Deep Learning from Scratch: Perceptron to Custom CNN

Secret Assignment

Since you clicked on this link, you now get access to the secret [assignment], complete it and submit in place of SOP, I would give you preference :)

Description

In this 6-week project, you will learn deep learning collaboratively by building it step by step.

We start with a simple perceptron, then build a neural network (MLP), and finally create a full convolutional neural network (CNN). We will write our own versions of key components like linear layers, activation functions, and even an optimizer.

The project uses real images and runs on GPU. Each week will include hands-on coding, Kaggle contests, and group work on a shared model.

At the end, we will have a working CNN, based on your own architecture, and your own functions.

No prior knowledge needed, just curiosity

Timeline

Here is a tentative timeline: Week 1: Introduction to PyTorch, brushup on Python Week 2: Perceptron Week 3: Simple Neural Networks Week 4: Image Processing and Introduction to Convolutional Neural Networks Week 5: Implementing CNNs Week 6: Buffer

Homepage

Here is what to expect from this website: I will upload all the public resources and so on here, and I will further create one subpage for each week, where I’ll try to make an attempt to write my own explanation to the architecture.

Resources for Week 1

For Week 1, we will focus on Python brush-up, PyTorch basics, and the first real deep learning model Perceptron.

If you’re already comfortable with Python, skip to PyTorch and Perceptron.

Python Brush-Up

• w3schools Python – Up to Functions
  (Focus: Variables, Loops, Conditionals, Functions, Lists, Dictionaries)
• Automate the Boring Stuff – Chapter 1 & 2 (Free)

PyTorch Basics

• PyTorch 60-Minute Blitz
  (Do: Tensors, Autograd, nn.Module)

  Perceptron Theory

• Perceptron – Simple Explanation
• YouTube: “Perceptron”

Week 2-6

Coming Soon