This roadmap is designed for undergraduate students transitioning from Machine Learning to Deep Learning. It covers Neural Networks, CNNs, RNNs, LSTMs, and Transformers using PyTorch.

🔵 Phase 1: Deep Learning Foundations

1️⃣ Neural Network Fundamentals

✅ Core Concepts

• What is a Neural Network?
• Perceptron
• Activation Functions (ReLU, Sigmoid, Tanh, Softmax)
• Forward Propagation
• Loss Functions (MSE, Cross-Entropy)
• Backpropagation
• Gradient Descent
• Overfitting & Regularization (Dropout, L2)

2️⃣ Math Behind Deep Learning (Revision)

📌 Linear Algebra

• Matrix multiplication
• Dot product
• Vector spaces
• Eigenvalues (basic intuition)

📌 Calculus

• Partial derivatives
• Chain rule
• Gradient computation

📌 Probability

• Softmax as probability distribution
• Log-likelihood
• Cross-entropy loss

📚 Learning Resources (Foundations)

🎥 YouTube Playlists

• StatQuest – Neural Networks
• 3Blue1Brown – Neural Networks
• DeepLearning.AI (Andrew Ng)

🎓 Courses

• Google Deep Learning Crash Course
• Embeddings Module (Google ML Crash Course)
• Coursera – Deep Learning Specialization
• Fast.ai – Practical Deep Learning

3️⃣ PyTorch Basics

Students must learn:

• Working with tensors
• Autograd (automatic differentiation)
• Implementing a three-layer neural network
• Writing a full training loop
• Model evaluation
• GPU training

📚 Resources

• PyTorch Official Tutorials
• PyTorch Playlist (YouTube)
• PyTorch 60 Minute Blitz

🔵 Phase 2: Core Deep Learning Architectures

📘 Feed-Forward Neural Networks (FNN / MLP)

Topics

• Multi-layer neural networks
• Activation functions
• Loss functions
• Backpropagation
• Weight initialization

Recommended Resources

• Neural Networks & Deep Learning (Coursera)
• Dive into Deep Learning (Free Book with PyTorch Code)

📘 Convolutional Neural Networks (CNN)

Topics

• Convolution operation
• Filters and feature maps
• Stride and padding
• Pooling layers
• CNN architectures
• Training CNN in PyTorch

Recommended Resources

• Convolutional Neural Networks (Coursera)
• Dive into Deep Learning – CNN Chapter
• PyTorch CIFAR-10 Tutorial
• MNIST CNN Tutorial (DataCamp)

📘 Recurrent Neural Networks (RNN, LSTM, GRU)

Topics

• Sequential data
• Vanishing gradient problem
• Basic RNN
• LSTM
• GRU
• Next-word prediction

Recommended Resources

• Sequence Models (Coursera)
• Dive into Deep Learning – RNN Chapter
• PyTorch NLP Tutorial
• Next Word Prediction using LSTM

📘 Transformers

Topics

• Attention mechanism
• Self-attention
• Scaled dot-product attention
• Multi-head attention
• Positional encoding
• Encoder–Decoder architecture
• Fine-tuning pretrained models

Recommended Resources

• Attention Is All You Need (Original Paper)
• Dive into Deep Learning – Attention Mechanisms Chapter
• Transformers Explained (Visual Guide 1)
• Transformers Explained (Visual Guide 2)
• Transformers Explained (Visual Guide 3)
• Transformers for Machine Translation
• My Lecture Slides on Transformer

🔵 Phase 3: Mini Projects (Choose Any 2)

🔹 Option A: Image Classification (CNN)

Dataset:
CIFAR-10 Dataset

Deliverables

• Data preprocessing
• CNN architecture
• Training curves
• Accuracy
• Confusion matrix
• Conclusion

🔹 Option B: Sentiment Analysis (RNN / LSTM)

Dataset:
IMDB Reviews Dataset

Deliverables

• Text preprocessing
• Tokenization
• LSTM model
• F1-score evaluation
• Error analysis

🔹 Option C: Text Summarization (Transformer)

Dataset:
CNN/DailyMail Dataset

Deliverables

• Fine-tuning pretrained model
• ROUGE evaluation
• Generated samples

🎯 Expected Outcomes

After completing this roadmap, students will be able to:

• Build neural networks from scratch
• Train CNN, RNN, and LSTM models
• Understand and implement Transformers
• Fine-tune pretrained models
• Write structured DL project reports
• Apply Deep Learning in internships and projects