https://github.com/avanishd-3/tf-neural-nets
Neural networks written in TensorFlow
https://github.com/avanishd-3/tf-neural-nets
tensorflow
Last synced: 20 days ago
JSON representation
Neural networks written in TensorFlow
- Host: GitHub
- URL: https://github.com/avanishd-3/tf-neural-nets
- Owner: avanishd-3
- Created: 2025-02-24T06:15:47.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2025-04-22T06:06:26.000Z (about 1 year ago)
- Last Synced: 2026-05-05T22:40:09.534Z (2 months ago)
- Topics: tensorflow
- Language: Jupyter Notebook
- Homepage:
- Size: 97.5 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# tf-neural-nets
Collection of neural netwworks made with TensorFlow and Keras.
Each network also includes the model weights in a .keras file.
## Usage Instructions
```
git clone https://github.com/avanishd-3/tf-neural-networks.git
```
Then run the notebooks
Models can also be loaded with
```
model = tf.keras.models.load_model("model_filename.keras")
```
## Model Quality
Model positions the best models (for their respective tasks) would have on the [Papers with Code](https://paperswithcode.com/) leaderboard as of April 7, 2025
| Model | Accuracy (%) | Ranking |
| ----------------- | ------------ | ------------- |
| Fashion MNIST CNN | 91.78 | 15 |
| MNIST ANN | 98.44 | 53 |
| CIFAR-10 CNN | 81.12 | 245 |
## ANNs
There are 3 models here.
1. Classification model based on the Fashion MNIST dataset (~88.43% accuracy).
2. Regression model based on the Auto MPG dataset.
3. High-accuracy classification model based on the MNIST dataset (~98.44% accuracy).
4. Classification model based on the CIFAR-10 dataset (~49.20% accuracy, which is pretty high for a non-CNN model on an image-recognition dataset).
## CNNs
There are 3 models here.
1. Classification model based on the Fashion MNIST dataset (~91.78% accuracy).
2. Classification model based on the CIFAR-10 dataset (~81.12% accuracy).
- Note: I made a [Transformer-based model](https://huggingface.co/avanishd/vit-base-patch16-224-in21k-finetuned-cifar10) that has a 97.93% accuray after only 1 epoch of training.
4. Transfer learning (based on Xception) binary classification model trained on TensorFlow cat or dog dataset (~95.53% accuracy).
- No weight for this one, because it exceeds the GitHub file size.
- Training is pretty quick, though, because most of the training has already been done.
## VAE
There is 1 model here.
1. Image generation VAE model trained on Fashion MNIST