Dense Neural Networks are not Universal Approximators
Levi Rauchwerger, Stefanie Jegelka, Ron Levie · 2026
Read Dense Neural Networks are not Universal Approximators by Levi Rauchwerger, Stefanie Jegelka, Ron Levie (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-07618.txt.
Description
Universal approximation theorems say large networks can approximate any continuous function, but only without restrictions on weights. Levi Rauchwerger and two co-authors show that dense neural networks, with natural constraints framing dense links, do not share that universality.
Combining a model compression argument, the weak regularity lemma, and a view of feedforward networks as message-passing graph networks, they exhibit Lipschitz functions that dense ReLU networks cannot match. The result highlights an intrinsic limitation of dense layers and motivates sparse connectivity as a necessary ingredient for true breadth. Network theorists get a precise boundary for what dense architectures can express.
Catalog record
| CREATOR | Levi Rauchwerger, Stefanie Jegelka, Ron Levie |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 92 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-07618