FURI | Fall 2026, Needs Review

Neuron Activation Network Optimizer

Data icon, disabled. Four grey bars arranged like a vertical bar chart.

This research investigates whether neurons that learn the same function can be identified and removed during training without a meaningful loss of accuracy. The Neuron Activation Network Optimizer (NANO) records each hidden neuron’s responses, groups neurons with matching behavior, and deletes the duplicates while training continues. The resulting smaller network is intended to reduce the cost of running artificial intelligence on phones and other devices with limited power. Future work will test NANO on larger image datasets and compare it with existing pruning methods.