Deep Learning with TensorFlow/PyTorch

Course outline and learning expectations

This is a tutor-led course. The outline shows what your tutor will cover; teaching materials and examinations are provided directly to enrolled students.

What you will learn

• Compare TensorFlow and PyTorch and select a suitable framework for a project
• Build and train deeper neural networks for multi-class image classification
• Configure optimizers, learning-rate schedules and batch-training strategies
• Use CPU and GPU runtimes in Google Colab or Kaggle Notebooks while managing compute limits
• Save, load, checkpoint and version TensorFlow and PyTorch models reproducibly
• Diagnose overfitting, gradient instability, shape errors, device errors and data-quality problems
• Evaluate and present a crop-disease classifier using appropriate classification metrics

Modules

1. 1

   TensorFlow vs. PyTorch - choosing a framework

   Compare core workflows and select a framework based on project needs and available resources

   Tensors and automatic differentiationKeras workflowPyTorch workflowFramework selection criteriaReproducible project setup
2. 2

   Building deeper architectures

   Design and train deeper networks for image classification without losing stability

   Convolutional neural networksActivation functionsBatch normalizationDropoutResidual connectionsTransfer learning and fine-tuning
3. 3

   Optimizers, learning rates and batch training

   Control model convergence through informed training choices

   SGD, Adam and AdamWLearning-rate schedulesBatch size trade-offsLoss functionsGradient clippingTraining and validation curves
4. 4

   Working with GPUs and free cloud compute

   Run TensorFlow and PyTorch notebooks on available accelerators and manage platform limits

   CPU vs GPU executionGoogle Colab setupKaggle Notebooks setupDevice placementGPU memory monitoringEfficient session and quota use
5. 5

   Saving, loading and versioning models properly

   Preserve training progress and make experiments repeatable

   Weights vs full modelsCheckpointsOptimizer stateTensorFlow saving formatsPyTorch state dictionariesExperiment namingGit-based version control
6. 6

   Debugging training problems

   Use systematic checks to find and correct data, model and training failures

   Overfitting and underfittingVanishing and exploding gradientsShape and device errorsData leakage and label issuesLearning-curve diagnosisReproducibility checks
7. 7

   Training under hardware constraints common in Uganda

   Apply practical methods for limited local GPU access, memory and connectivity

   Transfer learningFreezing layersSmaller input sizesBatch-size tuningGradient accumulationMixed precision where supportedCPU fallbackCheckpointing for interrupted sessions
8. 8

   Capstone - multi-class crop disease image classifier

   Build, train, evaluate and document a practical classifier using TensorFlow or PyTorch

   Dataset preparationData augmentationModel selectionCloud GPU trainingAccuracy, precision, recall and confusion matrixModel checkpointingFinal technical report and demonstration