iScience · 2025 · Journal article

Evolutionary learning in neural networks by heterosynaptic plasticity

Zedong Bi, Ruiqi Fu, Guozhang Chen, Dongping Yang, Yu Zhou, Liang Tian

Combining evolutionary routing with dopamine-guided heterosynaptic plasticity trains biophysical spiking and analog neural networks without backpropagation.

The scientific question

How can complex biophysical neural networks learn effectively without gradient backpropagation?

Backpropagation struggles with instability and non-differentiable dynamics when optimizing detailed biophysical and spiking neuron models.

Network architecture and evolutionary learning mechanism.
Figure 1 from the original paper ↗ · CC BY-NC 4.0

The core idea

Dopamine-Guided Evolutionary Heterosynaptic Plasticity

An evolutionary algorithm evaluates distinct neural routing pathways via alternating gating while guiding synaptic updates through dopamine modulation, dendritic meta-plasticity, and memory replay.

What we found

Gradient-Free Biophysical Network Training

The framework trains both feedforward and recurrent spiking or analog networks, reproducing cognitive-like neural dynamics and matching gradient-based performance on classification and control tasks.

Continue the question

Related reading

Explore our research

Cite this work

Zedong Bi, Ruiqi Fu, Guozhang Chen, Dongping Yang, Yu Zhou, Liang Tian. Evolutionary learning in neural networks by heterosynaptic plasticity. iScience, 2025. doi:10.1016/j.isci.2025.112340.