Associated Learning

Associated Learning modularizes neural network into smaller components, each of which has a local objective. Because the objectives are mutually independent, Associated Learning can learn the parameters at different components independently and simultaneously.

See the following paper for details:

[1] Associated Learning: Decomposing End-to-end Backpropagation based on Auto-encoders and Target Propagation by Yu-Wei Kao and Hung-Hsuan Chen

Citing the paper

Yu-Wei Kao and Hung-Hsuan Chen. "Associated Learning: Decomposing End-to-End Backpropagation Based on Autoencoders and Target Propagation." Neural Computation 33, no. 1 (2021): 174-193.

BibTeX:

@article{kao2021associated,
    title={Associated Learning: Decomposing End-to-End Backpropagation Based on Autoencoders and Target Propagation},
    author={Kao, Yu-Wei and Chen, Hung-Hsuan},
    journal={Neural Computation},
    volume={33},
    number={1},
    pages={174--193},
    year={2021},
    publisher={MIT Press}
}

Tested Environment

• Windows 10 (1903)
• Tensorflow 1.13.1 and 1.14.0

Usage

Each folder contains a main.py and a models.py. The folder name corresponds to the experimental dataset. Specifying the model you want to test with --model flag and run the following:

python main.py --model <MODEL>

For MNIST dataset, 4 models can be selected:

• MLP
• MLP_AL
• CNN
• CNN_AL

For CIFAR-10 and CIFAR-100, 10 models can be selected:

• MLP
• MLP_AL
• CNN
• CNN_AL
• ResNet_20
• ResNet_20_AL
• ResNet_32
• ResNet_32_AL
• VGG
• VGG_AL