liyipeng-moon/categoryselectivetvae

Modeling Category-Selective Cortical Regions with Topographic Variational Autoencoders

Modeling Category-Selective Cortical Regions with Topographic Variational Autoencoders

Getting Started

Install requirements with Anaconda:

conda env create -f environment.yml

Activate the conda environment

conda activate tvae

Install the tvae package

Install the tvae package inside of your conda environment. This allows you to run experiments with the tvae command. At the root of the project directory run (using your environment's pip): pip3 install -e .

If you need help finding your environment's pip, try which python, which should point you to a directory such as .../anaconda3/envs/tvae/bin/ where it will be located.

(Optional) Setup Weights & Biases:

This repository uses Weight & Biases for experiment tracking. By deafult this is set to off. However, if you would like to use this (highly recommended!) functionality, all you have to do is set 'wandb_on': True in the experiment config, and set your account's project and entity names in the tvae/utils/logging.py file.

For more information on making a Weight & Biases account see (creating a weights and biases account) and the associated quickstart guide.

Running an experiment

To evaluate the selectivity of pretrained alexnet (the non-topographic baseline), you can run:

• tvae --name 'ffa_modeling_pretrained_alexnet'

To train and evaluate the selectivity of the TVAE for objects, faces, bodies, and places, you can run:

• tvae --name 'ffa_modeling_fc6'

To train and evaluate the selectivity of the the TDANN for objects, faces, bodies, and places, you can run:

• tvae --name 'ffa_modeling_tdann'

To evaluate the selectivity of the TVAE on abstract catagories (animacy vs. inanimacy):

• tvae --name 'ffa_modeling_fc6_functional'

To evaluate the selectivity of the TDANN on abstract catagories (animacy vs. inanimacy):

• tvae --name 'ffa_modeling_tdann_functional'

These 'functional' experiment files can also be easily modified to test selectivity to big vs. small objects by simply changing the directories of the input images.

Basics of the framework

• All experiments can be found in tvae/experiments/, and begin with the model specification, followed by the experiment config.

Model Architecutre Options

• 'mu_init': int, Initalization value for mu parameter
• 's_dim': int, Dimensionality of the latent space
• 'k': int, size of the summation kernel used to define the local topographic structure
• 'group_kernel': tuple of int, defines the size of the kernel used by the grouper, exact definition and relationship to W varies for each experiment.

Training Options

• 'wandb_on': bool, if True, use weights & biases logging
• 'lr': float, learning rate
• 'momentum': float, standard momentum used in SGD
• 'max_epochs': int, total training epochs
• 'eval_epochs': int, epochs between evaluation on the test (for MNIST)
• 'batch_size': int, number of samples per batch
• 'n_is_samples': int, number of importance samples when computing the log-likelihood on MNIST.

Acknowledgements

The Robert Bosch GmbH is acknowledged for financial support.