Whether the enterprise is completely new to machine learning (ML) or it has already trained and deployed a model from scratch, Google Cloud Platform has a variety of tools to help the company start using ML right now.
Sara Robinson, developer advocate, Google, centered her presentation around the basics of machine learning during her Data Summit 2019 presentation, “Exploring Machine Learning on the Google Cloud Platform.”
“If you boil it down machine learning is just basic multiplication,” Robinson said.
Machine learning on Google Cloud starts out with choosing the right model. Data scientists and machine learning engineers take care of this process, Robinson explained.
AI platform training and prediction can be done in house with the Google Cloud platform. This eventually leads to ML API’s which are pre-trained models with a single REST API request.
AutoML supports images, text, translation, and structured data. Data uploaded into this tool can automatically get predictions for data right away, she said. AutoML recently updated to include tables, numerical, or categorical inputs.
For data that isn’t a fit for AutoML, the Coud AI platform is the place for custom models on the Google Cloud.
Data labeling, built-in algorithms, deep learning VM images, training, notebooks, and predictions are offered in the Google AI platform.
AI isn’t infallible, however, models can show bias depending on the person compiling data for analysis, Robinson said.
Taking steps to avoid bias in ML system includes asking questions such as:
- Is your training data representative of the population using your product?
- Who is labeling your data? How are you accounting for bias introduced in the labeling phase?
Many presenters are making their slide decks available on the Data Summit 2019 website at www.dbta.com/DataSummit/2019/Presentations.aspx.