What question did this study set out to answer?

The central aim is to present Neural DNA (NDNA) as a compact genome that optimizes neural network growth.

March 29, 2026Open Access

Neural DNA: A Compact Genome for Growing Network Architecture

Key Points

The central aim is to present Neural DNA (NDNA) as a compact genome that optimizes neural network growth.
Development of Neural DNA with fewer than 300 parameters.
Application of type-based compatibility rules for network growth.
Initialization of networks in a default-disconnected state followed by metabolic cost pressure application.
Evaluation across three architectures: MLP, CNN, Transformer, and five datasets.
NDNA outperforms random sparsity by 0.39% to 7.01%.
Topology can transfer across tasks without modification.
Compression ratios achieve 8,384:1 in the largest tested architecture.

Abstract

We introduce Neural DNA (NDNA), a compact learned genome of fewer than 300 parameters that grows neural network topology through type-based compatibility rules, default-disconnected initialization, and metabolic cost pressure. NDNA consistently outperforms random sparsity by 0.39% to 7.01% across three architectures (MLP, CNN, Transformer) and five datasets (MNIST, CIFAR-10, CIFAR-100, Fashion-MNIST, IMDB). Topology transfers across tasks without modification. Compression ratios scale with network size, reaching 8,384:1 on the largest architecture tested.

Neural DNA: A Compact Genome for Growing Network Architecture

Key Points

Abstract

Cite This Study

Also Consider

Also Consider