Tweet
Tweet

Slide 1

Slide 1 text

େࡕެཱେֶɹ২ాେथ AI࠷৽࿦จಡΈձ 2022೥1೥·ͱΊ

Slide 2

Slide 2 text

2022೥·ͱΊ AI࠷৽࿦จಡΈձ ɾϝΠϯ ConvNeXt (2݄ൃද): ࢖͍΍͍͢࠷ۙͷߴੑೳϞσϧ GLIDE (1݄ൃද)ɹςΩετtoը૾ੜ੒ Imagic (11݄ൃද)ɹࡉ΍͔ͳमਖ਼ AudioLM (10݄ൃද): Ի੠ੜ੒Ϟσϧ Socratic Models (5݄ൃද): ൚༻AI (AGI) ɾͦͷଞ Wav2Vec 2 (7݄ൃද): NeuroAI?Brain-inspired AI (AIͱਓؒͷ೴ͷؔ܎Λ୳Δ෼໺) Algorithmic Imprint (7݄ൃද): AI࡞੒ऀͷྙཧ

Slide 3

Slide 3 text

2022೥·ͱΊ Text GLIDE, Imagic AGI ࣗવݴޠॲཧΛج൫ͱͨ͠൚ਓ޻஌ೳ(AGI)΁ͷνϟϨϯδͷ1೥ɻ CV ConvNeXt AudioLM Speech Socratic model Diffusion

Slide 4

Slide 4 text

2022೥·Ͱ·ͱΊ Self-Attention 2017೥ 2018೥ BERT 2020೥ DETR ViT GPT3 2021೥ CLIP wav2vec 2 w2v-BERT BigSSL 2019೥ GPT2 SwinT DDPM ADM

Slide 5

Slide 5 text

2022೥·ͱΊ Text GLIDE, Imagic AGI ࣗવݴޠॲཧΛج൫ͱͨ͠൚ਓ޻஌ೳ(AGI)΁ͷνϟϨϯδͷ1೥ɻ CV AudioLM Speech Socratic model Diffusion ConvNeXt

Slide 6

Slide 6 text

ConvNeXt: CNN x SwinTransformer ը૾෼ྨϞσϧͷstate-of-the-art

Slide 7

Slide 7 text

ConvNeXt: CNN x SwinTransformer ޻෉఺ͷ·ͱΊ: ϕʔε͸ResNet

Slide 8

Slide 8 text

ConvNeXt ·ͣॳΊʹɻ

Slide 9

Slide 9 text

ConvNeXt ֤εςʔδͷ܁Γฦ͠਺ΛSwinTʹ͚ۙͮΔ

Slide 10

Slide 10 text

ConvNeXt 4×4 non-overlapping convolution ৞ΈࠐΈͷύονԽ

Slide 11

Slide 11 text

ConvNeXt Depthwise convolutionಋೖޙɺ෯Λ޿͛Δ

Slide 12

Slide 12 text

ConvNeXt Inverted bottleneck(Narrow→Wide→Narrow)ߏ଄ͷಋೖ TransformerͰ͸֦େ཰4ഒΛ࢖༻ɻ※MobileNetͰ͸֦େ཰͸6ഒɻ શମͱͯ͠ͷܭࢉྔ͸ݮΔ͕ɺConvͷԋࢉ͸૿Ճɻ SwinTͰ͸ίί

Slide 13

Slide 13 text

ConvNeXt Depthwise convolutionͷҠಈ ※Depthwise ConvolutionͰେ͖ͳΧʔωϧαΠζ࢖͏ͨΊ Ұ࣌తʹConvͷԋࢉྔݮগͰੑೳѱԽɻ SwinTͰ͸ίί MSAϒϩοΫ͕FFNΑΓ΋ઌ಄ʹ͋Δ

Slide 14

Slide 14 text

ConvNeXt SwinTransformerͷΧʔωϧαΠζ(7)ΛਅࣅΔ Depthwise convolutionͷ ΧʔωϧαΠζେ͖͍ͯ͘͘͠ɻ 7Ͱੑೳ͕๞࿨(SwinTͱಉ͡) ↓

Slide 15

Slide 15 text

ConvNeXt ࡉ͔ͳSwinT or ViTͷ޻෉Λಋೖ ReLU→GELU NormalizationݮΒ͢ BN→LN μ΢ϯαϯϓϧ૚Λ੾Γ཭͠

Slide 16

Slide 16 text

ConvNeXt ݁Ռ

Slide 17

Slide 17 text

ConvNeXt ResNetΛSwinTransformerԽͯ͠ɺ CNN͚ͩͰState-of-the-artग़ͨΑɻ

Slide 18

Slide 18 text

No content

Slide 19

Slide 19 text

2022೥·ͱΊ Text AGI ࣗવݴޠॲཧΛج൫ͱͨ͠൚ਓ޻஌ೳ(AGI)΁ͷνϟϨϯδͷ1೥ɻ CV ConvNeXt AudioLM Speech Socratic model Diffusion GLIDE, Imagic

Slide 20

Slide 20 text

GLIDE Stable Di ff usionͷجૅϞσϧ

Slide 21

Slide 21 text

Diffusion model ੜ੒Ϟσϧ

Slide 22

Slide 22 text

Diffusion modelͷྺ࢙ DDPM ADM GLIDE CLIP ↓ ҆ఆԽɺߴղ૾౓Խ ݴޠΛѻ͏

Slide 23

Slide 23 text

Diffusion modelͷྺ࢙ DDPM ADM GLIDE CLIP ↓

Slide 24

Slide 24 text

DDPM: diffusion modelͷ࢝·Γ DNN Image Noise Image 
 + 
 Noise ਪ࿦ͨ͠ Noise ࣌ࠁ৘ใ ೋ৐ޡࠩ ࠷খԽ

Slide 25

Slide 25 text

Diffusion modelͷྺ࢙ DDPM ADM GLIDE CLIP ↓

Slide 26

Slide 26 text

ADM: ϞσϧΛ2ͭʹ෼͚ͯɺߴղ૾౓Խʹ੒ޭɻ Base Upsampler ෼ྨ ߴղ૾ Classi fi er guidance (CNN)

Slide 27

Slide 27 text

GLIDE = CLIP x Diffusion model Di ff usion modelͷྺ࢙ DDPM ADM GLIDE CLIP ↓

Slide 28

Slide 28 text

CLIP: ը૾ͱςΩετͷڮ౉͠

Slide 29

Slide 29 text

CLIP: ը૾ͱςΩετͷڮ౉͠ ը૾ͱςΩετΛ௚઀ൺֱͰ͖ΔΑ͏ʹಛ௃ม׵Ͱ͖ΔϞσϧ ViT: Image Transformer: Text ίαΠϯྨࣅ౓

Slide 30

Slide 30 text

ADM Base Upsampler ෼ྨ ߴղ૾ Classi fi er guidance (CNN)

Slide 31

Slide 31 text

GLIDE = ADM-basedʹCNNΛCLIPʹมߋ ADM-basedʹCNNΛCLIPʹมߋ Base Upsampler ෼ྨ ߴղ૾ Classi fi er guidance (CLIP)

Slide 32

Slide 32 text

Imagic: Stable DiffusionͷվྑςΫχοΫ Stable Di ff usionͷվྑςΫχοΫ

Slide 33

Slide 33 text

Imagic Overview

Slide 34

Slide 34 text

No content

Slide 35

Slide 35 text

2022೥·ͱΊ Text GLIDE, Imagic AGI ࣗવݴޠॲཧΛج൫ͱͨ͠൚ਓ޻஌ೳ(AGI)΁ͷνϟϨϯδͷ1೥ɻ CV ConvNeXt Speech Socratic model Diffusion AudioLM

Slide 36

Slide 36 text

AudioLM Իͷੜ੒Ϟσϧ

Slide 37

Slide 37 text

AudioLM = w2v-BERT x SoundStream Overview ɾจষͱΦʔσΟΦͷؒʹ͸Ұରଟͷؔ܎͕͋Δɻ ɾΦʔσΟΦ͸ςΩετʹൺͯ͠σʔλྔ͕ଟ͍ɻ

Slide 38

Slide 38 text

SoundStream ԻΛྔࢠԽ͢Δ

Slide 39

Slide 39 text

w2v-BERT Contrastive LearningͱMasked Language Modelingͷ૊Έ߹Θͤ

Slide 40

Slide 40 text

No content

Slide 41

Slide 41 text

2022೥·ͱΊ Text GLIDE, Imagic AGI ࣗવݴޠॲཧΛج൫ͱͨ͠൚ਓ޻஌ೳ(AGI)΁ͷνϟϨϯδͷ1೥ɻ CV ConvNeXt AudioLM Speech Diffusion Socratic model

Slide 42

Slide 42 text

Socratic models طଘֶशࡁΈϞσϧΛ૊Έ߹Θͤͨ(४ʁ)൚ਓ޻஌ೳϞσϧ

Slide 43

Slide 43 text

Socratic models Overview Language is an intermediate representation

Slide 44

Slide 44 text

Socratic models Overview طଘͷVLM (Visual Language Model)ɺLMs (Large Language Model) ɺ ALMs (Audio Language Model)ͷಉ࢜ が ɺߏ଄Խ͞Εͨର࿩Λߦ͏ɻ ͦͯ͠ɺ ビデ ΦαʔνɺΩϟ プ γϣϯੜ੒ɺ ビデ ΦQ&A (ະ஌ͷλεΫ)ɺকདྷͷߦಈ༧ଌΛ͜ͷର࿩ۭؒ΁ͷ৽͍͠ࢀՃऀͱͯ͠ѻ͏ ɻ

Slide 45

Slide 45 text

Socratic models ྫࣔ̍ɿجຊฤ

Slide 46

Slide 46 text

Socratic models ྫࣔ̎ɿԠ༻ฤ

Slide 47

Slide 47 text

Socratic models ιΫϥςεର࿩ͱ͸ʁ

Slide 48

Slide 48 text

No content

Slide 49

Slide 49 text

Others: NeuroAIᶃ ೴ͷػೳͱݴޠϞσϧͷରԠΛ୳Δ

Slide 50

Slide 50 text

Others: NeuroAIᶃ શମ૾: Wav2Vec 2Λֶश͠ɺͦͷ݁Ռ͔ΒfMRIͷBOLDΛ༧ଌ͢ΔWΛ࡞੒ɾ݁Ռݕূ

Slide 51

Slide 51 text

Others: NeuroAIᶃ ฏۉԽͨ͠೴ͷ׆ੑͷදݱɻ

Slide 52

Slide 52 text

Others: NeuroAIᶃ ϞσϧͷϨΠϠʔͷਂ͞ͱ೴ͷ෦ҐʹରԠ͕͋ͬͨɻ

Slide 53

Slide 53 text

Others: NeuroAIᶄ ೴೾͔ΒݴޠΛੜ੒͢Δ

Slide 54

Slide 54 text

Others: NeuroAIᶄ ϞσϧͷτϨʔχϯάηογϣϯ 81िؒʹΘͨΓ50ճͷηογϣϯ ݽཱޠλεΫͱจষλεΫ λʔήοτͷ୯ޠ΍จষ͕ը໘্ͷจࣈͱͯ͠ ඃݧऀʹࢹ֮తʹఏࣔ͞Εඃݧऀ͸ ͦͷ୯ޠ΍จষΛੜ੒͠Α͏ͱͨ͠ɻ ݽཱޠλεΫͰ͸ɺ50ݸͷӳ୯ޠηοτ͔Βݸʑͷ୯ޠΛੜ੒ɻ จষλεΫͰ͸ɺ50୯ޠηοτ͔ΒͳΔӳޠจ͔Β୯ޠྻΛੜ੒ɻ

Slide 55

Slide 55 text

Others: NeuroAIᶄ Ϟσϧͷ݁Ռ จষ͸75%ͷਫ਼౓ ୯ޠ͸93%ͷਫ਼౓

Slide 56

Slide 56 text

No content

Slide 57

Slide 57 text

Others: AI Ethics ྙཧ

Slide 58

Slide 58 text

Algorithmic Imprint Ξϧ ゴ Ϧ ズ ϜʹΑΔ֐ が ൃੜͨ͠৔߹ͷҰൠత で ߹ཧతͳରࡦͱͯ͠ɺͦͷ༗֐ͳӨڹ が ͞Βʹ఻ൖ͢ΔͷΛ๷ ぐ ͨΊʹ Ξϧ ゴ Ϧ ズ Ϝͷ࢖༻ఀࢭ が Α͘ߦΘΕΔ が ɺఀࢭ͔ͨ͠Βͱݴͬͯެฏੑɺઆ໌੹೚ɺಁ໌ੑɺྙཧͷ໰୊ が ͳ͘ͳΔ Θ͚ で ͸ͳ͍ →͜ͷ༗֐ͳΞϧ ゴ Ϧ ズ ϜͷӨڹ͸ɺΞϧ ゴ Ϧ ズ Ϝ࡟আҎ߱΋௕͘Өڹ͠ଓ͚Δ(Ξϧ ゴ Ϧ ズ Ϝͷࠟ੻) ྫ: ӳࠃΛڌ఺ͱ͢Δߴߍͷଔۀূॻࢼݧ で ͋ΔGCEࢼݧͷΞϧ ゴ Ϧ ズ ϜʹΑΔධՁΛऔΓר͘໰୊(2020) ■ ど ͷΑ͏ͳࢼݧ͔? ɾ 160͔ࠃҎ্ で ࣮ࢪ͞Ε͍ͯΔ(ͦͷଟ͘͸ӳࠃͷݩ২ຽ஍)ࠃࡍతʹೝΊΒΕͨࢼݧ ɾ AϨ ベ ϧͷ੒੷͸ඞવత で ͋Γɺେֶ΁ͷೖֶʹෆՄܽͳ໾ׂΛՌͨ͢ ■ܦҢ ɾCOVID-19ͷେྲྀߦʹΑΓGCEࢼݧΛ؂ಜ͢ΔӳࠃʹຊڌΛஔ͘४੓෎ػؔ で ͋ΔOfqual͸ର໘ࢼݧΛதࢭͨ͠ ɾࢼݧͷ୅ΘΓʹɺֶߍ で ͷੜెͷաڈͷ੒੷ɺڭࢣͷධՁΛ࢖༻ͯ͠Ξϧ ゴ Ϧ ズ Ϝ で ੒੷Λ࡞੒ͨ͠ →݁Ռɺੈքతͳ߅ٞߦಈ が ຄൃ͠ɺΞϧ ゴ Ϧ ズ Ϝ͸࡟আ͞Εͨ ɹڭࢣଆ: ͦ΋ͦ΋աڈͷੜెͷධՁΛه࿥͍ͯ͠ͳ͍ ɹੜెଆ: ੒੷ʹରͯ͠ਅ݋ʹऔΓ૊Μ で ͍ͳ͔ͬͨ(ࢼݧ が શͯͳͷ で ௚લͷ30~60೔ʹ໠ษڧ͢Δੜె が ଟ͍) ɾΞϧ ゴ Ϧ ズ Ϝ͸࡟আ͞Εͨ が ɺֶੜͷ࠶ධՁ͸ߦΘΕͳ͔ͬͨɻ ͢ͳΘͪɺ࠾఺ํ๏͸มΘͬͨ が ɺΞϧ ゴ Ϧ ズ ϜͷӨڹΛେ͖͘ड͚͍ͯͨ(Ξϧ ゴ Ϧ ズ Ϝͷࠟ੻)

Slide 59

Slide 59 text

Algorithmic Imprint ■Algorithmic Imprint(Ξϧ ゴ Ϧ ズ Ϝͷࠟ੻)Λҙࣝͨ͠Ξϧ ゴ Ϧ ズ Ϝͷઃܭ ʮΞϧ ゴ Ϧ ズ Ϝͷࠟ੻ʯΛҙࣝͨ͠ઃܭͷߟ͑ํʹΑΓɺΞϧ ゴ Ϧ ズ Ϝ։ൃ プ ϩηεΛΑΓެฏ で ࣾձٕज़తͳ ৘ใʹج づ ͍ͨ΋ͷʹ͢Δ͜ͱ がで ͖Δɻ (1)Ξϧ ゴ Ϧ ズ ϜͷӨڹ Ξϧ ゴ Ϧ ズ Ϝ͸࡟আͨ͠ޙʹ΋ར֐ؔ܎ऀʹӨڹΛٴ ぼ ͢ɻ։ൃऀͱӡӦऀ͸Ξϧ ゴ Ϧ ズ ϜΛ࡟আ ͢Δ だ ͚ で ͳ͘ɺΞϧ ゴ Ϧ ズ ϜʹΑΔة֐Λੋਖ਼͠ɺઆ໌੹೚ が ࣋ଓͯ͠ཁٻ͞ΕΔɻ (2)Ξϧ ゴ Ϧ ズ Ϝઃܭͷઆ໌੹೚ ։ൃऀ͸ʮΞϧ ゴ Ϧ ズ Ϝͷࠟ੻ʯͷӨڹΛड͚Δਓʹ֐ΛΑΓೝࣝ で ͖ΔΑ͏ʹ͢Δ べ ͖ で ͋Δɻ (3)AIྙཧ ガ バ φϯε で ิڧ͢Δ
 ٕज़తͳհೖ だ ͚ で ͸֐Λ࡟ݮ͢Δ͜ͱ͸ で ͖ͳ͍ɻ ʮΞϧ ゴ Ϧ ズ Ϝͷࠟ੻ʯΛҙࣝͨ͠Ξϧ ゴ Ϧ ズ ϜઃܭΛ ద੾ͳAI ྙཧ ガ バ φϯε で ิ׬͢Δɻ

Slide 60

Slide 60 text

No content

Slide 61

Slide 61 text

2023೥ʹ͍ͭͯ ʮզʑͷݚڀࣨʹ͔͠Ͱ͖ͳ͍͜ͱʯΛɻ Ҿ͖ଓ͖ษڧձ։࠵͢Δɻ ҩֶ΁ͷൺॏΛॏ͘͢Δɻ ҩྍը૾ݚڀ༻ϞσϧͷνϡʔτϦΞϧɾϋϯζΦϯ