Dpp1/nested-learning-paper

Nested Learning: The Illusion of Deep Learning Architectures

Authors: Ali Behrouz, Meisam Razaviyayn, Peiling Zhong, Vahab Mirrokni
Affiliation: Google Research
Uploaded by: Independent researcher (not affiliated with the authors)

PDF: NL.pdf

Abstract

Nested Learning (NL) is a learning paradigm that replaces deep neural networks with nested optimization problems. It unifies supervised, unsupervised, and continual learning without traditional deep architectures. This work explores how nested optimizers can be used as a general framework for learning loops and alternative approaches to deep learning.

Key Points

• Nested optimization replaces layers
• Works for supervised and unsupervised tasks
• Demonstrates alternatives to traditional deep learning

Citation