Authors: John D. Co-Reyes, Yingjie Miao, George Tucker, Aleksandra Faust, Esteban Real

Published on: February 08, 2024

Impact Score: 8.07

Arxiv code: Arxiv:2402.05821

Summary

• What is new: The use of a binary discriminator to guide the evolution of machine learning programs, dramatically speeding up the optimization process.
• Why this is important: The challenge of automatically designing better machine learning programs quickly and efficiently.
• What the research proposes: A binary discriminator trained online to select better ML programs without costly evaluations, combined with a uniform representation for various ML components and an adaptive mutation strategy.
• Results: Achieved a 3.7x speedup in the evolution of ML optimizers and a 4x speedup in the evolution of RL loss functions.

Technical Details

Technological frameworks used: GNNs

Models used: Binary discriminators, directed acyclic graph representation for ML components

Data used: nan

Potential Impact

Automated machine learning (AutoML) providers, enterprise companies with internal machine learning operations, cloud computing services offering ML as a service.

Want to implement this idea in a business?

We have generated a startup concept here: EvoLearnAI.