BlockNet10 is a neural network architecture designed for image classification tasks using the CIFAR-10 dataset. This model implements a sequence of intermediate blocks (B1, B2, ..., BK) followed by an output block (O).
Each intermediate block receives an input image x and outputs an image x'. The block comprises L independent convolutional layers, denoted as C1, C2, ..., CL.
Each convolutional layer Cl in a block operates on the input image x and outputs an image Cl(x).
The output image x' is computed as x' = a1C1(x) + a2C2(x) + ... + aLCL(x), where a = [a1, a2, ..., aL]T is a vector computed by the block.
The vector a is obtained by computing the average value of each channel of x and passing it through a fully connected layer with the same number of units as convolutional layers in the block.
The output block processes the final output image from the intermediate blocks for classification.
| Epoch Number | Train Accuracy | Test Accuracy | Average Loss |
|---|---|---|---|
| 50 | 75.43 | 80.56 | 0.685 |
You can contribute to the advancement of this architecture, changes in hyperparameter, or solve issues here.
If you use BlockNet10 in your research or work, please cite it as follows:
@article{blocknet10,
title={BlockNet10: CIFAR-10 Image Classifier},
author={Siddhesh Kulthe},
year={2024},
publisher={Hugging Face},
url={https://huggingface.co/siddheshtv/BlockNet10}
}