Self-motivated growing neural network for adaptive architecture via local structural plasticity

Yiyang Jia and Chengxu Zhou

Volume 702, article 134581

Cite this paper

We asked whether a controller could adjust its own capacity during training. SMGrNN combines gradient-based learning with a structural plasticity module: the network changes not only its weights, but also which neurons are present.

We compared adaptive and fixed networks on control benchmarks, and examined what changed when growth or pruning was switched off. This work concerns adaptation within a task; retaining skills across tasks is a question in my current research.

@article{jia2026smgrnn,
  author  = {Jia, Yiyang and Zhou, Chengxu},
  title   = {Self-motivated growing neural network for adaptive architecture via local structural plasticity},
  journal = {Neurocomputing},
  year    = {2026},
  volume  = {702},
  pages   = {134581},
  doi     = {10.1016/j.neucom.2026.134581}
}

All publications