Core Deep Learning

A complete hands-on journey through the foundations, architectures, and systems that power modern AI and AGI research.

This repo contains both:

• 📘 A comprehensive conceptual guide (math_and_explanations.pdf)
• 💻 Interactive PyTorch notebooks implementing every stage, from perceptrons to genrative model.

---

Overview

This project is structured around the 7 Core Deep Learning Topics.

#	Topic	Focus
1️⃣	Foundations	Perceptron, MLP, activations, loss functions, optimization
2️⃣	Convolutional Neural Networks (CNNs)	Convolutions, pooling, feature hierarchies
3️⃣	Sequence Models	RNNs, LSTMs, GRUs, embeddings (pre-Transformer)
4️⃣	Transformers	Self-attention, positional encoding, encoder vs decoder, pretraining
5️⃣	Embeddings & Representation Learning	Autoencoders, VAEs, contrastive learning (SimCLR, CLIP)
6️⃣	Generative Models	GANs, VAEs, Diffusion Models (Stable Diffusion, Imagen)
7️⃣	Reinforcement Learning	Q-learning, Policy Gradients, PPO, RLHF

---

What’s Inside

/core_DL.ipynb

Hands-on Jupyter notebook covering:

• MLP, CNN, RNN, Transformer demos
• Autoencoders, VAEs, GANs,

includes:

• Math → Code mapping
• Visualizations
• Practical datasets (MNIST, CIFAR, IMDB, CartPole)

---

/math_and_explanations.pdf

A 22-page conceptual guide with diagrams, equations, and intuition behind each topic.
Perfect as a quick reference or for revision.

Covers theory from scratch — activation functions, convolutions, backprop, self-attention, scaling laws, and beyond.

---

Great for:

• Understanding how every deep learning model works (mathematically and intuitively)
• Being able to implement and train them from scratch in PyTorch
• Having a complete, practical foundation for Deep learning

---

Setup

# Clone the repo
git clone https://github.com/scorchinghot/core_DL.ipynb.git
cd core_DL.ipynb

# Install dependencies
pip install -r requirements.txt

Common packages:

torch torchvision torchaudio
transformers datasets
matplotlib

---

Running the Notebooks

1. Open any notebook under /core_DL.ipynb
2. Run cells step-by-step (recommended order: 1 → 7)
3. View generated samples, loss curves, and embeddings

If you have a GPU:

export CUDA_VISIBLE_DEVICES=0

or in notebook:

device = 'cuda' if torch.cuda.is_available() else 'cpu'

---

References & Inspiration

• CS231n: Convolutional Neural Networks for Visual Recognition
• CS224n: Natural Language Processing with Deep Learning
• Deep Learning — Goodfellow, Bengio, Courville
• Spinning Up in Deep RL — OpenAI
• Illustrated Transformer — Jay Alammar
• Lil’Log — Lilian Weng (https://lilianweng.github.io)

---

License

MIT License © 2025 Scorchinghot Use freely for learning and teaching. Attribution appreciated.