Overview
- Problem statement: training and deploying ML models is time-consuming
- No-code solution
- Allows fast prototyping, or exploration of new datasets before investing in development
- Supports: tabular data, images, video, text
- Solves different types of problems (objectives)
- Upload data from Cloud Storage, BigQuery or local storage
Model Options
Image
- Classification—list of categories
- Object detection—labels and bounding boxes
Tabular
- Classification—list of categories
- Regression—returns numeric values
- Forecasting—time-dependent, predict series of numeric values in the future
Text
- Classification—list of categories
- Entity extraction—extract and label known entities
- Sentiment analysis—identify emotional opinion
Video
- Classification—list of categorized shots and segments
- Object tracking—list of shots and segments where objects are detected
- Action recognition—list of categorized actions with moments they happened
Key Technologies
- Transfer learning
- Uses a small dataset alongside model pretrained on a large, similar dataset
- Neural architecture search
- Trains and evaluates multiple models, compares, and chooses best one