English | 🇮🇷 فارسی
A complete Deep Learning curriculum written as 76 structured Markdown lectures across 14 chapters
The lectures are written in Persian with English technical terms, designed to be read in Obsidian while running every code block yourself in Google Colab. Every concept follows the same philosophy: intuition first, minimal math where it matters, runnable code always.
✨ Obsidian-native formatting: colorized collapsible callouts (
> [!success]answer boxes,[!tip],[!warning],[!summary]), LaTeX math, and[[wiki-links]]chaining every lecture to the next — so the whole course forms one connected graph in the Graph View.
| Feature | Description |
|---|---|
| 🧭 Intuition-first | Every concept starts with why, not formulas |
| 💻 100% runnable | Every code block is tested and ready for Colab |
| 📝 Exercises everywhere | Each lecture ends with exercises & hidden answers |
| 🔗 Cross-linked | Concepts revisit each other across chapters (U-Net → Pix2Pix → Diffusion) |
| 💼 Career-oriented | Interview tips (💡) and résumé-ready project framing |
| 🧪 Honest engineering | Failure modes, limitations, and real debugging workflows |
| # | Chapter | Lectures | Highlights |
|---|---|---|---|
| 01 | Deep Learning Fundamentals | 3 | DL vs ML, NN refresher, first PyTorch training |
| 02 | Tensors and Autograd | 5 | Tensor ops, broadcasting, reshaping, autograd engine |
| 03 | Building Neural Networks from Scratch | 6 | Perceptron → MLP, DataLoaders, init, Project: MNIST |
| 04 | Learning Mechanisms | 5 | Loss functions, GD variants, optimizers, backprop, LR schedules |
| 05 | Regularization & Training Techniques | 5 | Overfitting, L2/AdamW, dropout, batch norm, early stopping |
| 06 | Convolutional Neural Networks | 8 | Convolutions → ResNet → augmentation, Project: CIFAR-10 (~90%) |
| 07 | Advanced Computer Vision | 7 | Transfer learning, YOLO, U-Net, DeepLab, Project: Pet Segmentation |
| 08 | Recurrent Neural Networks | 6 | RNN, LSTM/GRU, seq2seq, time series, embeddings, Project: IMDB Sentiment |
| 09 | Autoencoders | 4 | AE, denoising/sparse AE, VAE — gateway to generative models |
| 10 | Natural Language Processing | 7 | Tokenization/BPE, word vectors, NER, HuggingFace, Project: Semantic Search |
| 11 | Attention and Transformers | 6 | Self-attention, Transformer, GPT from scratch, BERT/MLM, prompting |
| 12 | Generative Adversarial Networks | 6 | GAN theory, WGAN-GP, DCGAN, CycleGAN, StyleGAN → Diffusion, Project: CelebA Faces + FID |
| 13 | Experiment Tracking | 4 | TensorBoard, Weights & Biases sweeps, reproducible checkpoints |
| 14 | Model Deployment | 4 | TorchScript/ONNX, FastAPI + Docker, quantization/distillation, Capstone |
Every lecture follows the same proven structure:
- چرا این مبحث مهم است؟ — motivation & real-world context
- اهداف یادگیری — what you'll be able to do after reading
- Core concepts — intuition first, math only where needed
- Runnable code — complete, tested, Colab-ready blocks
- Key takeaways — summarized in a
[!summary]callout - Exercises — analytical + practical, with collapsible answers
💡 Interview tips are sprinkled throughout — the kind of questions actually asked in ML job interviews.
- Python + NumPy basics
- Fundamentals of Machine Learning (equivalent of Pytopia's ML course)
- No prior deep learning experience required
- Read the lecture
.mdfile in Obsidian (or any Markdown editor) - Run every code block yourself in Google Colab — free GPU included
- Solve the exercises before opening the answer boxes 😉
- Follow the ➡️ wiki-link at the bottom of each lecture to continue
🕸️ Tip: open Obsidian's Graph View after cloning — all 76 lectures form one connected chain from fundamentals to the capstone.
| Project | Chapter | Skills |
|---|---|---|
| MNIST classifier (full pipeline) | 03 | Dataset/DataLoader, training loop, save/load |
| CIFAR-10 (~90% acc, ResNet-18) | 06 | CNNs, augmentation, LR scheduling, TTA |
| Pet Segmentation | 07 | U-Net, IoU/Dice losses, transfer learning |
| IMDB Sentiment Analysis | 08 | Text pipeline, BiLSTM, packed sequences |
| Semantic Search Engine | 10 | Embeddings, retrieval, evaluation metrics |
| mini-GPT (from scratch) | 11 | Causal attention, sampling strategies |
| CelebA Face Generator + FID | 12 | DCGAN, latent interpolation, FID tracking |
| Capstone: deployable model service | 14 | ONNX export, FastAPI, Docker, INT8 quantization |
Structure inspired by Pytopia's Machine-Learning repository. Built with ❤️ for Persian-speaking learners entering deep learning.
⭐ If this course helped you, consider starring the repo!