Deep Learning (CS 701) - Semester 7 BTech CSE at NITH

Your Progress

0 / 28 topics

0% complete

Overview

🎯

Why it matters

Deep Learning powers everything cutting-edge: ChatGPT (transformers), DALL-E (diffusion models), self-driving cars (CNNs), AlphaGo (reinforcement learning). This is THE most advanced AI skill. Master this, you're unstoppable.

💼

Placement relevance

HIGHEST paying AI roles. Research positions at Google Brain, OpenAI, DeepMind. ML Engineer at top startups. ₹40-80 LPA for DL specialists. Publications boost PhD admissions.

🔗

Prerequisites for

AI Research · Generative AI · Computer Vision Research · NLP Research · Reinforcement Learning · PhD Programs

📚

Recommended books

Deep Learning by Ian Goodfellow, Yoshua Bengio, Aaron Courville · Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow by Aurélien Géron · Deep Learning with Python by François Chollet · Neural Networks and Deep Learning by Michael Nielsen

Curriculum — 4 Units

U1

Unit 1 · 7 Topics · 0% complete

Neural Network Fundamentals

⚡ Key Formulae

Backprop:∂L/∂w = (∂L/∂a)(∂a/∂z)(∂z/∂w) — chain rule

ReLU:f(x) = max(0, x)

Batch Norm:x̂ = (x - μ) / √(σ² + ε)

Perceptron & MLP

Activation Functions (ReLU, Sigmoid, Tanh)

Backpropagation Algorithm

Gradient Descent Variants (SGD, Adam)

Vanishing/Exploding Gradients

Weight Initialization

Batch Normalization

U2

Unit 2 · 7 Topics · 0% complete

Convolutional Neural Networks

⚡ Key Formulae

Conv Output:O = (W - K + 2P)/S + 1

Receptive Field:Filters detect local patterns (edges, textures)

Convolution Operation

Pooling (Max, Average)

CNN Architectures (LeNet, AlexNet, VGG, ResNet)

Transfer Learning

Image Classification

Object Detection (YOLO, R-CNN)

Data Augmentation

U3

Unit 3 · 7 Topics · 0% complete

Recurrent Neural Networks & Transformers

⚡ Key Formulae

LSTM Gates:Forget, Input, Output gates (control memory flow)

Attention:Attention(Q,K,V) = softmax(QK^T/√d_k)V

RNN Architecture

LSTM (Long Short-Term Memory)

GRU (Gated Recurrent Unit)

Sequence-to-Sequence

Attention Mechanism

Transformers (Self-Attention)

BERT, GPT Models

U4

Unit 4 · 7 Topics · 0% complete

Advanced Architectures

⚡ Key Formulae

GAN:Generator G vs Discriminator D (adversarial training)

Autoencoder:Encoder: x → z; Decoder: z → x̂ (dimensionality reduction)

Autoencoders

Variational Autoencoders (VAE)

GANs (Generative Adversarial Networks)

Diffusion Models

Reinforcement Learning Basics

Neural Style Transfer

Model Optimization & Deployment

Previous Year Questions

Unit 22023 · End Semester10 marks

Design a CNN architecture for CIFAR-10 image classification. Specify layer types, filter sizes, pooling, activation functions. Calculate output dimensions at each layer.

Unit 32023 · End Semester8 marks

Explain LSTM architecture with a diagram. How do forget, input, and output gates solve vanishing gradient problem? Give a sequence prediction use case.

Unit 42022 · End Semester6 marks

What are GANs? Explain Generator and Discriminator networks. How does adversarial training work? Give 2 real-world applications.

Exam Strategy

🧮

Calculate layer dimensions

CNN questions ask for output size calculations. Formula: O = (W-K+2P)/S + 1. Practice with 3-4 layer networks. Show step-by-step math.

📐

Draw architectures

ResNet, LSTM, Transformer diagrams are expected. Label components clearly. Show skip connections in ResNet, gates in LSTM, attention in Transformers.

💡

Understand WHY, not just WHAT

Why LSTM over RNN? (gradient flow). Why Attention? (long dependencies). Why BatchNorm? (stable training). Exams test conceptual understanding.

Related Subjects

Semester 5

Machine Learning

CS 501

Semester 7

Computer Vision

CS 704

Semester 7

Natural Language Processing

CS 703