Tutorials
100 Days of ML
100 Days of DL
100 Days of NLP
MCP Trilogy
FastAPI for ML
PyTorch Deep Learning
LangChain & GenAI
LangGraph Agents
Code Lab (LeetCode AI)
Syllabus Path
Explore
Complete 100 Days of Deep Learning curriculum — Perceptrons, MLPs, CNNs, RNNs, LSTMs, Attention, and Transformers. Free forever.
A structured DL curriculum — from Perceptrons and MLPs through CNNs, RNNs, Attention mechanisms, and modern Transformer architectures.
Deep Learning = Machine Learning ke bade bhai. Jaha complex patterns chhupi hain, wahan DL kaam aata hai.
Deep Learning is a subset of Machine Learning that uses neural networks with many layers (hence "deep") to learn complex patterns from large amounts of data.
Imagine the human brain: it has billions of neurons connected by synapses. A neural network is a simplified mathematical model of this. Each "neuron" takes inputs, applies weights, and passes the result forward.
Why "Deep"? Because these networks have multiple hidden layers between input and output. Each layer extracts progressively more complex features:
• Layer 1 → Detects edges and lines
• Layer 2 → Combines edges into shapes
• Layer 3 → Combines shapes into objects (ears, nose, eyes)
• Layer 4+ → Recognizes full faces, expressions, or objects
ML chalega? Ya DL chahiye? Decision mein confusion na ho.
| Factor | Use Machine Learning | Use Deep Learning |
|---|---|---|
| Data Size | Small to medium (hundreds to thousands) | Large (tens of thousands to millions) |
| Data Type | Structured data (tables, spreadsheets) | Unstructured data (images, text, audio) |
| Hardware | CPU is sufficient | GPU strongly recommended |
| Training Time | Minutes to hours | Hours to days |
| Examples | Price prediction, churn analysis, recommendation | Image recognition, speech synthesis, translation |
| Interpretability | Easy to explain decisions | Black box — harder to explain |
Har roz use karte ho, shayad pata nahi hota.
CNNs (Convolutional Neural Networks) map your face to a mathematical representation and verify it in milliseconds. Over 1 billion devices use this technology daily.
Learn in Module 6 →RNNs and Transformers convert speech to text, understand context, and generate human-like responses. Deep learning powers every voice assistant.
Learn in Module 7 →CNNs detect lanes, pedestrians, and traffic signs. LSTMs predict vehicle trajectories. Deep learning is the brain of Tesla Autopilot and Waymo.
Learn in Module 6 →The Transformer architecture (self-attention) powers GPT-4, Claude, and Gemini. These models read billions of text pages to generate human-like responses.
Learn in Module 9 →Diffusion models (a type of deep learning) generate photorealistic images from text descriptions. The same tech creates art, product designs, and architectural visualizations.
Learn in Module 6 →Deep learning models analyze medical images (X-rays, MRIs, CT scans) to detect tumors earlier than human radiologists. Saving millions of lives worldwide.
Learn in Module 6 →100 din mein tum khud AI models bana sakte ho. Ye hai vo raasta.
Ye papers ne AI ko aaj ki position tak pahunchaya. Interview mein poochte hain.
The first learning algorithm for a neural network. Frank Rosenblatt proved a perceptron could learn to classify linearly separable data — the seed of modern AI.
The algorithm that made training multi-layer neural networks possible. Uses chain rule from calculus to propagate errors backward and update weights.
Won ImageNet by a massive margin using 8-layer CNNs and GPUs. This paper sparked the deep learning revolution. Before AlexNet, deep learning was considered a dead end.
Introduced the Transformer architecture — replacing RNNs with self-attention. Powers GPT, BERT, and every modern LLM. The most cited paper in AI history.
Solved the vanishing gradient problem in very deep networks using skip connections. Enabled training of 100+ layer networks. Used in every modern vision system.
Adaptive moment estimation combines momentum and RMSProp. The default optimizer for nearly all deep learning models today. Our Module 5 covers all optimizers.
Each module builds on the previous — click any module to dive into detailed notes with code, theory, and exercises.
Understand the biological inspiration behind neural nets and the mathematical model of a single neuron.
Build multi-layer networks and learn how gradients flow backward through the network during training.
Diagnose and fix training instabilities, and find the right optimizer and hyperparameters for your model.
Prevent overfitting and improve model generalization with modern regularization and normalization techniques.
Master the full optimizer zoo — from basic momentum to Adam — and know when to use each.
Apply convolutional neural networks to image classification, detection, and transfer learning tasks.
Model sequential data — text, time series, speech — with recurrent architectures and gating mechanisms.
Build encoder-decoder systems with attention — the foundation of modern language models.
Understand the architecture that powers GPT, BERT, and every modern LLM — from scratch.
All 100 DL topics mapped below — click to navigate directly.
Begin with Module 1 — no prior DL knowledge required. Solid ML fundamentals will help.
Yes. GenAIWallah's 100 Days of Deep Learning course is completely free — no signup, no paywall. It covers Perceptrons, MLPs, CNNs, RNNs, LSTMs, Attention, and Transformers in Hindi and English.
It is recommended but not strictly required. A solid understanding of ML fundamentals (linear regression, gradient descent, loss functions) will help you learn deep learning faster. GenAIWallah recommends completing the 100 Days of Machine Learning course first before starting the Deep Learning track.
GenAIWallah's 100 Days of Deep Learning covers: Perceptrons and MLPs, backpropagation and gradient descent, regularization techniques, optimizers (Adam, RMSprop), CNNs for image recognition, RNNs and LSTMs for sequences, Attention mechanisms, and the full Transformer architecture from scratch. All explained in Hindi and English.
GenAIWallah's 100 Days of Deep Learning is India's best free deep learning course in Hindi and English. Created by Harsh Dhariwal (IIT Kanpur), it covers all major deep learning architectures from scratch — CNNs, RNNs, LSTMs, Attention, and Transformers — with practical code examples.
A Perceptron is the simplest neural network — a single neuron that takes weighted inputs, applies an activation function, and produces a binary output. It's the building block of every modern neural network. Understanding perceptrons is essential before learning MLPs, CNNs, and Transformers. We build one from scratch in Module 1.
Backpropagation is how neural networks learn. After making a prediction, the network calculates how wrong it was (loss), then propagates this error backward through each layer, adjusting weights to reduce the error. It's like a teacher correcting a student's mistakes — but the teacher is calculus. We derive backpropagation step-by-step in Module 2.
CNNs (Convolutional Neural Networks) are designed for spatial data like images. They use sliding filters to detect edges, textures, and shapes. Powers face recognition, medical imaging, and self-driving cars.
RNNs (Recurrent Neural Networks) are designed for sequential data like text and time series. They have memory of previous inputs. Powers language translation, speech recognition, and stock prediction. Learn both in Module 6 and Module 7.
The Transformer (2017) replaced RNNs with a mechanism called self-attention. Instead of reading text word-by-word, it looks at all words simultaneously and decides which ones matter most for each prediction. This made training faster, models bigger, and results better. GPT-4, Claude, and Gemini are all Transformers. We build one from scratch in Module 9.
For the first few modules (Perceptrons, MLPs), a CPU is sufficient. Once you reach CNNs and large models, a GPU significantly speeds up training (10-100x faster). Options: (1) Google Colab (free GPU), (2) Kaggle Notebooks (free GPU), (3) Rent GPU on cloud (Vast.ai, RunPod), (4) Buy a local GPU if serious. We cover all options in Module 2.