Beyond the Loss Curve: Scaling Laws, Active Learning, and the Limits of Learning from Exact Posteriors
Abstract
Class-conditional normalizing flows provide tractable posterior probabilities for studying learning behavior. This framework examines scaling laws, soft-label training, distribution shifts, and active learning while separating epistemic error from the uncertainty that remains even for an optimal predictor.
Type
Publication
arXiv preprint
Ravid Shwartz Ziv
AI Researcher
AI researcher at Meta MSL with a background in information theory and computational neuroscience, working on world models, memory, and compression. Former Assistant Professor and Faculty Fellow at NYU. My industry experience spans Wand AI, Intel, Google AI, and Wikipedia.