← All projects
Courseworkcoursework

Deep Learning — CS 7643

CNNs, RNNs and attention implemented from scratch in PyTorch.

The problem

Georgia Tech's CS 7643 implements the deep-learning curriculum hands-on rather than as a survey. Each homework rebuilds a foundational architecture from scratch in PyTorch, then benchmarks the from-scratch version against the pretrained baseline.

Who this is for

GT prospective students, anyone deciding whether the OMSCS deep-learning track is worth taking.

Architecture

CNNs
Built from conv / pool / batchnorm primitives for image classification.
RNNs
Vanilla RNN → LSTM for sequence modelling.
Attention
Scaled dot-product attention from scratch, then transfer-learning on top.

Request / data flow

  1. 01Read the paper.
  2. 02Implement the primitive in raw PyTorch.
  3. 03Train on the assigned dataset.
  4. 04Compare against the pretrained baseline — explain the gap.

Stack

PyTorchCNNRNNAttentionTransfer Learning