Courses › Course05

Course05 · Deep Learning

Neural networks from first principles to transformers: build them, train them, debug them, and understand the hardware they run on.

42 days 42 written 42 complete

Days 197-203 · Neural Network Foundations 7/7 complete
  1. Day 197 The Perceptron complete
  2. Day 198 Activation Functions complete
  3. Day 199 Forward Propagation complete
  4. Day 200 Backpropagation complete
  5. Day 201 A Neural Network in Pure NumPy complete
  6. Day 202 PyTorch Tensors complete
  7. Day 203 Training MNIST from Scratch complete

Project: MNIST from Scratch — Implement a two-layer neural network in pure NumPy that reaches at least 95% accuracy on MNIST digits.

Days 204-210 · Training Deep Networks 7/7 complete
  1. Day 204 PyTorch: autograd and nn.Module complete
  2. Day 205 Datasets and DataLoaders complete
  3. Day 206 Optimizers: SGD to Adam complete
  4. Day 207 Learning Rate Schedules complete
  5. Day 208 Dropout, Batch Norm, and Regularization complete
  6. Day 209 Debugging Training Runs complete
  7. Day 210 A Disciplined Training Project complete

Project: Fashion-MNIST Classifier — Train a PyTorch classifier with a proper training loop, LR schedule, and regularization, beating a stated baseline.

Days 211-217 · Convolutional Networks and Vision 7/7 complete
  1. Day 211 Convolutions complete
  2. Day 212 CNN Architectures complete
  3. Day 213 Transfer Learning complete
  4. Day 214 Data Augmentation complete
  5. Day 215 Training a Vision Model End to End complete
  6. Day 216 Beyond Classification: Detection and Segmentation complete
  7. Day 217 Your Own Image Classifier complete

Project: Custom Image Classifier — Fine-tune a pretrained CNN on your own small image dataset with augmentation and an error-case gallery.

Days 218-224 · Sequences and Text 7/7 complete
  1. Day 218 Text Preprocessing and Tokenization complete
  2. Day 219 Word Embeddings complete
  3. Day 220 Recurrent Neural Networks complete
  4. Day 221 LSTMs and GRUs complete
  5. Day 222 Sequence-to-Sequence and Early Attention complete
  6. Day 223 Text Classification with Embeddings complete
  7. Day 224 A Sentiment Analysis Project complete

Project: Sentiment Analyzer — Build a review-sentiment model comparing bag-of-words, embeddings, and a small recurrent network.

Days 225-231 · Transformers 7/7 complete
  1. Day 225 “Attention Is All You Need” complete
  2. Day 226 Self-Attention, Step by Step complete
  3. Day 227 The Transformer Architecture complete
  4. Day 228 Encoder Models: BERT and Friends complete
  5. Day 229 Decoder Models: The GPT Family complete
  6. Day 230 Hugging Face Transformers in Practice complete
  7. Day 231 Fine-Tuning a Small Transformer complete

Project: Fine-Tuned Transformer — Fine-tune a small pretrained transformer on a text classification task with Hugging Face and report results honestly.

Days 232-238 · Training at Scale 7/7 complete
  1. Day 232 GPUs and AI Hardware complete
  2. Day 233 Mixed Precision and Performance complete
  3. Day 234 Distributed Training Concepts complete
  4. Day 235 Experiment Tracking complete
  5. Day 236 Quantization and Distillation complete
  6. Day 237 Scaling Laws and What They Bought Us complete
  7. Day 238 Section Project: Reproducing a Paper complete

Project: Section Project: Reproduce a Result — Reproduce a small published deep-learning result end to end, with experiment tracking and a reproduction report.