2023 Google I/O [10 ways of using machine learning with Google Cloud]

prsper2001·2023년 5월 19일

GDSC Yonsei

목록 보기

This page summarizes the following Google I/O video


1. Teachable Machine

  • Browser based tool for image or audio classification
  • Data from webcam, microphone, or via upload
  • Great for prototyping
  • Models are exportable
  • Exported model is TensortFlow-like so it can be deployed anywhere such as on mobile, web, or even Cloud

2. AutoML in Vertex AI

  • Machine-learning tool for developers on Google cloud
  • Used to train custom enterprise grade models
  • No ML expertise required
  • High quality models
  • Build models to predict on images, video, text, or even tabular data
  • Easy to deploy the models for large scale batch and real time predictions
3. ML APIs
  • ML APIs for developers
  • Embed machine learning capabilities directly into application with just a single API call
  • Use powerful pre-trained models by Google
  • Vision API, Natural Language API, Video Intelligence API
  • Includes latest generative AI APIs like text to image and code completion
  • Great for common tasks like sentiment analysis and text and person detection in video
4. Generative AI Studio
  • Use prompts to generate images and text
  • Chat, sumarize text, and more
  • User-Interface or API
  • Prompt-tune and fine tune models
5. Vertex AI Model Garden
  • Single environment to interact with a variet of model types
  • Models will span across modalities
  • Deploy to endpoints with ease
6. Four New Foundation Models in Model Garden
  1. Code Generation & Completion: Enhance the software development leveraging your own code base
  2. Image Generation: Generate high-quality images with low latency, edit and iterate to your specifications
  3. Universal Speech: Unlocks the power of voice and highly accurate speech transcriptions
  4. Embeddings: Extract high quality semantic information from unstructured data
7. New Tuning Capability: Reinforcement Learning from Human Feedback (RLHF)
  • Use human feedback to increase your model's usefulness
  • Makes easy to optimize machine learning model performance with human feedback
  • Improve pre-trained LLM without increasing the model size
8. BigQuery ML
  • ML directly in your data warehouse
  • Use popular ML models in SQL
  • Use Vertex AI endpoints or ML APIs on your BigQuery data
  • MLOps integrations
9. Vertex AI custom training
  • JupyterLab with fully customizable compute
  • Enterprise-grade
  • Use Colab for quick use of Python
10. Vertex AI Matching Engine
  • Fast, scalable similarity search
  • Vector database for embeddings
  • Great for variety of data modalities
  • Enterprise-grade

0개의 댓글