Automatic selection of CNN using genetic algorithm

This repository contains an implementation (in PyTorch) of the DENSER ( Deep Evolutionary Network Structured Representation) approach for automatic selection of a Convoluational Neural Network (CNN) architecture. The followed approach is further described in this paper: DENSER: Automatic Selection of Convolutional Neural Network Architectures.

DENSER

The method proposed in the paper has to do with Neuroevolution, a field which deals with the automatic optimisation of ANN architectures and parametrization. The DENSER approach, in particular, combines Genetic Algorithms (GA) with Dynamic structured grammatical evolution (SGDE), which represents the genotype of each individual (a DNN in our case) through the use of a grammar, expressed in backus-naur form (BNF). (DSGE is slightly different from GE and SGE and these differences are explained in detail in the paper)

The grammar is used to dynamically generate a population of CNN architectures. Each individual of the population is then evaluated using a fitness function. As fitness function the accuracy of the CNN on a validation set is used. Evolution is performed in the following way:

• an initial population (of size $n$) of randomly generated CNNs is created (following the rules of the given grammar)

then, for the number of generations we initially set, we repeat the following steps:

• a fitness score is computed for each individual
• a pair of parents is chosen randomly from the fittest individuals
• through crossover operation we create two new individuals and we choose the longest one
• mutations are applied to the newly generated individual

The last three steps are repeated $n$ times for each generation, in order to obtain again a population of $n$ individuals.

The whole process is repeated until a CNN is found that meets the desired accuracy on the validation set.

Grammar

The grammar used is almost the same as the one reported in the paper and can be found in 'src/cnn.grammar.txt'.

Requirements

• numpy >= 1.21.5
• python >= 3.8.13
• pytorch >= 1.11.0
• tqdm >= 4.64.0
• matplotlib >= 3.5.2
• torchvision >= 0.12.0

Usage

To run type:

$ python3 main.py {cifar10 │ MNIST} {pop_size} {num_gen} {batch_size} {subpath-on-you-want-to-save}

The programm will print onf best_organisms the best performing CNNs found during evolution.

Structure of the repository

├── data
├── main.py
├── main_test.py
├── README.md
├── results
│   ├── MNIST
│   └── cifar10
│ 
├── scripts
│   ├── dataloader.py
│   ├── train.py
│   └── utils.py
├── src
│   ├── cnn.grammar.txt
│   ├── dsge_level.py
│   ├── evolution.py
│   ├── ga_level.py
│   ├── grammar.py
│   ├── mutations.py
│   ├── nn_encoding.py
|
└── tests
    └── tests.py