Anirudh Goyal Proposes Ranking Candidates for AI Experiments
Google DeepMind scientist frames experiment selection as a ranking task using candidates and history.
TLDR
Anirudh Goyal, Senior Staff Research Scientist at Google DeepMind with a PhD from Mila under Yoshua Bengio, outlined an approach to choosing experiments in machine learning research. He noted that predicting the exact score an experiment will achieve is a hard forecasting problem. Instead, he suggested turning the task into a preference problem: given several candidate solutions plus the research history so far, rank which candidate is most worth executing next.
Combined views
37.8K
14 Sources, first seen 30d ago