AutoML-Zero to One: Discovering Machine Learning
from Simple Primitives
Staff Research Scientist, Google Deepmind
Date: Thursday, Sep 14, 2023, 14:00 - 15:00
Can learning emerge from a search process in-silico? Our AutoML-Zero work at Google DeepMind shows that a simple evolutionary search process can automatically discover modern learning techniques from scratch. Without knowledge of machine learning, the process discovers algorithms by combining simple primitives such as additions and multiplications. In this type of primitive search, learning emerges naturally when survival depends on more than one task. The methods transfer to realistic setups, where primitive search can discover novel algorithms for ML optimization (the Lion optimizer) and robot adaptation (Auto-Robotics-Zero). In my talk, I will introduce this type of primitive symbolic search, present an overview of toy and state-of-the-art results, and motivate further work into this exciting research direction
Dr. Esteban Real is a research scientist at Google DeepMind, where he studies bio-inspired computing, especially automated machine learning, evolutionary search, and the relationship between natural and artificial neural networks. He currently focuses on simultaneous symbolic–continuous optimization, such as is needed for architecture search or program discovery.