Depth induces an implicit low-rank bias in deep unconstrained feature models trained with unregularized multiclass cross-entropy, promoting softmax codes over neural collapse via more efficient norm propagation.
Title resolution pending
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
years
2026 3verdicts
UNVERDICTED 3representative citing papers
Three new robust error models for catalytic tape resetting are characterized with equivalences to standard classes and collapse under derandomization.
L¹ polynomial regression achieves Õ(n^{O(log(1/ε)/σ)}) for smoothed agnostic halfspace learning, with nearly matching SQ lower bound n^{Ω(log(1+σ/ε²)/σ)}.
citing papers explorer
-
The Implicit Bias of Depth: From Neural Collapse to Softmax Codes
Depth induces an implicit low-rank bias in deep unconstrained feature models trained with unregularized multiclass cross-entropy, promoting softmax codes over neural collapse via more efficient norm propagation.
-
Understanding Robust Catalytic Computing
Three new robust error models for catalytic tape resetting are characterized with equivalences to standard classes and collapse under derandomization.
-
A Near-optimal SQ Lower Bound for Smoothed Agnostic Learning of Boolean Halfspaces
L¹ polynomial regression achieves Õ(n^{O(log(1/ε)/σ)}) for smoothed agnostic halfspace learning, with nearly matching SQ lower bound n^{Ω(log(1+σ/ε²)/σ)}.