Conference Google: Construire des solutions plus intelligentes sans expertise en machine learning
Sans expertise en ML != Sans ML
Intro
Who are we ?
Laurent Picard
- Developer advocate - Google Cloud
- Ebook pioner
Any sufficiently advanced technology is indistiguishable from magic
What is machine learning ?
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Why is machine learning now possible ?
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Three ways we can benefit from ML today
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Ne reiventez pas la roue !
Nouveau champ: auto-ML
- on peut construire nos propres models sans expertise
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Machine learning API
Ready-to-use models
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Vision API
Computer vision before ML:
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Landmark detection:
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Capable de determiner ou a ete prise la photo (quel endroit)
La photo originale a ete modifiee (symetrie horizontale)
- Toujours capable de determiner l'origine de la photo
Object detection:
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Face detection:
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Vue 3D de Gollum donc pas un vrai visage humain (mais marche quand meme !)
Text detection:
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Meme avec une legere rotation, on detecte toujours le texte
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Detecte egalement l'ecriture manuscrite (quelques erreurs)
Web entity detection and image matching:
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
La photo ci-dessus de Tolkien est totalement inedite pour l'API utilisee, capable de reconnaitre Tolkien + determiner que l'origine est un journal espagnol
OSS Client libraries
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Librairies clientes en open-source sur GitHub dans plusieurs langages.
Video Intelligence API
Demo:
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
OSS Client libraries
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Natural Language API
Analyze text with a simple request
Syntax analysis:
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Entity detection
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Content classification
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Sentiment analysis:
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Le ML se plante totalement sur la detection du sarcasme.
Translation API
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Google Translate par exemple!
On peut les ameliorer regulierement en fournissant de plus en plus d'exemples et de contre-exemples.
Speech-to-Text API
Convert text to speech in 120 languages with a simple request.
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Fonctionne en temps reel.
Ex: il y a quelques annees repeter a un bot en appelant une banque "Je veux un conseiller" en esperant qu'il comprenne.
Consequence sympa des reseaux neuronaux: aujourd'hui les speech-to-text API sont resistants aux bruits car ils apprennent a partir de vrais echantillon.
Speech timestamps:
Search for text within your audio
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
OSS Client libraries
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Text-to-speech (TTS) API
Generate natural speech with a simple request
WaveNet natural voices, par Deepmind
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
C'est le modele le plus avance de tous, qui reproduit le mieux une voix humaine.
Demo: "Quelle est la temperature a Paris ?" avec un accent anglais
Image Not Showing
Possible Reasons
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
Learn More →
Tadaaaaaa
OSS Client libraries

Tuto pour generer des voix
AutoML
Build your custom model with no expertise
Generic results with the Vision API

Cloud AutoML

Demo:

Utilisation de ~250 images en moyennes

"Juste" 3h de calculs.

Auto-generate a custom model from your data


Demo
Evaluation

Transfert learning

Hyperparameter tuning

Conlusion
How can I build smarter solutions ?

Liens utiles
→ Présentation
→ BD Google AI
→ ML codelabs