trainability-not-capacity-was-the-depth-bottleneck

IN derived (depth 2)

Created 2026-06-21T10:09:45+00:00 · Reviewed 2026-06-21T15:37:01+00:00

The theory-practice gap in neural network depth — where universal approximation says shallow networks suffice but practice demands deep ones — reveals that the real bottleneck was never representational capacity but trainability, and residual connections resolved exactly this optimization barrier, enabling the hierarchical representations that shallow networks could theoretically but not practically learn.

Justifications

SL — Skip connections proved that depth's limitation was optimization, not representation

Antecedents (all must be IN):

  • IN residual-connections-enable-arbitrary-depth — Residual connections (F(x) + x) are the key architectural innovation enabling arbitrarily deep networks — ResNet solved the degradation problem for CNNs, and the same mechanism was adopted by transformers to enable deep stacking without vanishing gradients.
  • IN universal-approximation-vs-practical-depth — Universal approximation theory (Cybenko, 1989) establishes that shallow networks with sigmoidal activations can approximate any continuous function on compact subsets of R^n, but this existence guarantee has practical caveats: deep networks leverage hierarchical feature learning across multiple layers, and ReLU networks require width strictly greater than input dimension to maintain universal approximation — suggesting that theoretical approximation capacity alone does not determine architectural effectiveness.

Dependents

These beliefs depend on this one: