autogen/tutorials/flaml-tutorial-aaai-23.md

AAAI 2023 Lab Forum - LSHP2: Automated Machine Learning & Tuning with FLAML

Session Information

Date and Time: February 8, 2023 at 2-6pm ET.

Location: Walter E. Washington Convention Center, Washington DC, USA

Duration: 4 hours (3.5 hours + 0.5 hour break)

For the most up-to-date information, see the AAAI'23 Program Agenda

Lab Forum Slides

What Will You Learn?

• What FLAML is and how to use FLAML to
  • find accurate ML models with low computational resources for common ML tasks
  • tune hyperparameters generically
• How to leverage the flexible and rich customization choices
  • finish the last mile for deployment
  • create new applications
• Code examples, demos, use cases
• Research & development opportunities

Session Agenda

Part 1. Overview of FLAML

• Overview of AutoML and FLAML
• Basic usages of FLAML
  • Task-oriented AutoML
    • Documentation
    • Notebook: A classification task with AutoML; Open In Colab
  • Tune User-Defined-functions with FLAML
    • Documentation
    • Notebook: Tune user-defined function; Open In Colab
  • Zero-shot AutoML
    • Documentation
    • Notebook: Zeroshot AutoML; Open In Colab
• ML.NET demo

Break (15m)

Part 2. Deep Dive into FLAML

• The Science Behind FLAML’s Success

  • Economical hyperparameter optimization methods in FLAML
  • Other research in FLAML

• Maximize the Power of FLAML through Customization and Advanced Functionalities

  • Notebook: Customize your AutoML with FLAML; Open In Colab
  • Notebook: Further acceleration of AutoML with FLAML; Open In Colab
  • Notebook: Neural network model tuning with FLAML ; Open In Colab

Part 3. New features in FLAML

• Natural language processing
  • Notebook: AutoML for NLP tasks; Open In Colab
• Time Series Forecasting
  • Notebook: AutoML for Time Series Forecast tasks; Open In Colab
• Targeted Hyperparameter Optimization With Lexicographic Objectives
  • Documentation
  • Notebook: Find accurate and fast neural networks with lexicographic objectives; Open In Colab
• Online AutoML
  • Notebook: Online AutoML with Vowpal Wabbit; Open In Colab
• Fair AutoML

Challenges and open problems