Publications

@article{swaroop_bennabhaktula_pushpull-net_2024,

title = {PushPull-Net: Inhibition-driven ResNet robust to image corruptions},

author = {Guru Swaroop Bennabhaktula and Enrique Alegre and Nicola Strisciuglio and George Azzopardi},

url = {https://scholar.google.es/citations?view_op=view_citation&hl=en&user=opCbArQAAAAJ&cstart=100&pagesize=100&sortby=title&citation_for_view=opCbArQAAAAJ:NXb4pA-qfm4C},

year  = {2024},

date = {2024-01-01},

journal = {International Conference on Pattern Recognition},

pages = {391–408},

abstract = {This paper introduces a new computational unit called PushPull-Conv, applied in the first layer of a ResNet architecture. Inspired by anti-phase inhibition in the visual cortex, this unit uses a pair of complementary filters: a push kernel and a pull kernel. The push kernel learns to respond to specific stimuli, while the pull kernel reacts to opposite contrasts. This design enhances stimulus selectivity and improves robustness by inhibiting responses in regions without preferred stimuli. Integrating PushPull-Conv into ResNets improves their resilience to image corruption.},

note = {Publisher: Springer Nature Switzerland Cham},

keywords = {Convolutional Neural Networks, Image Corruption, ResNet, Visual Cortex Simulation},

pubstate = {published},

tppubtype = {article}

}