Demo codes for

"Invariance of object detection in untrained deep neural networks"

Jeonghwan Cheon, Seungdae Baek, and Se-Bum Paik*

*Contact: [email protected]

1. System requirements

• MATLAB 2021b or later version
• Installation of the Deep Learning Toolbox (https://www.mathworks.com/products/deep-learning.html)
• Installation of the pretrained AlexNet (https://de.mathworks.com/matlabcentral/fileexchange/59133-deep-learning-toolbox-model-for-alexnet-network)
• No non-standard hardware is required.
• Uploaded codes were tested using MATLAB 2021b.

2. Installation

• Download all files and folders. ("Clone or download" -> "Download ZIP")
• Download 'Image.zip' from below link and unzip files in the same directory
• Dataset Download:
• Expected Installation time is about 60 minutes, but may vary by system conditions.

3. Instructions for demo

• Edit "Main.m' to select the class of the object-selective units to which perform analysis.
• Select result numbers (from 1 to 5) that you want to perform a demo simulation.
• Expected running time is about 5 minutes for each figure, but may vary by system conditions.

4. Expected output for demo

• Below results for untrained AlexNet will be shown.
  • Run_Selectivity.m: Emergence of selectivity to various objects in untrained networks (Result 1)
  • Run_Invariance.m: Viewpoint-invariant object selectivity observed in untrained networks (Result 2a)
  • Run_PFI.m: Viewpoint-invariant unit and specific units and its visual feature encoding (Result 2b)
  • Run_Connectivity.m: Computational model explains spontaneous emergence of invariance in untrained networks (Result 3)
  • Run_SVM.m: Invariantly tuned unit responses enable invariant object detection (Result 4)

5. Citation

@ARTICLE{10.3389/fncom.2022.1030707,
  AUTHOR={Cheon, Jeonghwan and Baek, Seungdae and Paik, Se-Bum},
  TITLE={Invariance of object detection in untrained deep neural networks},
  JOURNAL={Frontiers in Computational Neuroscience},
  VOLUME={16},
  YEAR={2022},
  URL={https://www.frontiersin.org/articles/10.3389/fncom.2022.1030707},
  DOI={10.3389/fncom.2022.1030707},
  ISSN={1662-5188}
}