"Spiking GLOM: Bio-inspired Architecture for Next-generation Object Recognition", Peng Kang, Srutarshi Banerjee, Henry Chopp, Aggelos Katsaggelos, Oliver Cossairt. In review.
From the main directory run:
pipenv install
pip install spikingjelly
to install the required dependencies.
To run contrastive pre-training on CIFAR-10, execute:
python spiking_glom_new/main.py --flagfile spiking_glom_config_new/spiking_glom_contrast.cfg
After pre-training, to run supervised training on CIFAR-10, execute:
python spiking_glom_new/main.py --flagfile spiking_glom_config_new/spiking_glom_supervise.cfg
See spiking_glom_new/flags_Agglomerator_slom.py
to check all the flag meanings.
To run contrastive pre-training on CIFAR-10, execute:
python hybrid_spiking_glom_new/main.py --flagfile hybrid_spiking_glom_config_new/spiking_glom_contrast.cfg
After pre-training, to run supervised training on CIFAR-10, execute:
python hybrid_spiking_glom_new/main.py --flagfile hybrid_spiking_glom_config_new/spiking_glom_supervise.cfg
For Potential-assisted Spiking GLOM:
python spiking_glom_new/main_energy.py --flagfile spiking_glom_config_new/spiking_glom_energy.cfg
For Hybrid Spiking GLOM:
python hybrid_spiking_glom_new/main.py --flagfile hybrid_spiking_glom_config_new/spiking_glom_energy.cfg
python spiking_glom_new/main.py --flagfile spiking_glom_config_new/spiking_glom_plot.cfg
We provide pre-trained models Potential-assisted Spiking GLOM and Hybrid Spiking GLOM for calculate the firing rates
and specific 5-level Potential-assisted Spiking GLOM for freeze to plot
.
- Agglomerator by Garau et al.
- SpikingJelly