Skip to Content
Learning PathsOverview

Learning Paths

Structured, sequential learning journeys through machine learning. Each path provides a curated curriculum with clear objectives, prerequisites, and hands-on practice.

Total time investment: 120-160 hours for complete curriculum

How Learning Paths Work

Each path includes:

  • Clear learning objectives - What you’ll master
  • Prerequisites - Required background knowledge
  • Week-by-week structure - Organized timeline with estimated hours
  • Content references - Links to concepts, papers, and examples
  • Hands-on practice - Exercises and projects
  • Assessment criteria - Self-check understanding

Philosophy: Build knowledge progressively through structured practice, not random exploration.

Foundation Paths

Core deep learning fundamentals. Start here if you’re new to deep learning.

Deep Learning Foundations

Duration: 5 weeks | 45-60 hours | Beginner

Complete introduction to deep learning covering neural networks, CNNs, attention, and language models. This is the recommended entry point.

Covers: Modules 1-4 (Neural Networks, CNNs, Transformers, GPT)

Neural Network Foundations

Duration: 2 weeks | 15-20 hours | Beginner

Master the fundamentals of neural networks, backpropagation, and optimization.

Topics:

  • Perceptrons and MLPs
  • Backpropagation and gradient descent
  • Optimization algorithms (SGD, Adam)
  • Regularization (dropout, weight decay)
  • Practical training techniques

Hands-on: Build MNIST classifier from scratch in NumPy

Computer Vision with CNNs

Duration: 2 weeks | 12-18 hours | Beginner to Intermediate

Learn convolutional networks for computer vision.

Topics:

  • Convolution and pooling operations
  • CNN architectures (AlexNet, VGG, ResNet)
  • Transfer learning strategies
  • Medical imaging applications

Papers: AlexNet (2012), VGG (2014), ResNet (2015)

Attention and Transformers

Duration: 2 weeks | 12-16 hours | Intermediate

Master attention mechanisms and transformer architecture.

Topics:

  • RNN limitations and attention motivation
  • Scaled dot-product attention
  • Multi-head attention
  • Transformer architecture (Vaswani et al., 2017)

Paper: Attention Is All You Need (most important paper in modern AI)

Language Models with NanoGPT

Duration: 2 weeks | 15-25 hours | Intermediate

Build GPT from scratch following Andrej Karpathy’s approach.

Topics:

  • GPT architecture (decoder-only transformer)
  • Causal attention and autoregressive generation
  • Tokenization (BPE)
  • Language model training techniques
  • Text generation strategies

Hands-on: Implement GPT-2 (124M params) and train on custom data

Advanced Paths

Cutting-edge techniques and modern AI systems.

Advanced Topics Overview

Duration: 3 weeks | 26-38 hours | Advanced

Survey of advanced deep learning topics.

Covers: Multimodal learning, generative models, advanced training

Multimodal Vision-Language Models

Duration: 2 weeks | 12-18 hours | Advanced

Master vision-language models like CLIP, ViT, and modern VLMs.

Topics:

  • Multimodal fusion strategies
  • Contrastive learning (InfoNCE, CLIP)
  • Vision Transformers (ViT)
  • Zero-shot transfer
  • Advanced VLMs (Flamingo, BLIP-2, LLaVA)

Papers: CLIP (2021), ViT (2021)

Applications: Visual search, VQA, accessibility tools

Generative Diffusion Models

Duration: 2 weeks | 12-18 hours | Advanced

Learn diffusion models for high-quality generation.

Topics:

  • Generative models comparison (GANs, VAEs, Diffusion)
  • Diffusion fundamentals (forward/reverse process)
  • DDPM training and DDIM sampling
  • Classifier-free guidance for text-to-image
  • Healthcare applications (synthetic medical data)

Papers: DDPM (2020), DDIM (2021), DALL-E 2 (2022)

Advanced Training Topics

Duration: 1 week | 8-12 hours | Advanced

Modern training techniques and dynamics.

Topics:

  • Self-supervised learning (contrastive + masked)
  • Masked prediction (BERT, MAE)
  • Training dynamics (double descent, overparameterization)
  • Scaling laws and compute-optimal training

Specialized Paths

Domain-specific applications and skills.

Healthcare AI & Electronic Health Records

Duration: 2 weeks | 16-24 hours | Intermediate to Advanced

Apply deep learning to healthcare, focusing on EHR analysis.

Topics:

  • EHR structure and medical coding (ICD-10, ATC, CPT)
  • Tokenization for medical events
  • Healthcare foundation models (ETHOS, BEHRT, GatorTron)
  • Clinical decision support
  • Interpretability and fairness in medical AI

Datasets: MIMIC-III, MIMIC-IV, EmergAI

Thesis connection: Directly supports multimodal EHR research

Research Methodology & Academic Writing

Duration: 1 week | 6-10 hours | Intermediate

Essential skills for conducting and publishing ML research.

Topics:

  • Reading research papers (three-pass method)
  • Formulating research questions (PICOT framework)
  • Experimental design (baselines, ablations, statistics)
  • Structuring papers (IMRaD format)
  • Publication strategy

Outcome: Ready to conduct rigorous ML research and write papers

For Beginners

Start from scratch and build to advanced topics:

  1. Deep Learning Foundations (5 weeks)
    • Or individual modules: NN → CNNs → Transformers → GPT
  2. Advanced Topics (3 weeks)
    • VLMs → Diffusion → Advanced Training
  3. Specialize: Healthcare AI or domain of choice

Total: 8-10 weeks for solid foundation + specialization

For Researchers

Focus on modern techniques and research skills:

  1. Attention and Transformers (2 weeks)
  2. Vision-Language Models (2 weeks)
  3. Research Methodology (1 week)
  4. Healthcare AI (2 weeks) - if applicable

Total: 5-7 weeks to research-ready

For Healthcare AI

Specialized curriculum for medical AI:

  1. Neural Network Foundations (2 weeks)
  2. Computer Vision (2 weeks)
  3. Transformers (2 weeks)
  4. Healthcare AI & EHR (2 weeks)
  5. Multimodal VLMs (2 weeks) - for multimodal clinical AI

Total: 10 weeks for healthcare AI specialization

Progress Tracking

Recommended approach:

  1. Start with prerequisites - Ensure you have required background
  2. Follow week-by-week structure - Don’t skip ahead
  3. Complete hands-on exercises - Theory alone isn’t enough
  4. Self-assess - Check understanding before moving on
  5. Build projects - Apply knowledge to real problems

Completion criteria: Can explain concepts clearly, implement algorithms from scratch, and apply to new problems.

Explore Library Content

While paths provide structure, the Library offers reference material:

Difference: Library = reference, Paths = curriculum

ResearchDeep Learning