A collection of deep learning implementations in PyTorch, covering fundamental architectures and training pipelines.
This repository serves as a learning and reference resource for building neural networks from scratch and using PyTorch modules.
In this Repository we have build simple Neural Networks Using Pytorch. For more detailed understanding you may visit This Youtube Playlist
You can visit all the Notebook for Implementation in Google Colab.
-
Simple Neural Network (NN Module)
Basic implementation of a neural network using PyTorch'snn.Module. -
ANN Classification in Fashion-MNIST
Fully connected neural network for classifying Fashion-MNIST images.
-
CNN Classification in Fashion-MNIST
Convolutional Neural Network for image classification. -
UNet
Popular encoder-decoder architecture for image segmentation tasks.
-
Fourier Neural Operator (FNO)
Implementation of FNO for operator learning in PDE-based problems. -
Generative Adversarial Network (GAN)
GAN for generating synthetic images. -
Autoencoder
Dimensionality reduction and feature learning with an autoencoder.
-
Implementing Dataset & Dataloader
Custom dataset and dataloader implementations for flexible training. -
Implementation of Training Pipeline
A reusable training pipeline with training, validation, and checkpointing.
Exploring applications of neural networks in approximation and generalization:
- All Applications – Combined experiments and demonstrations
- Approximation of sin(x) – Training a network to approximate the sine function
- Approximation of x² – Neural network approximation of a quadratic function
- Extrapolation – Testing network generalization outside the training range
- Vary Depth and Width – Effect of network size on approximation performance
- Python 3.8+
- PyTorch
- torchvision
- matplotlib, numpy
Install dependencies:
pip install torch torchvision matplotlib numpy