DistHD

This is the source code for paper "DistHD: A Learner-Aware Dynamic Encoding Method for Hyperdimensional Classification".

Test Environment

• Python: 3.8.10
• Torch: 1.13.1+cu117
• Numpy: 1.23.5

Getting Started

Dataset

We run DistHD on the following dataset:

• MNIST
• UCIHAR
• ISOLET
• PAMAP2
• DIABETE

To try these dataset, download them using links above and store them in data/NAME.py. (NAME = Name of the dataset)

DistHD usage

To execute the code, please preprocess your dataset with data_preprocess.py and then run the corresponding main function in DistHD.py.

The following code generates dummy data and trains a DistHD classification model with it. The result provided by model.test() includes training accuracy and inference accuracy.

dim = 10000
n_samples = 1000
n_features = 100
n_classes = 5
learning_rate = 0.35
epochs = 10
nRegen = 20
x = torch.randn(n_samples, features) # dummy data
y = torch.randint(0, classes, [n_samples]) # dummy data
x_train, x_test, y_train, y_test = sklearn.model_selection.train_test_split(x , y)
model = DistHD(dim, n_features, n_classes, learning_rate)
model.train(x_train, y_train, epochs, nRegen)
model.test(x_train, y_train, x_test, y_test)

The following code generates top1, top2, top3 accuracy using baselineHD:

baseline = DistHD(dim, n_features, n_classes, learning_rate)
baseline_result = DistHD.compare(x, y, epochs)

Citation Request

If you find the code useful, please cite the following paper:

Junyao Wang, Sitao Huang, Mohsen Imani "DistHD: A Learner-Aware Dynamic Encoding Method for Hyperdimensional Classification", IEEE/ACM Design Automation Conference (DAC), 2023.

@inproceedings{wang2023disthd,
  title={DistHD: A Learner-Aware Dynamic Encoding Method for Hyperdimensional Classification},
  author={Wang, Junyao and Huang, Sitao and Imani, Mohsen},
  booktitle={Proceedings of the 60th Annual Design Automation Conference},
  year={2023}
}

Thank you for using our code!