Machine Learning の新機能✨ / What's New in Machine Learning

by 神武

Slide 1

Slide 1 text

.BDIJOF-FBSOJOHͷ৽ػೳ✨

Slide 2

Slide 2 text

!LPPPPPUBLF ͡ΜΉ ͓அΓ ͜ͷεϥΠυ͸Ұൠެ։͞Ε͍ͯΔ88%$ͷεϥΠυͷΈҾ༻͍ͯ͠·͢

Slide 3

Slide 3 text

No content

Slide 4

Slide 4 text

ηογϣϯ

Slide 5

Slide 5 text

7JEFP63- IUUQTEFWFMPQFSBQQMFDPNWJEFPTQMBZXXED 7JEFP ৄ͘͠஌Γ͍ͨ࣌͸

Slide 6

Slide 6 text

Summary Create ML Domain API Core ML 3 7JEFP

Slide 7

Slide 7 text

Create ML $SFBUF.- .-ϞσϧΛ4XJGUͰ࡞Δ͜ͱ͕Ͱ͖Δ

Slide 8

Slide 8 text

Create ML ࠓ

Slide 9

Slide 9 text

Sound Activity Tabular Image Text Create ML ৽

Slide 10

Slide 10 text

<9DPEF><0QFO%FWFMPQFS5PPM><$SFBUF.-> Create ML ৽

Slide 11

Slide 11 text

Create ML ৽ υϥοάυϩοϓͰ؆୯ʹ

Slide 12

Slide 12 text

Image Classifier Object Detector

Slide 13

Slide 13 text

%&.0

Slide 14

Slide 14 text

෯޿͍.-ϞσϧΛ ؆୯ʹ࡞Δ͜ͱ͕Ͱ͖Δྫ Create ML

Slide 15

Slide 15 text

Image Classifier Object Detector 7JEFP

Slide 16

Slide 16 text

αΠίϩͷ໨ΛσΟςΫτ Dice Detection Model 5 5 4 4 6

Slide 17

Slide 17 text

Sound Classifier 7JEFP

Slide 18

Slide 18 text

ԻΛฉ͖෼͚Δ

Slide 19

Slide 19 text

Text Classifier Word Tagger 7JEFP

Slide 20

Slide 20 text

จষͷײ৘εύϜτϐοΫΛ෼ྨ Text Classification Topic Classification Spam/Not Spam Sentiment Analysis ! " Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Label 1 Label 2 Label 3 Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum

Slide 21

Slide 21 text

Tabular Classifier Tabular Regressor Recommender 7JEFP

Slide 22

Slide 22 text

Hardtrail Mountain Aspen Valley Pleasant Cove For You: Rocky Peak Cactus Ridge ͜ͷࢁΛߴධՁʹͨ͠ਓʹ ࣍ʹొΔࢁΛϨίϝϯυ

Slide 23

Slide 23 text

Activity Classifier 7JEFP

Slide 24

Slide 24 text

"QQMF8BUDIͷηϯαΛར༻ Device Sensors g a r Accelerometer Gyroscope Magnetometer Altimeter

Slide 25

Slide 25 text

ͲͷߦಈΛ͍ͯ͠Δ͔Λ෼ྨ Jogging Standing Gestures Gaming Golf Swimming

Slide 26

Slide 26 text

.-Y"3 + 7JEFP

Slide 27

Slide 27 text

"QQMF1FODJMͰखॻ͖จࣈࣝผ

Slide 28

Slide 28 text

"3͢͝Ζ͘ͱ૊Έ߹Θͤ

Slide 29

Slide 29 text

✨؆୯ʹ࣮ݱͰ͖Δ✨

Slide 30

Slide 30 text

͡ΜΉͷਪ͠ ϑϦεϏʔͷ౤͛ํ෼ྨͷ%&.0 Frisbee Motion Classifier Hammer Bowler Chicken Wing Backhand Forehand 7JEFP

Slide 31

Slide 31 text

Summary Create ML Domain API Core ML 3

Slide 32

Slide 32 text

%PNBJO"1* ֶशࡁΈͷϞσϧѻ͑Δ

Slide 33

Slide 33 text

Face capture quality Sound analysis Speech Saliency Speech on Mac Sentiment classification Animal Detection Text Recognition Object Tracking Attention Saliency Document Camera On device speech Word Tagging Image similarity Face Landmark Image Similarity Landmark Detection Image saliency Rectangle Detection Text catalog Word Embeddings Image Classification NL Transfer learning

Slide 34

Slide 34 text

7JTJPO'SBNFXPSL *NBHF 7JEFP

Slide 35

Slide 35 text

ਓ͕Ͳ͜ʹண໨͢Δ͔Λֶश Attention and Objectness Based Saliency

Slide 36

Slide 36 text

લܠͱޙܠΛֶश Attention and Objectness Based Saliency

Slide 37

Slide 37 text

ˠը૾͔ΒॏཁͳΦϒδΣΫτΛநग़ Bounding Boxes (0,0)

Slide 38

Slide 38 text

ը૾ͷྨࣅ౓Λܭࢉ Precision and Recall Recall Percentage of Target Class retrieved from entire library

Slide 39

Slide 39 text

'BDF-BOENBSLTͷվળ 76pt Face Landmarks 65pt Single Confidence Score Confidence Score per point New

Slide 40

Slide 40 text

ਓؒݕ஌ New Detectors - Human Detector

Slide 41

Slide 41 text

New Detectors - Cat and Dog Detectors Cat Dog Dog Dog Cat

Slide 42

Slide 42 text

τϥοΩϯάਫ਼౓ͷվળ New Object Tracker

Slide 43

Slide 43 text

7JTJPO'SBNFXPSL 5FYU 7JEFP

Slide 44

Slide 44 text

จࣈ͓͜͠

Slide 45

Slide 45 text

଎͞WTਖ਼֬͞ Fast Versus Accurate Fast Accurate Processing time Optimized for real-time Asynchronous processing Memory footprint Smallest Larger Support for rotated text Limited Broad Support for variety of fonts Limited Diverse font styles Accuracy for natural language Good Best

Slide 46

Slide 46 text

ద੾ͳύϥϝʔλઃఆΛ͢΂͠ Use case • Read codes/serial numbers just like a barcode reader • Constrained camera usage • Interactivity is key request = VNRecognizeTextRequest(completionHandler: recognizeTextHandler) request.recognitionLevel = .fast

Slide 47

Slide 47 text

ӳޠͷΈ

Slide 48

Slide 48 text

/BUVSBM-BOHVBHF 7JEFP

Slide 49

Slide 49 text

จষͷײ৘ղੳ Natural Language Sentiment Analysis I was so excited for the season finale, $ Natural Language Sentiment Analysis but it was a bit disappointing. I was so excited for the season finale, $ ☹

Slide 50

Slide 50 text

Sentiment Analysis Text Classification Sentiment Analysis -1.0 1.0 Natural Language 0 “We had a not so fun time in Hawaii cause mom twisted her ankle.” -0.8 Natural Language Text ෛPSਖ਼ͷײ৘ͷ਺஋Խ

Slide 51

Slide 51 text

4QFFDIBOE4PVOE 7JEFP

Slide 52

Slide 52 text

ΦϯσόΠεͰԻ੠ೝࣝ On-Device Device Support iPhone 6s and later iPad (5th generation) and later All

Slide 53

Slide 53 text

೔ຊޠͳ͍ On-Device Language Support English  United States, Canada,   Great Britain, India Spanish  United States, Mexico,   Spain Italian Brazilian Portuguese Russian Turkish Chinese  Mandarin and Cantonese

Slide 54

Slide 54 text

Ի੠ೝࣝ ˙ൃ࿩଎౓ ˙Ұ࣌ఀࢭ ˙੠ͷಛ௃ ৽

Slide 55

Slide 55 text

੠ͷಛ௃ ˙+JUUFS໎͍ͷ͋Δ੠ʁ ˙4IJNNFSԒͷ͋Δ੠ʁ // Printing new results when recognizing pre-recorded audio if result.isFinal { let formattedString = result.bestTranscription.formattedString let speakingRate = result.bestTranscription.speakingRate let averagePauseDuration = result.bestTranscription.averagePauseDuration for segment in recognitionResult.bestTranscription.segments { let jitter = segment.voiceAnalytics?.jitter.acousticFeatureValuePerFrame let shimmer = segment.voiceAnalytics?.shimmer.acousticFeatureValuePerFrame let pitch = segment.voiceAnalytics?.pitch.acousticFeatureValuePerFrame let voicing = segment.voiceAnalytics?.voicing.acousticFeatureValuePerFrame } } ৽

Slide 56

Slide 56 text

Summary Create ML Domain API Core ML 3

Slide 57

Slide 57 text

$PSF.- 7JEFP

Slide 58

Slide 58 text

Core ML 3 Model Flexibility Model  Personalization NEW

Slide 59

Slide 59 text

Neural Network layers 100+ αϙʔτ

Slide 60

Slide 60 text

഑෍ϞσϧϚγϚγ

Slide 61

Slide 61 text

ΦϯσόΠεΞοϓσʔτ Update Training examples UPDATED NEW

Slide 62

Slide 62 text

User Differences Ñ ൚༻తͳϞσϧͰ͸ਏ͍࣌

Slide 63

Slide 63 text

Update Task Ö ❤ á StickerClassifier Update Task CustomStickerClassifer UPDATED طଘͷϞσϧΛΞοϓσʔτ

Slide 64

Slide 64 text

On-Device Privacy Available No server ΦϯσόΠεͷྑ͞

Slide 65

Slide 65 text

5VSJ$SFBUF 1ZUIPOͰ $PSF.-ϞσϧΛ࡞Δ͜ͱ͕Ͱ͖Δ ϥΠϒϥϦ 7JEFP

Slide 66

Slide 66 text

୔ࢁͷը૾͕ඞཁ Original Approach ࠓ

Slide 67

Slide 67 text

Ұຕͷը૾Ͱ෺ମݕ஌ Synthetic Data Augmentation ৽

Slide 68

Slide 68 text

Synthetic Data Augmentation ৽ Ұຕͷը૾Ͱ෺ମݕ஌

Slide 69

Slide 69 text

No content

Slide 70

Slide 70 text

88%$ʹࢀՃ͢Δ͔೰ΜͰ͍Δਓ޲͚ IUUQTOPUFNVLPPPPPUBLFOOFBF

Slide 71

Slide 71 text

!LPPPPPUBLF ͡ΜΉ

Slide 72

Slide 72 text

%F/"͸ɺຊΠϕϯτͷ಺༰ɺฒͼʹ͓٬༷͕ຊΠϕϯτΛ௨ͯ͡ೖखͨ͠৘ใ౳ʹ͍ͭͯɺ ͦͷ׬શੑɺਖ਼֬ੑɺ࣮֬ੑɺ༗༻ੑ౳ʹ͖ͭɺ͍͔ͳΔ੹೚΋ෛΘͳ͍΋ͷͱ͠·͢ɻ