12 Week Deep Learning Syllabus
Written on
Research truly is a winding path. I started along a path of interoperability of digital health records looking at protocols, current systems and considering methods for modifying protocols, filesystems with very high level goals of:
- Putting control of the full health record including charts, results, diagnostic imaging etc into the hands of the patient. Allowing the patient to control the data with the ability to approve and revoke access across multiple healthcare systems and providers.
- Being able to extend existing filesystems, protocols and databases to support #1 so that fork lifting new systems wouldn't be required and the technology could be adopted with minimal cost and change.
As I read literature and consume other media it's impossible to avoid AI right now since it's such a hot topic so I'm going to deviate from my current path and spend some time getting up to speed on AI. What better way to do that than have ChatGPT recommend a syllabus? Here's the recomendation and my goals for the next 12 weeks.
12-Week Deep Learning Learning Path
Overview
Duration: 12 weeks (adjustable)
Goal: Understand core deep learning concepts, build basic projects, and prepare to apply DL in real-world research — especially in healthcare/security contexts.
Outcome: Be ready to build a thesis prototype or dive into applied research using deep learning tools.
Phase 1: Foundations (Weeks 1–4)
| Week | Focus | Resources | Milestones |
|---|---|---|---|
| 1 | What is Deep Learning? + Intro to ANNs | Coursera: Neural Networks (Andrew Ng) YouTube: 3Blue1Brown “Neural Networks” |
Train a tiny neural net (e.g. XOR classifier) |
| 2 | Loss Functions, Optimizers, Backpropagation | Continue Ng’s course Blog: Gradient Descent Visualized |
Understand how a model “learns” |
| 3 | Deep Neural Networks + Overfitting | Try a basic image classifier (MNIST) Tool: Google Colab |
Use dropout & regularization |
| 4 | Convolutional Neural Networks (CNNs) | Fast.ai CNN lesson Visualize filters from a CNN |
Train CNN on CIFAR10 or similar |
Phase 2: Applied Techniques (Weeks 5–8)
| Week | Focus | Resources | Projects |
|---|---|---|---|
| 5 | Transfer Learning | Fine-tune a ResNet using PyTorch Hugging Face “Transformers for Vision” |
Classify a small medical image dataset (e.g. skin lesions) |
| 6 | Intro to NLP & Word Embeddings | Hugging Face NLP Course Jay Alammar’s GPT visuals |
Use BERT or DistilBERT to classify text |
| 7 | Recurrent Neural Networks (RNNs), LSTMs | TensorFlow tutorial: text generation Try GPT-2 in Colab |
Generate medical-style notes |
| 8 | Transformers & Attention | Read “Attention is All You Need” (lite version) Play with a transformer model (e.g. via Hugging Face) |
Modify prompt-tuning for EMR-style text |
Phase 3: Domain Application (Weeks 9–12)
| Week | Focus | Resources | Project Ideas |
|---|---|---|---|
| 9 | Federated Learning | PySyft (OpenMined) tutorial Read: “Federated Learning in Healthcare” review |
Simulate FL with 2 “hospitals” sharing EMR data |
| 10 | Anomaly Detection w/ DL | Blog: Autoencoders for anomaly detection Paper: Deep learning for IDS survey |
Train AE to flag unusual device traffic |
| 11 | Explainable AI (XAI) | SHAP, LIME tutorials Read: “Explainable DL in Healthcare” |
Visualize which features influence model predictions |
| 12 | Mini Thesis Draft + Prep | Draft short proposal Submit paper to arXiv or blog findings |
Write: intro, lit review, methodology |
Tools You’ll Use
- Google Colab / Jupyter
- PyTorch or TensorFlow
- Hugging Face Transformers
- Scikit-learn / Pandas / NumPy
- Kaggle for datasets and practice
Optional Reading List
- Deep Learning by Goodfellow et al.
- Grokking Deep Learning by Andrew Trask
- Key Papers:
- “Attention is All You Need”
- “Federated Learning in Healthcare”
- “XAI in Medical AI”