݸਓ޲͚ͷ໊ࢗΛى఺ʹͨ͠4/4ΞϓϦ ϑΟʔυ΍ϝοηʔδͱ͍ͬͨػೳʹՃ໊͑ࢗަ׵ػೳɺاۀϖʔδ &JHIUͰϏδωεͷ΍ΓऔΓ͕׬݁

Slide 8

Slide 8 text

8 ໊ࢗσʔλԽϑϩʔ ໨ࢹΦϖϨʔλ Ϋϥ΢υΦϖϨʔλ Ξ΢τιʔγϯά ໨ࢹΦϖϨʔλ ͦΕͧΕ3%͕"1*ఏڙ ϚΠΫϩαʔϏε

Slide 9

Slide 9 text

0$3 ࣾ಺֎ ΦϖϨʔλʹΑΔखೖྗ ࣙॻͷࢀর ෳ਺ͷ৚݅ʹΑΔϚονϯά จࣈɺإɺϩΰ౳ͷݕग़ ߲໨಺༰ͷਪఆ
 ࣗવݴޠॲཧ ػցֶश ΦϖϨʔλʹΑΔ໨ࢹ ࣙॻͷࢀরɺϚονϯά ۣܗݕग़ Τοδݕग़ ϗϫΠτχϯά ը૾֦େ ௒ղ૾ XBJGVY 9 ໊ࢗσʔλԽͷཁૉٕज़

Slide 10

Slide 10 text

10 4BOTBOͷ໊ࢗσʔλԽ ਓɺٕज़ɺσʔλϕʔεʹΑΔ໊ࢗͷσʔλԽ

Slide 11

Slide 11 text

ػցֶशͷऔΓ૊Έ 11

Slide 12

Slide 12 text

12 ػցֶशϓϩδΣΫτͷ঺հ %FFQ)FBU.BQ.PEFM
 %FFQ-FBSOJOHʹΑΔը૾෼ׂͷֶश
 %FFQ-BOH%FUFDUPS.PEFM
 ໊ࢗը૾͔Βͷݴޠ൑ఆ
 σʔλԽγεςϜͷ࠷దԽ
 *OTQFDUPS
 ༳ΕͷٵऩͱσʔλϕʔεͷΫϨϯδϯά
 ໊دͤνΣοΫ
 ϛεݕ஌

Slide 13

Slide 13 text

13 %FFQ)FBU.BQ.PEFM ࢀߟz1ZSBNJE4DFOF1BSTJOH/FUXPSL )FOHTIVBOH;IBP +JBOQJOH4IJ 9JBPKVBO2J 9JBPHBOH8BOH +JBZB+JB *&&&$POGFSFODFPO$PNQVUFS7JTJPOBOE1BUUFSO3FDPHOJUJPO $713 IUUQTIT[IBPHJUIVCJPQSPKFDUTQTQOFU %FFQ-FBSOJOHϞσϧ ෼ׂͷ݁ՌΛֶशͤ͞Δ ͭͷϞσϧ
 ໊ࢗ಺ͷ߲໨ۣܗͷਪఆ
 ߲໨໊ͷਪఆ Ϟσϧ୯ମͰ΋ͷਫ਼౓
 طଘͷγεςϜͱ૊Έ߹Θͤ

Slide 14

Slide 14 text

%FFQ-BOH%FUFDUPS.PEFM z+BQBOFTFz ࢀߟz%FFQ3FTJEVBM-FBSOJOHGPS*NBHF3FDPHOJUJPO ,BJNJOH)F 9JBOHZV;IBOH 4IBPRJOH3FO +JBO4VO IUUQTBSYJWPSHBCT ໊͔ࢗΒݴޠΛ൑ఆ
 ೔ຊޠɺӳޠɺதࠃޠɺ೔ຊޠӳޠɺʜ
 ΦϖϨʔλೖྗ৘ใͱϑΟʔυόοΫ͔Βֶश σʔλԽϑϩʔͷ࠷దԽ ɹͦͷݴޠͷυϝΠϯΛ࣋ͭΦϖϨʔλ΁ͷৼΓ෼͚ ɹ0$3౳΁ͷࣄલ৘ใ΁ ֶशγεςϜΛճ͍ͯ͘͠ࣄͰଟݴޠԽʹରԠ

Slide 15

Slide 15 text

15 ଞػցֶशͷऔΓ૊Έ σʔλԽγεςϜͷ࠷దԽ - *OTQFDUPSνΣοΫ ‣ΦϖϨʔλͷೖྗͷϛεΛֶश ࣗવݴޠॲཧ ػցֶश ‣ܾఆ໦ "VUP&ODPEFSͳͲʮ΋͔ͯ͠͠ɿʯ͕ՄೳͳϞσϧ ‣ਫ਼౓Λ୲อ͠ͳ͕ΒίετΛ࠷దԽ - औΓࠐΈຕ਺ͷ༧ଌ ‣աڈͷձࣾͷऔΓࠐΈ΍ن໛ɺӦۀൺ཰ ‣#PPTUJOHʹΑΔઢܗճؼ HCMJOFS ‣ΦϖϨʔλௐ੔ɺΩϟϯϖʔϯ΍ަবࡐྉ
 ༳ΕٵऩͱσʔλϕʔεΫϨϯδϯά - ໊دͤνΣοΫ ‣اۀ৘ใɺ໊ࢗऔΓࠐΈຕ਺ɺਓ໊ͷ௝͠͞౳Λಛ௃ྔԽ ‣ܾఆ໦ͳͲઆ໌ม਺͕෼͔Γ΍͍͢ϞσϧͰ࡞ΓˠΦϖϨʔλ΁ - ϛεݕ஌ ‣σʔλϕʔεͷೖྗ࣌ظ౳ͷ৘ใΛݩʹϛεݕ஌ ‣ݹ͍γεςϜʹΑΔσʔλΛΫϨϯδϯά 4POTPOגࣜձࣾ ʔʔʔ ŠŠŠ

Slide 16

Slide 16 text

ػցֶशϓϩδΣΫτͷεεϝ 16

Slide 17

Slide 17 text

17 4BOTBOʹ͓͚Δػցֶशج൫ طଘͷߴਫ਼౓σʔλԽՄೳͳγεςϜʹ৐ͤΔ طଘج൫ͱ૊Έ߹Θͤɺൺֱ zػցֶशཁΒͳ͍Ͷzͱ͍͏൑அ΋༰қʹ
 ػցֶशϓϩδΣΫτ͸ܧଓతͳ1%$"͕ඞਢ
 ղੳ݁ՌΛཷΊࠐΉͨΊͷผͷ؀ڥ σʔλϕʔεج൫ͱ෼ੳج൫ ࠷దͳαʔϏεબ୒ͱػցֶशΠϯϑϥ
 3%಺ʹΞʔΩςΫτ5FBN (JU΍%PDLFSͷར༻
 3%಺ͷڞ௨Խ͞Εͨ෼ੳج൫

Slide 18

Slide 18 text

18 ػցֶश༻ͷσʔλ ೖྗ ೖྗ ෼ׂνΣοΫ ࣾ಺ͰͷਓʹΑΔσʔλνΣοΫ
 
 ʼػցֶश༻ͷUSBJO UFTUσʔλνΣοΫ
 ɹػցֶशΤϯδχΞͷ࢓ࣄͷେ൒Λ઎ΊΔσʔλ࡞Γ΋෼ࢄ
 ɹσʔλ͸ੜϞϊ
 σʔλΛ࡞ΔਓɺλΠϛϯάʹΑͬͯ൑அج४͕มԽͯ͠͠·͏
 ɹɹࣾ಺ͷίϛϡχέʔγϣϯͰυϝΠϯΛڞ༗Ͱ͖Δ
 ɹɹλεΫ΍σʔλɺλΠϛϯάʹԠͨ͡ܗࣜ
 ɹϓϩδΣΫτʹΑͬͯ਺ेສʙ਺ඦສͷֶशσʔλ
 
 ʼ%#಺ͷաڈσʔλνΣοΫ
 ɹ(BSCBHFJO (BSCBHFPVU
 ɹաڈͷσʔλ͕៉ྷʹͳ͍ͬͯ͘ࣄͰ
 ɹػցֶशϓϩδΣΫτͷεϐʔυ͸ܶతʹ޲্͢Δ * Garbage in, Garbage Out: How Purportedly Great Machine Learning Models Can Be Screwed Up By Bad Data [2017, Black Hat USA, Hillary Sanders] ໨ࢹνΣοΫ

Slide 19

Slide 19 text

ਓɺٕज़ɺσʔλϕʔεʹΑΔ໊ࢗͷσʔλԽ 19 ػցֶशϓϩδΣΫτͷαΠΫϧ ը૾ೝࣝ ػցֶश AWS ໨ࢹνΣοΫ ֶशσʔλ࡞Γ ΫϨϯδϯά ࣙॻ Ϛονϯά

Slide 20

Slide 20 text

20 ͦͯ͠σʔλαΠΤϯε΁ ໾৬มԽݕ஌ɺϦίϝϯυ اۀͷΫϥελϦϯάɺۀछ ਓɺձࣾͷΩʔϫʔυͷநग़ ޿ࠂ഑৴γεςϜͷ࠷దԽ ϑΟʔυͷ࠷దԽ ࣍ձ͏΂͖ਓɺձࣾɺ෦ॺͷϦίϝϯυ ΩʔϚϯͷϦίϝϯυ
 ɹɹɹɹɹɹɹɹɹɹɹʜ FUD

Slide 21

Slide 21 text

͓ΘΓʹ 21

Slide 22

Slide 22 text

22 4BOTBOͱ໊ࢗͷσʔλԽͱػցֶशͷऔΓ૊ΈΛૡ͍ఠΜͰ঺հ ػցֶशΛϓϩδΣΫτͰ࢖͏ͨΊʹඞཁͳ΋ͷ
 ५୔ͳ؀ڥ
 େྔͷ៉ྷͳσʔλ
 ૬Ԡͷج൫ٕज़ͱ%#ͱΠϯϑϥ
 ࣾ಺ͷཧղ ༏लͳ3%ϝϯόʔɺڞಉݚڀઌΛืूத
 ɹ५୔ͳσʔλɺ؀ڥɺֶձࢀՃ΍ొஃɺ,BHHMF෦ɺษڧձʜFUD
 ɹ,BHHMF(SBOE.BTUFS .BTUFS 0QFO$74IBSQ࡞ऀͳͲ໊͕ࡏ੶ ͓ΘΓʹ

Slide 23

Slide 23 text

No content