成長を止めない機械学習のやり方 / Don't stop 'til you get enough (data).

੒௕ΛࢭΊͳ͍  ػցֶशͷ΍Γํ ΫοΫύουגࣜձࣾɹછ୩༔Ұ࿠ ."/"#*:"

ࣗݾ঺հ છ୩༔Ұ࿠<:VJDIJSP4PNFZB> ౦ژ޻ۀେֶେֶӃܭࢉ޻ֶम࢜ ΫοΫύουגࣜձࣾݚڀ։ൃ෦  ϦαʔνΤϯδχΞ೥݄d ػցֶशج൫
Ϩγϐσʔλͷ෼ੳ౳ UXJUUFSDPN!BZFNPT@Z HJUIVCDPNBZFNPT IUUQTXXXBZFNPTNF

Ϩγϐ਺ɿສ඼Ҏ্ ࠃ಺ͷ݄ؒར༻ऀ਺ɿ໿ສਓ

ରԠݴޠɿݴޠϲࠃ ւ֎ͷ݄ؒར༻ऀ਺ɿສਓҎ্

ΫοΫύουݚڀ։ൃ෦ ೥݄ʹൃ଍ ࢲ͕഑ଐ͞Εͨͷ΋ಉ࣌ظ ౰࣌͸໊ͷϝϯόʔ ೥݄ݱࡏ͸ࠃ಺ʹ໊ ւ֎ʹ໊

ྉཧ͖Ζ͘ ৞ΈࠐΈχϡʔϥϧωοτϫʔΫ ʹΑΔྉཧը૾ͷࣗಈೝࣝ εϚʔτϑΥϯͷࣸਅͷ಺ɺ ྉཧࣸਅΛࣗಈతʹऩूه࿥

ྉཧ͖Ζ͘ ສਓҎ্ͷϢʔβʔ͕ར༻ ສຕҎ্ͷྉཧࣸਅΛه࿥ ೥݄೔ݱࡏ

ͦͷଞػցֶशϓϩδΣΫτ ը૾෼ੳʹΑΔϨγϐͷ෼ྨ ࡐྉ໊ͷਖ਼نԽ <͓͠ΐ͏Ώ ে༉ γϣʔϢ><͠ΐ͏Ώ>
ϑΟʔυόοΫͷࣗಈλά෇͚  FUD

લஔ͖

l લུ ʮ࣭໰ʯ·Ͱ͸͍͔ͳ͍͕ؾʹͳ͍ͬͯΔ͜ͱɺ  ʮճ౴ʯΛ஌Γ͍͕ͨͳ͔ͳ͔஌Δػձ͕ͳ͔ͬͨ͜ͱɻ    ͦ͏͍ͬͨٙ໰Λ೔ຊத͔ΒूΊɺ  ΧϯϑΝϨϯεΛ௨ͯ͡ղܾࡦΛݟग़͠ɺ஌ܙΛੜΈग़͢ɻ  ͜Ε͕ɺ."/"#*:"Λ։࠵͢Δ૝͍Ͱ͢ɻz IUUQTNBOBCJZBUFDI

l લུ ʮ࣭໰ʯ·Ͱ͸͍͔ͳ͍͕ؾʹͳ͍ͬͯΔ͜ͱɺ  ʮճ౴ʯΛ஌Γ͍͕ͨͳ͔ͳ͔஌Δػձ͕ͳ͔ͬͨ͜ͱɻ    ͦ͏͍ͬͨٙ໰Λ೔ຊத͔ΒूΊɺ  ΧϯϑΝϨϯεΛ௨ͯ͡ղܾࡦΛݟग़͠ɺ஌ܙΛੜΈग़͢ɻ  ͜Ε͕ɺ."/"#*:"Λ։࠵͢Δ૝͍Ͱ͢ɻz IUUQTNBOBCJZBUFDI ՝୊ͷڞ༗
ղܾࡦͷྻڍͱ෼ੳ

lຊηογϣϯͰ͸·ͣػցֶशΛऔΓೖΕͨ૊৫ʹ͓͚Δ  Ձ஋૑଄Λ્֐͋Δ͍͸ಷԽͤ͞Δز͔ͭͷཁҼΛࢦఠ͠ɺ  ͦΕΒͷཁҼʹ͍ͭͯ։ൃج൫΍ΤϯδχΞϦϯάɺ  νʔϜߏ੒౳༷ʑͳ֯౓͔Βߟ࡯͠·͢ɻ·ͨ͞Βʹɺ  ͦͷΑ͏ͳ՝୊ʹର͢ΔΫοΫύουͷऔΓ૊ΈΛ঺հ͠·͢ɻz

lຊηογϣϯͰ͸·ͣػցֶशΛऔΓೖΕͨ૊৫ʹ͓͚Δ  Ձ஋૑଄Λ્֐͋Δ͍͸ಷԽͤ͞Δز͔ͭͷཁҼΛࢦఠ͠ɺ  ͦΕΒͷཁҼʹ͍ͭͯ։ൃج൫΍ΤϯδχΞϦϯάɺ  νʔϜߏ੒౳༷ʑͳ֯౓͔Βߟ࡯͠·͢ɻ·ͨ͞Βʹɺ  ͦͷΑ͏ͳ՝୊ʹର͢ΔΫοΫύουͷऔΓ૊ΈΛ঺հ͠·͢ɻz ՝୊ͷڞ༗ ղܾࡦͷྻڍͱ෼ੳ

ʰ੒௕ΛࢭΊͳ͍ػցֶशͷ΍Γํʱ ػցֶशʹऔΓ૊Ήݱ৔ʹ͓͚Δ՝୊Λࢦఠ͠ɺ෼ੳ͠·͢ɻ

ΞδΣϯμ ͸͡Ίʹਓ޻஌ೳϒʔϜͱݬ໓ظ ੒௕ΛࢭΊͳ͍ػցֶशͷ΍Γํʙ૊৫ɾνʔϜฤʙ αʔϏε։ൃͱػցֶश ػցֶशϓϩδΣΫτ
੒௕ΛࢭΊͳ͍ػցֶशͷ΍Γํʙݸਓฤʙ ૯߹֨ಆٕͱͯ͠ͷػցֶश

ਓ޻஌ೳϒʔϜͱݬ໓ظ

ਓ޻஌ೳϒʔϜ ʰୈ࣍ਓ޻஌ೳϒʔϜʱ ը૾෼ྨ෼໺ʹ͓͚Δਂ૚ֶशͷ੒ՌΛ͖͔͚ͬͱ͢Δ ਓ޻஌ೳ࣮ݱͷखஈͷ̍ͭͱͯ͠ػցֶश΁ͷ౤ࢿʗظ଴΋ߴ·Δ ϋʔυ΢ΣΞʗιϑτ΢ΣΞʹ͔͔ΘΒͣ͋ΒΏΔྖҬʹԠ༻͞ΕΑ ͏ͱ͍ͯ͠Δ
ΫοΫύουͷ༷ͳαʔϏε։ൃͷݱ৔΋ྫ֎Ͱ͸ͳ͍

IUUQXXXHBSUOFSDPKQQSFTTIUNMQSIUNM

ϒʔϜ͔Βݬ໓ظ΁ "*ʹ͸ظ଴͕ա౓ʹൃ೤ͨ͠ͷͪʰౙʱ͕๚ΕΔͱ͍͏ྺ࢙͕͋Δ ݬ໓ظΛ৐Γӽ͍ͯͨ͘Ίʹݱ৔ͷզʑ͕͢΂͖͜ͱ ʰ੒௕ΛࢭΊͳ͍ػցֶशͷ΍Γํʱ ࠓճ͸ɺαʔϏε։ൃݱ৔ͱͯ͠ͷΫοΫύουͱ͍͏؀ڥͰಘΒΕ ͨ஌ݟΛத৺ʹɺ͜ͷ՝୊ʹ͍ͭͯߟ͍͖͑ͯ·͢

੒௕ΛࢭΊͳ͍ػցֶशͷ΍Γํ ૊৫ɾνʔϜฤ

ΞδΣϯμ ͸͡Ίʹਓ޻஌ೳϒʔϜͱݬ໓ظ ੒௕ΛࢭΊͳ͍ػցֶशͷ΍Γํʙ૊৫ɾνʔϜฤʙ αʔϏε։ൃͱػցֶश ػցֶशϓϩδΣΫτ
੒௕ΛࢭΊͳ͍ػցֶशͷ΍Γํʙݸਓฤʙ ૯߹֨ಆٕͱͯ͠ͷػցֶश ͸͡Ίʹਓ޻஌ೳϒʔϜͱݬ໓ظ ੒௕ΛࢭΊͳ͍ػցֶशͷ΍Γํʙ૊৫ɾνʔϜฤʙ αʔϏε։ൃͱػցֶश ػցֶशϓϩδΣΫτ ੒௕ΛࢭΊͳ͍ػցֶशͷ΍Γํʙݸਓฤʙ ૯߹֨ಆٕͱͯ͠ͷػցֶश

αʔϏε։ൃͱػցֶश αʔϏε։ൃͰػցֶशΛ࢖͍͍ͨͱ͍͏χʔζ͕૿͍͑ͯΔ ਓ޻஌ೳϒʔϜʹґΔͱ͜Ζ͕େ͖͍ ଟ෼ ʰػցֶशΛ࢖͏ʱͱ͸ʁ

՝୊ൃݟ  ΰʔϧઃఆ σʔλऩू σʔλՃ޻  ෼ੳ &%" ֶश ධՁݕূ ຊ൪ ධՁݕূ

࢖͏ ࡞Δ ՝୊ൃݟ  ΰʔϧઃఆ σʔλऩू σʔλՃ޻  ෼ੳ &%" ֶश ධՁݕূ
ຊ൪ ධՁݕূ σ ϓ ϩ Π

ྫϞσϧͷ"1*Խʹඞཁͳ΋ͷ ֶशࡁΈϞσϧ ͷEVNQ NPEFM\QC I ^ "1* BQQQZ  
)551 T &OEQPJOU QSFEJDU JOGPͱ͔   ϞσϧͷಡΈࠐΈͱ*0ϋϯυϥ ࡞ۀ طଘγεςϜʗΠϯϑϥ΁ͷΠϯςάϨʔγϣϯ  -#ɺϞχλϦϯάɺ%/4౳ʑ ࣮ߦ؀ڥ %PDLFSpMF

ྫϞσϧͷ"1*Խʹඞཁͳ΋ͷ ֶशࡁΈϞσϧ ͷEVNQ NPEFM\QC I ^ "1* BQQQZ  
)551 T &OEQPJOU QSFEJDU JOGPͱ͔   ϞσϧͷಡΈࠐΈͱ*0ϋϯυϥ ࡞ۀ طଘγεςϜʗΠϯϑϥ΁ͷΠϯςάϨʔγϣϯ  -#ɺϞχλϦϯάɺ%/4౳ʑ ࣮ߦ؀ڥ %PDLFSpMF ػցֶशϞσϧͷσϓϩΠʹ͸  खؒͱ͕͔͔࣌ؒΔ

def is_food_image(self, image_io): image = Image.open(image_io) result = self.classifier.infer(image) return
result == Category.food ར༻ྫྉཧ͖Ζ͘ͷ৔߹ ύοέʔδԽ͞ΕͨػցֶशϞσϧͷݺͼग़͠  ͋Δ͍͸҉໧ͳ"1*ίʔϧ

def is_food_image(self, image_io): image = Image.open(image_io) result = self.classifier.infer(image) return
result == Category.food ྫྉཧ͖Ζ͘ͷ৔߹ ύοέʔδԽ͞ΕͨػցֶशϞσϧͷݺͼग़͠  ͋Δ͍͸҉໧ͳ"1*ίʔϧ έʔεʹΑΔ͕ αʔϏε։ൃͱ͍͏จ຺ʹ͓͍ͯ  ػցֶशϞσϧ͸ίʔυͰ࣮૷͞ΕͨϩδοΫ  ͱಉ౳ʹѻΘΕ͏Δ

࢖͏ ࡞Δ Ϩγϐ෼ྨثͷΧςΰϦʹ  ʰຑം౾෗ʱΛ଍ͯ͠Έ͍ͨ िؒ͘Β͍͔͔Δͳ Ϛδ͔ʂ if category == 'ΧϨʔ':
foo() + elif category == 'ຑം౾෗': + bar() ՝୊ൃݟ  ΰʔϧઃఆ σʔλऩू σʔλՃ޻  ෼ੳ &%" ֶश ධՁݕূ ຊ൪ ධՁݕূ

αʔϏε։ൃͱϞσϧͷ։ൃʗσϓϩΠ ͷεϐʔυײʹΪϟοϓ͕͋Δ

Ͳ͏͢Ε͹ʁ ਓࡐ഑ஔͷ࠷దԽ

࢖͏ ࡞Δ Ϩγϐ෼ྨثͷΧςΰϦʹ  ʰຑം౾෗ʱΛ଍ͯ͠Έ͍ͨ िؒ͘Β͍͔͔Δͳ Ϛδ͔ʂ if category == 'ΧϨʔ':
foo() + elif category == 'ຑം౾෗': + bar() ՝୊ൃݟ  ΰʔϧઃఆ σʔλऩू σʔλՃ޻  ෼ੳ &%" ֶश ධՁݕূ ຊ൪ ධՁݕূ

ਓࡐ഑ஔͷ࠷దԽ αʔϏε։ൃʹػցֶशΛԠ༻͢Δ͜ͱʹෆ׳Εͳݱ৔ͷ৔߹ αʔϏε։ൃʹٻΊΒΕΔ։ൃεϐʔυͱ  ػցֶशϞσϧͷ։ൃͱσϓϩΠʹ͔͔Δ࣌ؒ ྆ऀʹΪϟοϓ͕͋Δɺͱ͍͏໰୊Λڞ༗ͯ͠าΈدΔ
ྫ͑͹ਓతϦιʔεͷ഑ஔΛݟ௚ͯ͠ΈΔ ྫɿαʔϏε։ൃଆͷΤϯδχΞ͕ϞσϧͷσϓϩΠΛख఻͏

Ͳ͏͢Ε͹ʁ ਓࡐ഑ஔͷ࠷దԽ औΓ૊Ή໰୊ͷݟ௚͠ Ϟσϧ։ൃʗσϓϩΠͷޮ཰Խ

Ͳ͏͢Ε͹ʁ ਓࡐ഑ஔͷ࠷దԽ औΓ૊Ή໰୊ͷݟ௚͠ Ϟσϧ։ൃʗσϓϩΠͷޮ཰Խ ։ൃεϐʔυͷҧ͍΍ػցֶशಛ༗ͷίετΛՃຯ্ͨ͠Ͱɺ ػցֶशΛར༻ͯ͠औΓ૊Ή΂͖ͳͷ͔ߟ͑ͳ͓͢ IUUQTSFTFBSDIHPPHMFDPNQVCTQVCIUNM
ʰػցֶश͸ٕज़తෛ࠴ͷߴརି͠ ҙ༁ ʱ

Ͳ͏͢Ε͹ʁ ਓࡐ഑ஔͷ࠷దԽ औΓ૊Ή໰୊ͷݟ௚͠ Ϟσϧ։ൃʗσϓϩΠͷޮ཰Խ

Ϟσϧ։ൃʗσϓϩΠͷޮ཰Խ ։ൃσʔλͷ෼ੳɺ࣮ݧɺϞσϧͷֶश σϓϩΠύοέʔδԽɺ"1*Խ

ػցֶशج൫ ػցֶशʹؔΘΔ࣮ݧ΍։ൃͷεϐʔυΛ্͛Δҝͷج൫ ΫοΫύουݚڀ։ൃ෦͸෦ࣗ਎Ͱੵۃతʹվળ͍ͯ͠Δ IUUQTTQFBLFSEFDLDPNBZFNPTNBDIJOF MFBSOJOHQMBUGPSNBUDPPLQBE IUUQTTQFBLFSEFDLDPNBZFNPTBDDFMFSBUF NBDIJOFMFBSOJOHXJUIBXT

ΦϯσϚϯυ(16࣮ݧ؀ڥ

ΦϯσϚϯυ(16࣮ݧ؀ڥ ւ֎΋ؚΊ໊ͷϝϯόʔ͕͍Δ ෳ਺ͷϓϩδΣΫτʹෳ਺ͷਓ͕औΓ૊Ή ࣮ݧ༻ͷܭࢉػ؀ڥΛͲͷΑ͏ʹ༻ҙ͢Δ͔ $IBUCPUܦ༝ͰىಈఀࢭՄೳͳΦϯσϚϯυ(16࣮ݧ؀ڥ

ʰ࣮ݧ༻Πϯελϯε࡞ͬͯʱ ઃఆࡁΈΠϯελϯεͷ  εφοϓγϣοτ ".* ATTIBZFNPTXPSLCFODIEOTDPNQBOZA $SFBUF

ʰ࣮ݧ༻Πϯελϯε࡞ͬͯʱ ઃఆࡁΈΠϯελϯεͷ  εφοϓγϣοτ ".* ATTIBZFNPTXPSLCFODIEOTDPNQBOZA $SFBUF IUUQTTQFBLFSEFDLDPNBZFNPTNBDIJOF
MFBSOJOHQMBUGPSNBUDPPLQBE ৄ͘͠͸

ϓϩδΣΫτςϯϓϨʔτ ʹΑΔ࣮ݧ࠶ݱੑͷ޲্

ྫͷϓϩδΣΫτͷ  ௥࣮ݧ͕͍ͨ͠ OPUFCPPLͱεΫϦϓτ͸ϨϙδτϦʹɺ σʔλ͸ࠓ͸खݩʹ͔͠ͳ͍΍ʜ ΍͹͍

࣮ݧ࠶ݱੑͷ௿Լ ͍͍ͭͭ࠶ݱੑͷͳ͍ঢ়ଶΛ࡞ͬͯ͠·͏ ϫʔΫϑϩʔ؅ཧͱ͔େֻ͔Γͳͷ͸ͪΐͬͱʜ

$PPLJFDVUUFSʹΑΔςϯϓϨʔτԽ ϓϩδΣΫτͷςϯϓϨʔτԽ SVCZʹ͸CVOEMFS͕͋Δ͚Ͳʜ IUUQTHJUIVCDPNBVESFZSDPPLJFDVUUFS 1ZUIPO +JOKB
੡ͷ൚༻ͳϓϩδΣΫτςϯϓϨʔτੜ੒πʔϧ ಠࣗͷςϯϓϨʔτΛ૊Ή͜ͱ͕ग़དྷɺσʔλαΠΤϯε༻ͷ෺΋͋Δ IUUQTHJUIVCDPNESJWFOEBUBDPPLJFDVUUFSEBUBTDJFODF

ESJWFOEBUBDPPLJFDVUUFSEBUBTDJFODF A.BLFpMFA֤छศརUBSHFUͱ͔ઃఆ஋ͱ͔ ASFRVJSFNFOUTUYUAґଘϥΠϒϥϦͷϦετ ֤छط໿σΟϨΫτϦ ATSDEBUBA ATSDNPEFMAͦΕͧΕσʔλͷੜ੒ʗલॲཧ༻ɺ
Ϟσϧͷֶश༻εΫϦϓτΛஔ͘ AEBUBAHJUJHOPSF͞ΕΔɺ.BLFpMFͷUBSHFUͰ4ͱTZOD

ྫͷϓϩδΣΫτͷ  ௥࣮ݧ͕͍ͨ͠ HJUDPQBOZDPNSFTFBSDISFDJQFBOBMZTJT ͜͜ʙ HJUDMPOF NBLFTZOD@EBUB@GSPN@T

EPDLFSTDJFODFDPPLJFDVUUFSEPDLFSTDJFODF IUUQTHJUIVCDPNEPDLFSTDJFODFDPPLJFDVUUFSEPDLFS TDJFODF EPDLFSΛར༻͠ɺϓϩδΣΫτͷ࣮ݱੑΛ͞ΒʹߴΊΔ OPUFCPPLͷ্ཱͪ͛ 1PSUGPSXBSE౳ߦ͏UBSHFU΋

Ϟσϧ։ൃʗσϓϩΠͷޮ཰Խ ։ൃσʔλͷ෼ੳɺ࣮ݧɺϞσϧͷֶश Ϋϥ΢υ؀ڥΛ׆͔ͨ͠ΦϯσϚϯυ(16࣮ݧ؀ڥͷ࣮ݱ ϓϩδΣΫτͷςϯϓϨʔτԽʹΑΔ࣮ݧ࠶ݱੑͷ޲্ σϓϩΠύοέʔδԽɺ"1*Խ

ػցֶशϞσϧͷσϓϩΠ ͱ͸ αʔόʔ΁ͷຒΊࠐΈʗύοέʔδʗ"1*

ػցֶशϞσϧͷσϓϩΠ ͱ͸ αʔόʔ΁ͷຒΊࠐΈʗύοέʔδʗ"1* ϞσϧΛಡΈࠐΜͰɺ֎෦ϦΫΤετʹԠͯ͡ϞσϧΛ  ݺͼग़͢ബ͘খ͞ͳ"1*Λ࣮૷͢Δ

"1* .JDSPTFSWJDF ͕͍͍ཧ༝ ͦͷଞͷख๏ ྫ͑͹ΞϓϦαʔόʔ΁ͷ૊ΈࠐΈ ͱൺ΂ͯ  ػցֶशϞσϧͷσϓϩΠख๏ͱͯ͠ద͍ͯ͠Δཧ༝

"1* .JDSPTFSWJDF ͕͍͍ཧ༝ ػցֶशͱΞϓϦͰҟͳΔ࣮ߦ؀ڥ͕ར༻Ͱ͖Δ ҟͳΔݴޠɺ$16(16 "1*ͷεέʔϧʹΑͬͯߴεϧʔϓοτͷ࣮ݱ͸Մೳ
ػցֶशϞσϧʹΑΔਪ࿦͸ॏ͍ ෼ྨثɺ౳ͷϓϦϛςΟϒͳػೳͷ࠶ར༻ ։ൃʹ͓͚Δ୲౰ൣғͷڥքઢͱͯ͠࠷ద ͩͱࢥ͏

࠶ܝϞσϧͷ"1*Խʹඞཁͳ΋ͷ ֶशࡁΈϞσϧ ͷEVNQ NPEFM\QC I ^ "1* BQQQZ  
)551 T &OEQPJOU QSFEJDU JOGPͱ͔   ϞσϧͷಡΈࠐΈͱ*0ϋϯυϥ ࡞ۀ طଘγεςϜʗΠϯϑϥ΁ͷΠϯςάϨʔγϣϯ  -#ɺϞχλϦϯάɺ%/4౳ʑ ࣮ߦ؀ڥ %PDLFSpMF ػցֶशͱͷυϝΠϯڞ༗౓ େ খ

࠶ܝϞσϧͷ"1*Խʹඞཁͳ΋ͷ ֶशࡁΈϞσϧ ͷEVNQ NPEFM\QC I ^ "1* BQQQZ  
)551 T &OEQPJOU QSFEJDU JOGPͱ͔   ϞσϧͷಡΈࠐΈͱ*0ϋϯυϥ ࡞ۀ طଘγεςϜʗΠϯϑϥ΁ͷΠϯςάϨʔγϣϯ  -#ɺϞχλϦϯάɺ%/4౳ʑ ࣮ߦ؀ڥ %PDLFSpMF ػցֶशͱͷυϝΠϯڞ༗౓ େ খ ͜ͷล͕ಛʹਏ͍

ϚωʔδυͳσϓϩΠج൫ ͋Δ͍͸ࣗ࡞ͷͦΕ Ϟσϧͷ"1*Խɺ࣮ߦ؀ڥͷ੔උͳͲ͸ڞ௨Խ͢Δ༨஍͕͋Δ ڞ௨Խ͞Εͨج൫Λఏڙ͢ΔࣄͰɺσϓϩΠखॱΛ୹ॖग़དྷΔ ৄ͘͠͸ IUUQTTQFBLFSEFDLDPNBZFNPTNBDIJOF MFBSOJOHQMBUGPSNBUDPPLQBE

੒௕ΛࢭΊͳ͍ػցֶशͷ΍Γํ ݸਓฤ

૯߹֨ಆٕͱͯ͠ͷػցֶश ʰػցֶश͕ग़དྷΔΑ͏ʹͳΔʱͱ͸

1ZUIPO %PDLFS TDJLJUMFBSO 5FOTPS'MPX .F$BC 42- ,FSBT ϕΠζ౷ܭ χϡʔϥϧωοτ "84
($1 3 47. LBHHMF

૯߹֨ಆٕͱͯ͠ͷػցֶश ʰػցֶश͕ग़དྷΔΑ͏ʹͳΔʱͱ͸ ϒʔϜ΋ख఻͍ɺ৘ใ͕൙ཞ͍ͯ͠Δ Α͏ʹݟ͑Δ ʰ͋Ε΋͜Ε΋ֶͼͨա͗Δʱ
ݸਓʹͱͬͯɺ໨ඪʗ໨తͷ۩ମԽ͕ॏཁʹͳ͖͍ͬͯͯΔ ྫ͑͹ ʰαʔϏε։ൃʹػցֶशΛར༻͢Δதن໛ dਓ ͷݱ৔Ͱɺཱࣗͯ͠Ձ஋ΛੜΈग़ͤΔΤϯδχΞʹͳΔʱͱ͔

૯߹֨ಆٕͱͯ͠ͷػցֶश Ұํ͔֬ʹݱ৔Ͱ΍Δ΂͖͜ͱ͸ଟ͍ ͔͠΋ݱ৔ʹΑͬͯҟͳΔͷͰɺ ݄ฒΈ͕ͩ ຊͰֶ΂ͳ͍͜ͱ͕ଟ͍ ΍͸Γ࣮຿Ͱػցֶशʹ৮ΕΔͷ͕ۙಓ
σʔλ෼ੳίϯϖ΋ྑ͍ ʰݶΒΕͨظؒͷதͰσʔλΛ෼ੳ͠ɺείΞΛܧଓతʹվળ͢Δʱ  ·͋·࣮͋຿ͬΆ͍

·ͱΊ ʰ੒௕ΛࢭΊͳ͍ػցֶशͷ΍Γํʱ αʔϏεͷ࣮ݱखஈͱͯ͠ͷػցֶशʹର͢Δظ଴͸ӈݞ্͕Γ ݱ৔ͱͯ͠ظ଴ʹԠ͑ଓ͚͍͖͍ͯͨͰ͢Ͷɻ

8FSFIJSJOH IUUQTDPPLQBEKPCT

成長を止めない機械学習のやり方 / Don't stop 'til you get enoug...

成長を止めない機械学習のやり方 / Don't stop 'til you get enough (data).

More Decks by Yuichiro Someya

Other Decks in Programming

Featured

Transcript