Reproduce “Conditional Meta-Learning of Linear Representations” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Giulia Denevi, Massimiliano Pontil, Carlo Ciliberto
The authors introduce conditional meta-learning for prediction, using a task’s training examples to select its representation rather than sharing one fixed representation across tasks. In synthetic regression with 20-dimensional inputs and two-dimensional subspaces, conditional learning beats unconditional learning and independent task learning as clusters increase from two to six. A theoretical excess-risk bound improves by √m under covariance conditions, not generally.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.