Biologically Inspired Semantic Lateral Connectivity in Convolutional Neural Networks

Installation

This module was implemented for Python 3.8, PyTorch 1.7.1 and CUDA 10.1. Required packages can be found in the requirements.txt file.

Usage

The working directory for all main scripts is semlc/. The main script to run experiments is semlc/run.py.

Use for instance as follows:

cd semlc
python3 run.py alexnet semlc -e 180 -w 3 --data cifar10

# for help, use
python run.py --help

Inspecting Results

Every model is assigned a unique process id. Results and parameters of the model are stored in semlc/experiments/static/saved_models. We provide a Flask app which can be started from the root directory by calling ./monitor. It provides an interface to navigate and analyze models and their results.