CS 4782 Course Notes

CS 4782 Deep Learning · Cornell University

Comprehensive lecture notes compiled by students. Each topic covers key concepts, mathematical foundations, and practical insights from the course.

Foundations

Topic 0

Recap, Linear Models Available

Review of linear algebra, probability, and linear classifiers as a foundation for deep learning.

Topic 1

MLP, SGD, and Optimization Available

Multilayer perceptrons, forward pass, backpropagation, and optimization algorithms including SGD, momentum, AdaGrad, RMSProp, and Adam.

Topic 2

Regularization Available

Techniques to prevent overfitting: L1/L2 regularization, dropout, batch normalization, and data augmentation.

Computer Vision

Topic 3

Convolutional Neural Networks (CNNs) Available

Convolution operations, pooling, and the architecture of convolutional neural networks for image processing.

Topic 4

Modern ConvNets Available

Advanced architectures: AlexNet, VGG, ResNet, and recent developments in convolutional networks.

Natural Language Processing

Topic 5

Word Embeddings Available

Representing words as dense vectors: Word2Vec, GloVe, and contextual embeddings.

Topic 6

Recurrent Neural Networks (RNNs) Available

Sequential data processing with RNNs, LSTMs, and GRUs for language modeling and sequence prediction.

Topic 7

Attention and Transformers Available

Self-attention mechanisms, the Transformer architecture, and positional encodings.

Topic 8

Large Language Models (LLMs) Available

GPT, BERT, and modern large language models: pre-training, fine-tuning, and emergent capabilities.

Vision-Language & Pre-Training

Topic 9

Vision Pre-Training Available

Supervised and self-supervised pre-training strategies for visual representation learning.

Topic 10

Vision-Language Models Available

CLIP, Flamingo, and multimodal models that connect vision and language understanding.

Generative Models

Topic 11

Discriminators and GANs Available

Generative Adversarial Networks: generator-discriminator training, mode collapse, and GAN variants.

Topic 12

U-Nets and VAEs Available

Encoder-decoder architectures, skip connections, and variational autoencoders for generation.

Topic 13

Diffusion Models (Part I) Coming Soon

Denoising diffusion probabilistic models: forward process, reverse process, and training objectives.

Topic 14

Diffusion Models (Part II) Coming Soon

Score matching, guidance techniques, and applications of diffusion models in image synthesis.

Reinforcement Learning

Topic 15

RL Setup Coming Soon

Markov decision processes, rewards, value functions, and the reinforcement learning framework.

Topic 16

Deep Q-Learning (DQN) Coming Soon

Q-learning with neural networks, experience replay, and target networks.

Topic 17

Policy Gradient Coming Soon

REINFORCE algorithm, actor-critic methods, and variance reduction techniques.

Topic 18

RL with Human Feedback (RLHF) Coming Soon

Training language models with human preferences: reward modeling and PPO fine-tuning.

Ethics & Safety

Topic 19

Robustness, Bias, and AI Safety Coming Soon

Adversarial attacks, fairness in ML, and considerations for safe AI deployment.

Topic 20

Interpretability, Legal Issues, and Environmental Impacts Coming Soon

Explainable AI, regulatory considerations, and the carbon footprint of deep learning.