Deep Learning – Practical Approach
(Detailed Notes)
1. Lab & Pre-requirements
Software
- Python 3.8+
- Anaconda/Miniconda (environment management)
- IDE/Editors: Jupyter Notebook, VS Code, or PyCharm
- Git for version control
Libraries & Frameworks
- TensorFlow 2.x + Keras API
- NumPy & Pandas
- Matplotlib & Seaborn
- Scikit-learn
- OpenCV/PIL (optional for image processing)
Programming Knowledge Needed
- Python basics: loops, functions, pip
- Math: Linear Algebra, Calculus, Probability & Statistics
- ML basics (optional): Supervised/Unsupervised, Classification, Regression
2. Lab Environment Preparations
- Day 1: Install Python + TensorFlow, verify with 'Hello World' NN
- Day 2–4: Prepare notebooks for tensors, activations, losses
- Day 5–6: Preprocess datasets (MNIST, CIFAR-10)
- Day 7–8: Image augmentation, CNNs, transfer learning
- Day 9: Save/load models, deployment setup
3. 9-Day Session Plan
Day Module Focus Activities / Practical Duration
1 Introduction & Install TensorFlow 3 hrs
Setup & Keras, run first
NN
2 Neural Networks & Tensor ops, build 3 hrs
Tensors basic NN
3 Forward/Backward Code 3 hrs
Propagation forward/backprop,
activations
4 Loss, Optimization Loss fns, optimizers, 3 hrs
& Evaluation metrics
5 Keras Basics Build/train simple 3 hrs
NN (Sequential API)
6 Validation & Validation sets, 3 hrs
Prediction predictions,
confusion matrix
7 CNN Basics Build CNN, image 3 hrs
preprocessing
8 Transfer Learning Fine-tuning, 3 hrs
augmentation, pre-
trained models
9 Deployment Save/load/export 3 hrs
models, deployment
workflows
4. Practical Highlights
- Day 5 onwards: focus on Keras API.
- Work with datasets like MNIST, CIFAR-10.
- Apply CNNs for image classification.
- Explore fine-tuning & transfer learning.
- Learn saving/loading models & deployment.
