Workshop Learning Objectives (Convolutional Neural Networks)

Session 1

1. Discuss what focus or discipline or perspective that you come from.
2. Discuss what your learning goals are with respect to this workshop.
3. Discuss how vision research has influenced the area of machine learning.
4. Explain why we need convolutional neural networks (CNNs).
5. List and describe the various CNN domains.
6. Describe, configure, and request the Jupyter Lab notebook environment.
7. Explain the different types of cells within Jupyter.
8. Discuss Edge-Detection.
9. Show how to manipulate the Jupyter notebook environment.
10. Utilize Google's Teachable Machine to train an ML model.
11. Describe the two types of folders or image data involved with the ML.
12. Discuss outcomes and any surprises.
13. Define what a classifier is.
14. Discuss how you can learn more about the vocabulary and terminology with ML/DL.
15. Explain what a tensor is.

Session 2

1. Discuss how deep learning images are 3D tensors.
2. Describe what pixel or element is related to a tensor.
3. Discuss how color image imput is processed into a gray scale output.
4. Describe how neurons or kernels filter images.
5. Describe the purpose of weights.
6. Describe what a feature is.
7. Apply and discuss the basic math of filters.
8. Discuss padding and its purpose.
9. Discuss TensorFlow.
10. Implement a convolutional operation.
11. Explain stride.
12. Explain max and average pooling.
13. Discuss the some of the types of layers.
14. Discuss fully connected layers with respect to problem types and activation functions.

Session 3

1. Discuss how data augmentation increases accuracy of your model.
2. Discuss what is meant by a model that is robust versus brittle.
3. Discuss what keras is.
4. Discuss what matplotlib is.
5. Implement image classification with data augmentation.
6. Discuss the results and any interesting outcomes or surprises.
7. Discuss why it's important to flatten labels.
8. Discuss why an initial data analysis is important prior to training.
9. Discuss techniques of data augmentation.
10. Explain the Goldilocks problem in terms of proper fit with respect to training models.
11. Discuss the difference between overfit, underfit, and generalzing well.
12. Discuss how overfitting can be a byproduct of outliers.
13. Discuss the continuum of underfitting and overfitting through the duration of model training.
14. Discuss how overfit can be prevented.
15. Describe regularization.
16. Discuss how underfit can be prevented.